ui: Initialize preferred xemu NVIDIA application profile

---
updated-dependencies:
- dependency-name: astral-sh/setup-uv
  dependency-version: 6.3.1
  dependency-type: direct:production
  update-type: version-update:semver-patch
...

Signed-off-by: dependabot[bot] <support@github.com>

.github/workflows/build-xemu-win64-toolchain.yml

@@ -35,7 +35,7 @@ jobs:
             type=ref,event=branch
             type=sha
       - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@b5ca514318bd6ebac0fb2aedd5d36ec1b5c232a2 # v3
+        uses: docker/setup-buildx-action@e468171a9de216ec08956ac3ada2f0791b6bd435 # v3
       - name: Login to GitHub Container Registry
         uses: docker/login-action@74a5d142397b4f367a81961eba4e8cd7edddf772 # v3
         if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
@@ -44,7 +44,7 @@ jobs:
           username: ${{ github.actor }}
           password: ${{ secrets.GITHUB_TOKEN }}
       - name: Build and push image
-        uses: docker/build-push-action@471d1dc4e07e5cdedd4c2171150001c434f0b7a4 # v5
+        uses: docker/build-push-action@263435318d21b8e681c14492fe198d362a7d2c83 # v5
         with:
           context: ubuntu-win64-cross
           push: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}

.github/workflows/build.yml

@@ -83,7 +83,7 @@ jobs:
 
     steps:
     - name: Download source package
-      uses: actions/download-artifact@95815c38cf2ff2164869cbab79da8d1f422bc89e # v4
+      uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4
       with:
         name: src.tar.gz
     - name: Extract source package
@@ -140,7 +140,7 @@ jobs:
           arch: aarch64
     steps:
     - name: Download artifacts
-      uses: actions/download-artifact@95815c38cf2ff2164869cbab79da8d1f422bc89e # v4
+      uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4
       with:
         name: ${{ matrix.artifact_name }}
         path: ${{ matrix.artifact_name }}
@@ -174,25 +174,25 @@ jobs:
           build_param: --debug
           artifact_name: xemu-ubuntu-x86_64-debug
           artifact_filename: xemu-ubuntu-x86_64-debug.tgz
-          runs-on: ubuntu-24.04
+          runs-on: ubuntu-22.04
         - arch: x86_64
           configuration: Release
           build_param:
           artifact_name: xemu-ubuntu-x86_64-release
           artifact_filename: xemu-ubuntu-x86_64-release.tgz
-          runs-on: ubuntu-24.04
+          runs-on: ubuntu-22.04
         - arch: aarch64
           configuration: Debug
           build_param: --debug
           artifact_name: xemu-ubuntu-aarch64-debug
           artifact_filename: xemu-ubuntu-aarch64-debug.tgz
-          runs-on: ubuntu-24.04-arm
+          runs-on: ubuntu-22.04-arm
         - arch: aarch64
           configuration: Release
           build_param:
           artifact_name: xemu-ubuntu-aarch64-release
           artifact_filename: xemu-ubuntu-aarch64-release.tgz
-          runs-on: ubuntu-24.04-arm
+          runs-on: ubuntu-22.04-arm
     steps:
     - name: Initialize compiler cache
       id: cache
@@ -202,7 +202,7 @@ jobs:
         key: cache-${{ runner.os }}-${{ matrix.arch }}-${{ matrix.configuration }}-${{ github.sha }}
         restore-keys: cache-${{ runner.os }}-${{ matrix.arch }}-${{ matrix.configuration }}-
     - name: Download source package
-      uses: actions/download-artifact@95815c38cf2ff2164869cbab79da8d1f422bc89e # v4
+      uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4
       with:
         name: src.tar.gz
     - name: Extract source package
@@ -245,6 +245,11 @@ jobs:
 
         echo -e "\n\nCompiler Cache Stats:"
         ccache -s
+    - name: Test
+      run: |
+        pushd src/build
+        ./pyvenv/bin/meson test --suite xbox
+        popd
     - name: Generate AppImage
       run: |
         wget --no-verbose https://github.com/linuxdeploy/linuxdeploy/releases/latest/download/linuxdeploy-${{ matrix.arch }}.AppImage
@@ -300,12 +305,12 @@ jobs:
           artifact_filename: xemu-macos-arm64-release.zip
     steps:
     - name: Download source package
-      uses: actions/download-artifact@95815c38cf2ff2164869cbab79da8d1f422bc89e # v4
+      uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4
       with:
         name: src.tar.gz
     - name: Extract source package
       run: tar xf src.tar.gz
-    - uses: actions/setup-python@v5.5.0
+    - uses: actions/setup-python@v5.6.0
       with:
         python-version: '3.12'
     - name: Install dependencies
@@ -353,12 +358,12 @@ jobs:
         configuration: ["debug", "release"]
     steps:
     - name: Download x86_64 build
-      uses: actions/download-artifact@95815c38cf2ff2164869cbab79da8d1f422bc89e # v4
+      uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4
       with:
         name: xemu-macos-x86_64-${{ matrix.configuration }}
         path: xemu-macos-x86_64-${{ matrix.configuration }}
     - name: Download arm64 build
-      uses: actions/download-artifact@95815c38cf2ff2164869cbab79da8d1f422bc89e # v4
+      uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4
       with:
         name: xemu-macos-arm64-${{ matrix.configuration }}
         path: xemu-macos-arm64-${{ matrix.configuration }}
@@ -393,7 +398,7 @@ jobs:
     needs: [Ubuntu, macOSUniversal, Windows, WindowsPdb]
     steps:
     - name: Download artifacts
-      uses: actions/download-artifact@95815c38cf2ff2164869cbab79da8d1f422bc89e # v4
+      uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4
       with:
         path: dist
     - name: Extract source package
@@ -415,7 +420,7 @@ jobs:
       run: |
         cp dist/xemu-win-x86_64-release-pdb/xemu-win-x86_64-release.zip dist/xemu-win-x86_64-release-pdb/xemu-win-release.zip
     - name: Publish release
-      uses: softprops/action-gh-release@c95fe1489396fe8a9eb87c0abf8aa5b2ef267fda # v2.2.1
+      uses: softprops/action-gh-release@72f2c25fcb47643c292f7107632f7a47c1df5cd8 # v2.3.2
       with:
         tag_name: v${{ env.XEMU_VERSION }}
         name: v${{ env.XEMU_VERSION }}
@@ -457,7 +462,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
     - name: Download source package
-      uses: actions/download-artifact@95815c38cf2ff2164869cbab79da8d1f422bc89e # v4
+      uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4
       with:
         name: src.tar.gz
     - name: Extract source package
@@ -466,7 +471,7 @@ jobs:
         tar -C src -xf src.tar.gz
 
         # Ensure subprojects are uploaded
-        rm src/subprojects/.gitignore
+        find src/subprojects -name "*.gitignore" -exec rm {} \;
     - name: Integrate Debian packaging
       run: |
         pushd src

.github/workflows/bump-subproject-wraps.yml

@@ -0,0 +1,74 @@
+name: Bump Meson subprojects
+
+on:
+  workflow_dispatch:
+  schedule:
+    - cron: '0 6 * * 1'
+
+permissions:
+  contents: write
+  pull-requests: write
+
+jobs:
+  bump_wraps:
+    name: "Bump Meson subprojects"
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4
+
+      - name: Install the latest version of uv
+        uses: astral-sh/setup-uv@bd01e18f51369d5a26f1651c3cb451d3417e3bba # v6
+        with:
+          enable-cache: false
+
+      - name: Check for updates
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: |
+          uv run -s scripts/bump-subproject-wraps.py -m \
+              subprojects/curl.wrap \
+              subprojects/genconfig.wrap \
+              subprojects/glslang.wrap \
+              subprojects/imgui.wrap \
+              subprojects/implot.wrap \
+              subprojects/json.wrap \
+              subprojects/nv2a_vsh_cpu.wrap \
+              subprojects/SPIRV-Reflect.wrap \
+              subprojects/tomlplusplus.wrap \
+              subprojects/volk.wrap \
+              subprojects/VulkanMemoryAllocator.wrap \
+              subprojects/xxhash.wrap \
+            > updated.json
+
+      - name: Create PRs for updates
+        env:
+          GH_TOKEN: ${{ secrets.XEMU_ROBOT_TOKEN }}
+        run: |
+          set -euo pipefail
+          git config user.name  "xemu-robot"
+          git config user.email "robot@xemu.app"
+
+          jq -c '.[]' updated.json | while read -r item; do
+            path=$(echo "$item" | jq -r '.path')
+            file_basename=$(basename "$path")
+            name="${file_basename%%.*}"
+
+            owner=$(echo "$item" | jq -r '.owner')
+            repo=$(echo "$item" | jq -r '.repo')
+            old_rev=$(echo "$item" | jq -r '.old_rev')
+            new_rev=$(echo "$item" | jq -r '.new_rev')
+            new_tag=$(echo "$item" | jq -r '.new_tag')
+
+            echo "➤ Processing $name"
+            branch="sync/bump-${name//\//-}-${GITHUB_RUN_ID}"
+
+            git switch --quiet -c "$branch" origin/master
+            git add "$path"
+            git commit -m "meson: Bump ${name} to ${new_tag}"
+            git push -u origin "$branch"
+
+            gh pr create \
+              --title "meson: Bump ${name} to ${new_tag}" \
+              --body  "Automatic bump of \`${name}\` to [${new_tag}](https://github.com/${owner}/${repo}/compare/${old_rev}..${new_rev})." \
+              --base  master
+          done

build.sh

@@ -66,8 +66,14 @@ package_macos() {
 
     cp Info.plist dist/xemu.app/Contents/
 
-    plutil -replace CFBundleShortVersionString -string $(cat ${project_source_dir}/XEMU_VERSION | cut -f1 -d-) dist/xemu.app/Contents/Info.plist
-    plutil -replace CFBundleVersion            -string $(cat ${project_source_dir}/XEMU_VERSION | cut -f1 -d-) dist/xemu.app/Contents/Info.plist
+    if [[ -e "${project_source_dir}/XEMU_VERSION" ]]; then
+      xemu_version="$(cat ${project_source_dir}/XEMU_VERSION | cut -f1 -d-)"
+    else
+      xemu_version="0.0.0"
+    fi
+
+    plutil -replace CFBundleShortVersionString -string "${xemu_version}" dist/xemu.app/Contents/Info.plist
+    plutil -replace CFBundleVersion            -string "${xemu_version}" dist/xemu.app/Contents/Info.plist
 
     codesign --force --deep --preserve-metadata=entitlements,requirements,flags,runtime --sign - "${exe_path}"
     python3 ./scripts/gen-license.py --version-file=macos-libs/$target_arch/INSTALLED > dist/LICENSE.txt

config_spec.yml

@@ -54,6 +54,9 @@ input:
   auto_bind:
     type: bool
     default: true
+  allow_vibration:
+    type: bool
+    default: true
   background_input_capture: bool
   keyboard_controller_scancode_map:
     # Scancode reference : https://github.com/libsdl-org/SDL/blob/main/include/SDL_scancode.h
@@ -209,9 +212,19 @@ display:
       advanced_tree_state:
         type: bool
         default: false
+  setup_nvidia_profile:
+    type: bool
+    default: true
 
 audio:
+  vp:
+    num_workers:
+      type: integer
+      default: 0  # 0 = auto
   use_dsp: bool
+  hrtf:
+    type: bool
+    default: true
   volume_limit:
     type: number
     default: 1

debian/control

@@ -6,6 +6,8 @@ Build-Depends: debhelper (>= 11),
  cmake,
  git,
  python3:any,
+ python3-pip,
+ python3-tomli,
  python3-yaml,
  python3-venv,
  ninja-build,

hw/xbox/mcpx/apu/apu.c

@@ -0,0 +1,560 @@
+/*
+ * QEMU MCPX Audio Processing Unit implementation
+ *
+ * Copyright (c) 2012 espes
+ * Copyright (c) 2018-2019 Jannik Vogel
+ * Copyright (c) 2019-2025 Matt Borgerson
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "apu_int.h"
+
+MCPXAPUState *g_state; // Used via debug handlers
+
+static void update_irq(MCPXAPUState *d)
+{
+    if (d->regs[NV_PAPU_FECTL] & NV_PAPU_FECTL_FEMETHMODE_TRAPPED) {
+        qatomic_or(&d->regs[NV_PAPU_ISTS], NV_PAPU_ISTS_FETINTSTS);
+    }
+    if ((d->regs[NV_PAPU_IEN] & NV_PAPU_ISTS_GINTSTS) &&
+        ((d->regs[NV_PAPU_ISTS] & ~NV_PAPU_ISTS_GINTSTS) &
+         d->regs[NV_PAPU_IEN])) {
+        qatomic_or(&d->regs[NV_PAPU_ISTS], NV_PAPU_ISTS_GINTSTS);
+        // fprintf(stderr, "mcpx irq raise ien=%08x ists=%08x\n",
+        //         d->regs[NV_PAPU_IEN], d->regs[NV_PAPU_ISTS]);
+        pci_irq_assert(PCI_DEVICE(d));
+    } else {
+        qatomic_and(&d->regs[NV_PAPU_ISTS], ~NV_PAPU_ISTS_GINTSTS);
+        // fprintf(stderr, "mcpx irq lower ien=%08x ists=%08x\n",
+        //         d->regs[NV_PAPU_IEN], d->regs[NV_PAPU_ISTS]);
+        pci_irq_deassert(PCI_DEVICE(d));
+    }
+}
+
+static uint64_t mcpx_apu_read(void *opaque, hwaddr addr, unsigned int size)
+{
+    MCPXAPUState *d = opaque;
+
+    uint64_t r = 0;
+    switch (addr) {
+    case NV_PAPU_XGSCNT:
+        r = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) / 100; //???
+        break;
+    default:
+        if (addr < 0x20000) {
+            r = qatomic_read(&d->regs[addr]);
+        }
+        break;
+    }
+
+    trace_mcpx_apu_reg_read(addr, size, r);
+    return r;
+}
+
+static void mcpx_apu_write(void *opaque, hwaddr addr, uint64_t val,
+                           unsigned int size)
+{
+    MCPXAPUState *d = opaque;
+
+    trace_mcpx_apu_reg_write(addr, size, val);
+
+    switch (addr) {
+    case NV_PAPU_ISTS:
+        /* the bits of the interrupts to clear are written */
+        qatomic_and(&d->regs[NV_PAPU_ISTS], ~val);
+        update_irq(d);
+        qemu_cond_broadcast(&d->cond);
+        break;
+    case NV_PAPU_FECTL:
+    case NV_PAPU_SECTL:
+        qatomic_set(&d->regs[addr], val);
+        qemu_cond_broadcast(&d->cond);
+        break;
+    case NV_PAPU_FEMEMDATA:
+        /* 'magic write'
+         * This value is expected to be written to FEMEMADDR on completion of
+         * something to do with notifies. Just do it now :/ */
+        stl_le_phys(&address_space_memory, d->regs[NV_PAPU_FEMEMADDR], val);
+        // fprintf(stderr, "MAGIC WRITE\n");
+        qatomic_set(&d->regs[addr], val);
+        break;
+    default:
+        if (addr < 0x20000) {
+            qatomic_set(&d->regs[addr], val);
+        }
+        break;
+    }
+}
+
+static const MemoryRegionOps mcpx_apu_mmio_ops = {
+    .read = mcpx_apu_read,
+    .write = mcpx_apu_write,
+};
+
+static void se_frame(MCPXAPUState *d)
+{
+    mcpx_apu_update_dsp_preference(d);
+    mcpx_debug_begin_frame();
+    g_dbg.gp_realtime = d->gp.realtime;
+    g_dbg.ep_realtime = d->ep.realtime;
+
+    qemu_spin_lock(&d->monitor.fifo_lock);
+    int num_bytes_free = fifo8_num_free(&d->monitor.fifo);
+    qemu_spin_unlock(&d->monitor.fifo_lock);
+
+    /* A rudimentary calculation to determine approximately how taxed the APU
+     * thread is, by measuring how much time we spend waiting for FIFO to drain
+     * versus working on building frames.
+     * =1: thread is not sleeping and likely falling behind realtime
+     * <1: thread is able to complete work on time
+     */
+    if (num_bytes_free < sizeof(d->monitor.frame_buf)) {
+        int64_t sleep_start = qemu_clock_get_us(QEMU_CLOCK_REALTIME);
+        qemu_cond_wait(&d->cond, &d->lock);
+        int64_t sleep_end = qemu_clock_get_us(QEMU_CLOCK_REALTIME);
+        d->sleep_acc += (sleep_end - sleep_start);
+        return;
+    }
+    int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
+    if (now - d->frame_count_time >= 1000) {
+        g_dbg.frames_processed = d->frame_count;
+        float t = 1.0f - ((double)d->sleep_acc /
+                          (double)((now - d->frame_count_time) * 1000));
+        g_dbg.utilization = t;
+
+        d->frame_count_time = now;
+        d->frame_count = 0;
+        d->sleep_acc = 0;
+    }
+    d->frame_count++;
+
+    /* Buffer for all mixbins for this frame */
+    float mixbins[NUM_MIXBINS][NUM_SAMPLES_PER_FRAME] = { 0 };
+
+    mcpx_apu_vp_frame(d, mixbins);
+    mcpx_apu_dsp_frame(d, mixbins);
+
+    if ((d->ep_frame_div + 1) % 8 == 0) {
+#if 0
+        FILE *fd = fopen("ep.pcm", "a+");
+        assert(fd != NULL);
+        fwrite(d->apu_fifo_output, sizeof(d->apu_fifo_output), 1, fd);
+        fclose(fd);
+#endif
+
+        if (0 <= g_config.audio.volume_limit && g_config.audio.volume_limit < 1) {
+            float f = pow(g_config.audio.volume_limit, M_E);
+            for (int i = 0; i < 256; i++) {
+                d->monitor.frame_buf[i][0] *= f;
+                d->monitor.frame_buf[i][1] *= f;
+            }
+        }
+
+        qemu_spin_lock(&d->monitor.fifo_lock);
+        num_bytes_free = fifo8_num_free(&d->monitor.fifo);
+        assert(num_bytes_free >= sizeof(d->monitor.frame_buf));
+        fifo8_push_all(&d->monitor.fifo, (uint8_t *)d->monitor.frame_buf,
+                       sizeof(d->monitor.frame_buf));
+        qemu_spin_unlock(&d->monitor.fifo_lock);
+        memset(d->monitor.frame_buf, 0, sizeof(d->monitor.frame_buf));
+    }
+
+    d->ep_frame_div++;
+
+    mcpx_debug_end_frame();
+}
+
+/* Note: only supports millisecond resolution on Windows */
+static void sleep_ns(int64_t ns)
+{
+#ifndef _WIN32
+        struct timespec sleep_delay, rem_delay;
+        sleep_delay.tv_sec = ns / 1000000000LL;
+        sleep_delay.tv_nsec = ns % 1000000000LL;
+        nanosleep(&sleep_delay, &rem_delay);
+#else
+        Sleep(ns / SCALE_MS);
+#endif
+}
+
+static void monitor_sink_cb(void *opaque, uint8_t *stream, int free_b)
+{
+    MCPXAPUState *s = MCPX_APU_DEVICE(opaque);
+
+    if (!runstate_is_running()) {
+        memset(stream, 0, free_b);
+        return;
+    }
+
+    int avail = 0;
+    while (avail < free_b) {
+        qemu_spin_lock(&s->monitor.fifo_lock);
+        avail = fifo8_num_used(&s->monitor.fifo);
+        qemu_spin_unlock(&s->monitor.fifo_lock);
+        if (avail < free_b) {
+            sleep_ns(1000000);
+            qemu_cond_broadcast(&s->cond);
+        }
+    }
+
+    int to_copy = MIN(free_b, avail);
+    while (to_copy > 0) {
+        uint32_t chunk_len = 0;
+        qemu_spin_lock(&s->monitor.fifo_lock);
+        chunk_len = fifo8_pop_buf(&s->monitor.fifo, stream, to_copy);
+        assert(chunk_len <= to_copy);
+        qemu_spin_unlock(&s->monitor.fifo_lock);
+        stream += chunk_len;
+        to_copy -= chunk_len;
+    }
+
+    qemu_cond_broadcast(&s->cond);
+}
+
+static void monitor_init(MCPXAPUState *d)
+{
+    qemu_spin_init(&d->monitor.fifo_lock);
+    fifo8_create(&d->monitor.fifo, 3 * (256 * 2 * 2));
+
+    struct SDL_AudioSpec sdl_audio_spec = {
+        .freq = 48000,
+        .format = AUDIO_S16LSB,
+        .channels = 2,
+        .samples = 512,
+        .callback = monitor_sink_cb,
+        .userdata = d,
+    };
+
+    if (SDL_Init(SDL_INIT_AUDIO) < 0)  {
+        fprintf(stderr, "Failed to initialize SDL audio subsystem: %s\n", SDL_GetError());
+        exit(1);
+    }
+
+    SDL_AudioDeviceID sdl_audio_dev;
+    sdl_audio_dev = SDL_OpenAudioDevice(NULL, 0, &sdl_audio_spec, NULL, 0);
+    if (sdl_audio_dev == 0) {
+        fprintf(stderr, "SDL_OpenAudioDevice failed: %s\n", SDL_GetError());
+        assert(!"SDL_OpenAudioDevice failed");
+        exit(1);
+    }
+    SDL_PauseAudioDevice(sdl_audio_dev, 0);
+}
+
+static void mcpx_apu_realize(PCIDevice *dev, Error **errp)
+{
+    MCPXAPUState *d = MCPX_APU_DEVICE(dev);
+
+    dev->config[PCI_INTERRUPT_PIN] = 0x01;
+
+    memory_region_init_io(&d->mmio, OBJECT(dev), &mcpx_apu_mmio_ops, d,
+                          "mcpx-apu-mmio", 0x80000);
+
+    memory_region_init_io(&d->vp.mmio, OBJECT(dev), &vp_ops, d,
+                          "mcpx-apu-vp", 0x10000);
+    memory_region_add_subregion(&d->mmio, 0x20000, &d->vp.mmio);
+
+    memory_region_init_io(&d->gp.mmio, OBJECT(dev), &gp_ops, d,
+                          "mcpx-apu-gp", 0x10000);
+    memory_region_add_subregion(&d->mmio, 0x30000, &d->gp.mmio);
+
+    memory_region_init_io(&d->ep.mmio, OBJECT(dev), &ep_ops, d,
+                          "mcpx-apu-ep", 0x10000);
+    memory_region_add_subregion(&d->mmio, 0x50000, &d->ep.mmio);
+
+    pci_register_bar(dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY, &d->mmio);
+}
+
+static void mcpx_apu_exitfn(PCIDevice *dev)
+{
+    MCPXAPUState *d = MCPX_APU_DEVICE(dev);
+    d->exiting = true;
+    qemu_cond_broadcast(&d->cond);
+    qemu_thread_join(&d->apu_thread);
+    mcpx_apu_vp_finalize(d);
+}
+
+static void mcpx_apu_reset(MCPXAPUState *d)
+{
+    qemu_mutex_lock(&d->lock); // FIXME: Can fail if thread is pegged, add flag
+    memset(d->regs, 0, sizeof(d->regs));
+
+    mcpx_apu_vp_reset(d);
+
+    // FIXME: Reset DSP state
+    memset(d->gp.dsp->core.pram_opcache, 0,
+           sizeof(d->gp.dsp->core.pram_opcache));
+    memset(d->ep.dsp->core.pram_opcache, 0,
+           sizeof(d->ep.dsp->core.pram_opcache));
+    d->set_irq = false;
+    qemu_cond_signal(&d->cond);
+    qemu_mutex_unlock(&d->lock);
+}
+
+// Note: This is handled as a VM state change and not as a `pre_save` callback
+// because we want to halt the FIFO before any VM state is saved/restored to
+// avoid corruption.
+static void mcpx_apu_vm_state_change(void *opaque, bool running, RunState state)
+{
+    MCPXAPUState *d = opaque;
+
+    if (state == RUN_STATE_SAVE_VM) {
+        qemu_mutex_lock(&d->lock);
+    }
+}
+
+static int mcpx_apu_post_save(void *opaque)
+{
+    MCPXAPUState *d = opaque;
+    qemu_cond_signal(&d->cond);
+    qemu_mutex_unlock(&d->lock);
+    return 0;
+}
+
+static int mcpx_apu_pre_load(void *opaque)
+{
+    MCPXAPUState *d = opaque;
+    mcpx_apu_reset(d);
+    qemu_mutex_lock(&d->lock);
+    return 0;
+}
+
+static int mcpx_apu_post_load(void *opaque, int version_id)
+{
+    MCPXAPUState *d = opaque;
+    qemu_cond_signal(&d->cond);
+    qemu_mutex_unlock(&d->lock);
+    return 0;
+}
+
+static void mcpx_apu_reset_hold(Object *obj, ResetType type)
+{
+    MCPXAPUState *d = MCPX_APU_DEVICE(obj);
+    mcpx_apu_reset(d);
+}
+
+const VMStateDescription vmstate_vp_dsp_dma_state = {
+    .name = "mcpx-apu/dsp-state/dma",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .fields      = (VMStateField[]) {
+        VMSTATE_UINT32(configuration, DSPDMAState),
+        VMSTATE_UINT32(control, DSPDMAState),
+        VMSTATE_UINT32(start_block, DSPDMAState),
+        VMSTATE_UINT32(next_block, DSPDMAState),
+        VMSTATE_BOOL(error, DSPDMAState),
+        VMSTATE_BOOL(eol, DSPDMAState),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
+const VMStateDescription vmstate_vp_dsp_core_state = {
+    .name = "mcpx-apu/dsp-state/core",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .fields      = (VMStateField[]) {
+        // FIXME: Remove unnecessary fields
+        VMSTATE_UINT16(instr_cycle, dsp_core_t),
+        VMSTATE_UINT32(pc, dsp_core_t),
+        VMSTATE_UINT32_ARRAY(registers, dsp_core_t, DSP_REG_MAX),
+        VMSTATE_UINT32_2DARRAY(stack, dsp_core_t, 2, 16),
+        VMSTATE_UINT32_ARRAY(xram, dsp_core_t, DSP_XRAM_SIZE),
+        VMSTATE_UINT32_ARRAY(yram, dsp_core_t, DSP_YRAM_SIZE),
+        VMSTATE_UINT32_ARRAY(pram, dsp_core_t, DSP_PRAM_SIZE),
+        VMSTATE_UINT32_ARRAY(mixbuffer, dsp_core_t, DSP_MIXBUFFER_SIZE),
+        VMSTATE_UINT32_ARRAY(periph, dsp_core_t, DSP_PERIPH_SIZE),
+        VMSTATE_UINT32(loop_rep, dsp_core_t),
+        VMSTATE_UINT32(pc_on_rep, dsp_core_t),
+        VMSTATE_UINT16(interrupt_state, dsp_core_t),
+        VMSTATE_UINT16(interrupt_instr_fetch, dsp_core_t),
+        VMSTATE_UINT16(interrupt_save_pc, dsp_core_t),
+        VMSTATE_UINT16(interrupt_counter, dsp_core_t),
+        VMSTATE_UINT16(interrupt_ipl_to_raise, dsp_core_t),
+        VMSTATE_UINT16(interrupt_pipeline_count, dsp_core_t),
+        VMSTATE_INT16_ARRAY(interrupt_ipl, dsp_core_t, 12),
+        VMSTATE_UINT16_ARRAY(interrupt_is_pending, dsp_core_t, 12),
+        VMSTATE_UINT32(num_inst, dsp_core_t),
+        VMSTATE_UINT32(cur_inst_len, dsp_core_t),
+        VMSTATE_UINT32(cur_inst, dsp_core_t),
+        VMSTATE_UNUSED(1),
+        VMSTATE_UINT32(disasm_memory_ptr, dsp_core_t),
+        VMSTATE_BOOL(exception_debugging, dsp_core_t),
+        VMSTATE_UINT32(disasm_prev_inst_pc, dsp_core_t),
+        VMSTATE_BOOL(disasm_is_looping, dsp_core_t),
+        VMSTATE_UINT32(disasm_cur_inst, dsp_core_t),
+        VMSTATE_UINT16(disasm_cur_inst_len, dsp_core_t),
+        VMSTATE_UINT32_ARRAY(disasm_registers_save, dsp_core_t, 64),
+// #ifdef DSP_DISASM_REG_PC
+//         VMSTATE_UINT32(pc_save, dsp_core_t),
+// #endif
+        VMSTATE_END_OF_LIST()
+    }
+};
+
+const VMStateDescription vmstate_vp_dsp_state = {
+    .name = "mcpx-apu/dsp-state",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .fields = (VMStateField[]) {
+        VMSTATE_STRUCT(core, DSPState, 1, vmstate_vp_dsp_core_state, dsp_core_t),
+        VMSTATE_STRUCT(dma, DSPState, 1, vmstate_vp_dsp_dma_state, DSPDMAState),
+        VMSTATE_INT32(save_cycles, DSPState),
+        VMSTATE_UINT32(interrupts, DSPState),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
+
+const VMStateDescription vmstate_vp_ssl_data = {
+    .name = "mcpx_apu_voice_data",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .fields = (VMStateField[]) {
+        VMSTATE_UINT32_ARRAY(base, MCPXAPUVPSSLData, MCPX_HW_SSLS_PER_VOICE),
+        VMSTATE_UINT8_ARRAY(count, MCPXAPUVPSSLData, MCPX_HW_SSLS_PER_VOICE),
+        VMSTATE_INT32(ssl_index, MCPXAPUVPSSLData),
+        VMSTATE_INT32(ssl_seg, MCPXAPUVPSSLData),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
+static const VMStateDescription vmstate_mcpx_apu = {
+    .name = "mcpx-apu",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .post_save = mcpx_apu_post_save,
+    .pre_load = mcpx_apu_pre_load,
+    .post_load = mcpx_apu_post_load,
+    .fields = (VMStateField[]) {
+        VMSTATE_PCI_DEVICE(parent_obj, MCPXAPUState),
+        VMSTATE_STRUCT_POINTER(gp.dsp, MCPXAPUState, vmstate_vp_dsp_state,
+                               DSPState),
+        VMSTATE_UINT32_ARRAY(gp.regs, MCPXAPUState, 0x10000),
+        VMSTATE_STRUCT_POINTER(ep.dsp, MCPXAPUState, vmstate_vp_dsp_state,
+                               DSPState),
+        VMSTATE_UINT32_ARRAY(ep.regs, MCPXAPUState, 0x10000),
+        VMSTATE_UINT32_ARRAY(regs, MCPXAPUState, 0x20000),
+        VMSTATE_UINT32(vp.inbuf_sge_handle, MCPXAPUState),
+        VMSTATE_UINT32(vp.outbuf_sge_handle, MCPXAPUState),
+        VMSTATE_STRUCT_ARRAY(vp.ssl, MCPXAPUState, MCPX_HW_MAX_VOICES, 1,
+                             vmstate_vp_ssl_data, MCPXAPUVPSSLData),
+        VMSTATE_INT32(vp.ssl_base_page, MCPXAPUState),
+        VMSTATE_UINT8_ARRAY(vp.hrtf_submix, MCPXAPUState, 4),
+        VMSTATE_UINT8(vp.hrtf_headroom, MCPXAPUState),
+        VMSTATE_UINT8_ARRAY(vp.submix_headroom, MCPXAPUState, NUM_MIXBINS),
+        VMSTATE_UINT64_ARRAY(vp.voice_locked, MCPXAPUState, 4),
+        VMSTATE_END_OF_LIST()
+    },
+};
+
+static void mcpx_apu_class_init(ObjectClass *klass, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+    ResettableClass *rc = RESETTABLE_CLASS(klass);
+    PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
+
+    k->vendor_id = PCI_VENDOR_ID_NVIDIA;
+    k->device_id = PCI_DEVICE_ID_NVIDIA_MCPX_APU;
+    k->revision = 177;
+    k->class_id = PCI_CLASS_MULTIMEDIA_AUDIO;
+    k->realize = mcpx_apu_realize;
+    k->exit = mcpx_apu_exitfn;
+
+    rc->phases.hold = mcpx_apu_reset_hold;
+
+    dc->desc = "MCPX Audio Processing Unit";
+    dc->vmsd = &vmstate_mcpx_apu;
+}
+
+static const TypeInfo mcpx_apu_info = {
+    .name = "mcpx-apu",
+    .parent = TYPE_PCI_DEVICE,
+    .instance_size = sizeof(MCPXAPUState),
+    .class_init = mcpx_apu_class_init,
+    .interfaces =
+        (InterfaceInfo[]){
+            { INTERFACE_CONVENTIONAL_PCI_DEVICE },
+            {},
+        },
+};
+
+static void mcpx_apu_register(void)
+{
+    type_register_static(&mcpx_apu_info);
+}
+type_init(mcpx_apu_register);
+
+static void *mcpx_apu_frame_thread(void *arg)
+{
+    MCPXAPUState *d = MCPX_APU_DEVICE(arg);
+    qemu_mutex_lock(&d->lock);
+    while (!qatomic_read(&d->exiting)) {
+        int xcntmode = GET_MASK(qatomic_read(&d->regs[NV_PAPU_SECTL]),
+                                NV_PAPU_SECTL_XCNTMODE);
+        uint32_t fectl = qatomic_read(&d->regs[NV_PAPU_FECTL]);
+        if (xcntmode == NV_PAPU_SECTL_XCNTMODE_OFF ||
+            (fectl & NV_PAPU_FECTL_FEMETHMODE_TRAPPED) ||
+            (fectl & NV_PAPU_FECTL_FEMETHMODE_HALTED)) {
+            d->set_irq = true;
+        }
+
+        if (d->set_irq) {
+            qemu_mutex_unlock(&d->lock);
+            bql_lock();
+            update_irq(d);
+            bql_unlock();
+            qemu_mutex_lock(&d->lock);
+            d->set_irq = false;
+        }
+
+        xcntmode = GET_MASK(qatomic_read(&d->regs[NV_PAPU_SECTL]),
+                            NV_PAPU_SECTL_XCNTMODE);
+        fectl = qatomic_read(&d->regs[NV_PAPU_FECTL]);
+        if (xcntmode == NV_PAPU_SECTL_XCNTMODE_OFF ||
+            (fectl & NV_PAPU_FECTL_FEMETHMODE_TRAPPED) ||
+            (fectl & NV_PAPU_FECTL_FEMETHMODE_HALTED)) {
+            qemu_cond_wait(&d->cond, &d->lock);
+            continue;
+        }
+        se_frame((void *)d);
+    }
+    qemu_mutex_unlock(&d->lock);
+    return NULL;
+}
+
+void mcpx_apu_init(PCIBus *bus, int devfn, MemoryRegion *ram)
+{
+    PCIDevice *dev = pci_create_simple(bus, devfn, "mcpx-apu");
+    MCPXAPUState *d = MCPX_APU_DEVICE(dev);
+
+    g_state = d;
+
+    d->ram = ram;
+    d->ram_ptr = memory_region_get_ram_ptr(d->ram);
+
+    mcpx_apu_dsp_init(d);
+
+    d->set_irq = false;
+    d->exiting = false;
+
+    qemu_mutex_init(&d->lock);
+    qemu_cond_init(&d->cond);
+    qemu_add_vm_change_state_handler(mcpx_apu_vm_state_change, d);
+
+    mcpx_apu_vp_init(d);
+    qemu_thread_create(&d->apu_thread, "mcpx.apu_thread", mcpx_apu_frame_thread,
+                       d, QEMU_THREAD_JOINABLE);
+
+    monitor_init(d);
+}

hw/xbox/mcpx/apu/apu_debug.h

@@ -23,13 +23,15 @@
 #include <stdbool.h>
 #include <stdint.h>
 
-enum McpxApuDebugMon {
+#define MAX_VOICE_WORKERS 16
+
+typedef enum McpxApuDebugMonitorPoint {
     MCPX_APU_DEBUG_MON_AC97,
     MCPX_APU_DEBUG_MON_VP,
     MCPX_APU_DEBUG_MON_GP,
     MCPX_APU_DEBUG_MON_EP,
     MCPX_APU_DEBUG_MON_GP_OR_EP
-};
+} McpxApuDebugMonitorPoint;
 
 struct McpxApuDebugVoice
 {
@@ -55,6 +57,12 @@ struct McpxApuDebugVoice
 struct McpxApuDebugVp
 {
     struct McpxApuDebugVoice v[256];
+    int num_workers;
+    struct {
+        int num_voices;
+        int time_us;
+    } workers[MAX_VOICE_WORKERS];
+    int total_worker_time_us;
 };
 
 struct McpxApuDebugDsp
@@ -76,8 +84,8 @@ extern "C" {
 #endif
 
 const struct McpxApuDebug *mcpx_apu_get_debug_info(void);
-int mcpx_apu_debug_get_monitor(void);
-void mcpx_apu_debug_set_monitor(int mon);
+McpxApuDebugMonitorPoint mcpx_apu_debug_get_monitor(void);
+void mcpx_apu_debug_set_monitor(McpxApuDebugMonitorPoint monitor);
 void mcpx_apu_debug_isolate_voice(uint16_t v);
 void mcpx_apu_debug_clear_isolations(void);
 void mcpx_apu_debug_toggle_mute(uint16_t v);

hw/xbox/mcpx/apu/apu_int.h

@@ -0,0 +1,117 @@
+/*
+ * QEMU MCPX Audio Processing Unit implementation
+ *
+ * Copyright (c) 2012 espes
+ * Copyright (c) 2018-2019 Jannik Vogel
+ * Copyright (c) 2019-2025 Matt Borgerson
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, see <http://www.gnu.org/licenses/>.
+ */
+#ifndef HW_XBOX_MCPX_APU_INT_H
+#define HW_XBOX_MCPX_APU_INT_H
+
+#include "qemu/osdep.h"
+#include <math.h>
+#include <SDL.h>
+#include "hw/hw.h"
+#include "hw/pci/pci.h"
+#include "hw/pci/pci_device.h"
+#include "cpu.h"
+#include "migration/vmstate.h"
+#include "qemu/main-loop.h"
+#include "qemu/thread.h"
+#include "sysemu/runstate.h"
+#include "audio/audio.h"
+#include "qemu/fifo8.h"
+#include "ui/xemu-settings.h"
+
+#include "trace.h"
+#include "apu.h"
+#include "apu_regs.h"
+#include "apu_debug.h"
+#include "fpconv.h"
+#include "vp/vp.h"
+#include "dsp/gp_ep.h"
+
+#define GET_MASK(v, mask) (((v) & (mask)) >> ctz32(mask))
+
+#define SET_MASK(v, mask, val)                                       \
+    do {                                                             \
+        (v) &= ~(mask);                                              \
+        (v) |= ((val) << ctz32(mask)) & (mask);                      \
+    } while (0)
+
+#define CASE_4(v, step)                                              \
+    case (v):                                                        \
+    case (v)+(step):                                                 \
+    case (v)+(step)*2:                                               \
+    case (v)+(step)*3
+
+// #define DEBUG_MCPX
+
+#ifdef DEBUG_MCPX
+#define DPRINTF(fmt, ...) \
+    do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
+#else
+#define DPRINTF(fmt, ...) \
+    do { } while (0)
+#endif
+
+#define MCPX_APU_DEVICE(obj) \
+    OBJECT_CHECK(MCPXAPUState, (obj), "mcpx-apu")
+
+typedef struct MCPXAPUState {
+    /*< private >*/
+    PCIDevice parent_obj;
+    /*< public >*/
+
+    bool exiting;
+    bool set_irq;
+
+    QemuThread apu_thread;
+    QemuMutex lock;
+    QemuCond cond;
+
+    MemoryRegion *ram;
+    uint8_t *ram_ptr;
+    MemoryRegion mmio;
+
+    MCPXAPUVPState vp;
+    MCPXAPUGPState gp;
+    MCPXAPUEPState ep;
+
+    uint32_t regs[0x20000];
+
+    int ep_frame_div;
+    int sleep_acc;
+    int frame_count;
+    int64_t frame_count_time;
+
+    struct {
+        McpxApuDebugMonitorPoint point;
+        int16_t frame_buf[256][2]; // 1 EP frame (0x400 bytes), 8 buffered
+        QemuSpin fifo_lock;
+        Fifo8 fifo;
+    } monitor;
+} MCPXAPUState;
+
+extern MCPXAPUState *g_state; // Used via debug handlers
+extern struct McpxApuDebug g_dbg, g_dbg_cache;
+extern int g_dbg_voice_monitor;
+extern uint64_t g_dbg_muted_voices[4];
+
+void mcpx_debug_begin_frame(void);
+void mcpx_debug_end_frame(void);
+
+#endif

hw/xbox/mcpx/apu/apu_regs.h

@@ -145,6 +145,8 @@
 #define NV1BA0_PIO_VOICE_PAUSE                           0x00000140
 #   define NV1BA0_PIO_VOICE_PAUSE_HANDLE                    0x0000FFFF
 #   define NV1BA0_PIO_VOICE_PAUSE_ACTION                    (1 << 18)
+#define NV1BA0_PIO_SET_CURRENT_HRTF_ENTRY                0x00000160
+#   define NV1BA0_PIO_SET_CURRENT_HRTF_ENTRY_HANDLE         0x0000FFFF
 #define NV1BA0_PIO_SET_CONTEXT_DMA_NOTIFY                0x00000180
 #define NV1BA0_PIO_SET_CURRENT_SSL_CONTEXT_DMA           0x0000018C
 #define NV1BA0_PIO_SET_CURRENT_SSL                       0x00000190
@@ -153,7 +155,7 @@
 #define NV1BA0_PIO_SET_SSL_SEGMENT_LENGTH                0x00000604
 #define NV1BA0_PIO_SET_SUBMIX_HEADROOM                   0x00000200
 #   define NV1BA0_PIO_SET_SUBMIX_HEADROOM_AMOUNT            0x7
-#define NV1BA0_PIO_SET_HRTF_HEADROOM                       0x00000280
+#define NV1BA0_PIO_SET_HRTF_HEADROOM                     0x00000280
 #   define NV1BA0_PIO_SET_HRTF_HEADROOM_AMOUNT              0x7
 #define NV1BA0_PIO_SET_HRTF_SUBMIXES                     0x000002C0
 #define NV1BA0_PIO_SET_CURRENT_VOICE                     0x000002F8
@@ -165,6 +167,8 @@
 #define NV1BA0_PIO_SET_VOICE_CFG_ENV1                    0x00000310
 #define NV1BA0_PIO_SET_VOICE_CFG_ENVF                    0x00000314
 #define NV1BA0_PIO_SET_VOICE_CFG_MISC                    0x00000318
+#define NV1BA0_PIO_SET_VOICE_TAR_HRTF                    0x0000031C
+#   define NV1BA0_PIO_SET_VOICE_TAR_HRTF_HANDLE             0x0000FFFF
 #define NV1BA0_PIO_SET_VOICE_SSL_A                       0x00000320
 #   define NV1BA0_PIO_SET_VOICE_SSL_A_COUNT                 0x000000FF
 #   define NV1BA0_PIO_SET_VOICE_SSL_A_BASE                  0xFFFFFF00
@@ -185,6 +189,15 @@
 #   define NV1BA0_PIO_SET_VOICE_BUF_CBO_OFFSET              0x00FFFFFF
 #define NV1BA0_PIO_SET_VOICE_CFG_BUF_EBO                 0x000003DC
 #   define NV1BA0_PIO_SET_VOICE_CFG_BUF_EBO_OFFSET          0x00FFFFFF
+#define NV1BA0_PIO_SET_HRIR                              0x00000400
+#   define NV1BA0_PIO_SET_HRIR_LEFT0                        0x000000FF
+#   define NV1BA0_PIO_SET_HRIR_RIGHT0                       0x0000FF00
+#   define NV1BA0_PIO_SET_HRIR_LEFT1                        0x00FF0000
+#   define NV1BA0_PIO_SET_HRIR_RIGHT1                       0xFF000000
+#define NV1BA0_PIO_SET_HRIR_X                            0x0000043C
+#    define NV1BA0_PIO_SET_HRIR_X_LEFT30                     0x000000FF
+#    define NV1BA0_PIO_SET_HRIR_X_RIGHT30                    0x0000FF00
+#    define NV1BA0_PIO_SET_HRIR_X_ITD                        0xFFFF0000
 #define NV1BA0_PIO_SET_SSL_SEGMENT_OFFSET                0x00000600
 #define NV1BA0_PIO_SET_SSL_SEGMENT_LENGTH                0x00000604
 #define NV1BA0_PIO_SET_CURRENT_INBUF_SGE                 0x00000804
@@ -249,6 +262,8 @@
 #define NV_PAVS_VOICE_CFG_MISC                           0x00000018
 #   define NV_PAVS_VOICE_CFG_MISC_EF_RELEASERATE            (0xFFF << 0)
 #   define NV_PAVS_VOICE_CFG_MISC_FMODE                     (3 << 16)
+#define NV_PAVS_VOICE_CFG_HRTF_TARGET                    0x0000001C
+#   define NV_PAVS_VOICE_CFG_HRTF_TARGET_HANDLE             0x0000FFFF
 #define NV_PAVS_VOICE_CUR_PSL_START                      0x00000020
 #   define NV_PAVS_VOICE_CUR_PSL_START_BA                   0x00FFFFFF
 #define NV_PAVS_VOICE_CUR_PSH_SAMPLE                     0x00000024
@@ -313,6 +328,7 @@
 #define EP_INPUT_FIFO_COUNT   2
 
 #define MCPX_HW_MAX_VOICES 256
+#define MCPX_HW_MAX_3D_VOICES 64
 
 #define NUM_SAMPLES_PER_FRAME 32
 #define NUM_MIXBINS 32
@@ -336,6 +352,12 @@ enum MCPX_HW_NOTIFIER {
 #define NV1BA0_NOTIFICATION_STATUS_DONE_SUCCESS       0x01
 #define NV1BA0_NOTIFICATION_STATUS_IN_PROGRESS        0x80
 
+#define HRTF_NULL_HANDLE 0xFFFF
+#define HRTF_ENTRY_COUNT 128
+
+#define MULTIPASS_BIN      31
+#define MULTIPASS_BIN_MASK (1 << MULTIPASS_BIN)
+
 // clang-format on
 
 #endif

hw/xbox/mcpx/apu/debug.c

@@ -0,0 +1,86 @@
+/*
+ * QEMU MCPX Audio Processing Unit implementation
+ *
+ * Copyright (c) 2012 espes
+ * Copyright (c) 2018-2019 Jannik Vogel
+ * Copyright (c) 2019-2025 Matt Borgerson
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "apu_int.h"
+
+struct McpxApuDebug g_dbg, g_dbg_cache;
+int g_dbg_voice_monitor = -1;
+uint64_t g_dbg_muted_voices[4];
+
+const struct McpxApuDebug *mcpx_apu_get_debug_info(void)
+{
+    return &g_dbg_cache;
+}
+
+void mcpx_debug_begin_frame(void)
+{
+    for (int i = 0; i < MCPX_HW_MAX_VOICES; i++) {
+        g_dbg.vp.v[i].active = false;
+        g_dbg.vp.v[i].multipass_dst_voice = 0xFFFF;
+    }
+}
+
+void mcpx_debug_end_frame(void)
+{
+    g_dbg_cache = g_dbg;
+}
+
+void mcpx_apu_debug_set_gp_realtime_enabled(bool run)
+{
+    g_state->gp.realtime = run;
+}
+
+void mcpx_apu_debug_set_ep_realtime_enabled(bool run)
+{
+    g_state->ep.realtime = run;
+}
+
+McpxApuDebugMonitorPoint mcpx_apu_debug_get_monitor(void)
+{
+    return g_state->monitor.point;
+}
+
+void mcpx_apu_debug_set_monitor(McpxApuDebugMonitorPoint monitor)
+{
+    g_state->monitor.point = monitor;
+}
+
+void mcpx_apu_debug_isolate_voice(uint16_t v)
+{
+    g_dbg_voice_monitor = v;
+}
+
+void mcpx_apu_debug_clear_isolations(void)
+{
+    g_dbg_voice_monitor = -1;
+}
+
+bool mcpx_apu_debug_is_muted(uint16_t v)
+{
+    assert(v < MCPX_HW_MAX_VOICES);
+    return g_dbg_muted_voices[v / 64] & (1LL << (v % 64));
+}
+
+void mcpx_apu_debug_toggle_mute(uint16_t v)
+{
+    assert(v < MCPX_HW_MAX_VOICES);
+    g_dbg_muted_voices[v / 64] ^= (1LL << (v % 64));
+}

hw/xbox/mcpx/apu/dsp/debug.c

@@ -2,6 +2,7 @@
  * MCPX DSP emulator
  *
  * Copyright (c) 2015 espes
+ * Copyright (c) 2020-2025 Matt Borgerson
  *
  * Adapted from Hatari DSP M56001 emulation
  * (C) 2001-2008 ARAnyM developer team
@@ -34,242 +35,11 @@
 #include "dsp_state.h"
 
 #include "dsp.h"
+#include "debug.h"
 
-/* Defines */
+#if DEBUG_DSP
 #define BITMASK(x)  ((1<<(x))-1)
 
-#define INTERRUPT_ABORT_FRAME (1 << 0)
-#define INTERRUPT_START_FRAME (1 << 1)
-#define INTERRUPT_DMA_EOL (1 << 7)
-
-// #define DEBUG_DSP
-
-#ifdef DEBUG_DSP
-#define DPRINTF(fmt, ...) \
-    do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
-#else
-#define DPRINTF(fmt, ...) \
-    do { } while (0)
-#endif
-
-static uint32_t read_peripheral(dsp_core_t* core, uint32_t address);
-static void write_peripheral(dsp_core_t* core, uint32_t address, uint32_t value);
-
-DSPState *dsp_init(void *rw_opaque,
-                   dsp_scratch_rw_func scratch_rw,
-                   dsp_fifo_rw_func fifo_rw)
-{
-    DPRINTF("dsp_init\n");
-
-    DSPState* dsp = (DSPState*)malloc(sizeof(DSPState));
-    memset(dsp, 0, sizeof(*dsp));
-
-    dsp->core.read_peripheral = read_peripheral;
-    dsp->core.write_peripheral = write_peripheral;
-
-    dsp->dma.core = &dsp->core;
-    dsp->dma.rw_opaque = rw_opaque;
-    dsp->dma.scratch_rw = scratch_rw;
-    dsp->dma.fifo_rw = fifo_rw;
-
-    dsp_reset(dsp);
-
-    return dsp;
-}
-
-void dsp_reset(DSPState* dsp)
-{
-    dsp56k_reset_cpu(&dsp->core);
-    dsp->save_cycles = 0;
-}
-
-void dsp_destroy(DSPState* dsp)
-{
-    free(dsp);
-}
-
-static uint32_t read_peripheral(dsp_core_t* core, uint32_t address) {
-    DSPState* dsp = container_of(core, DSPState, core);
-
-    DPRINTF("read_peripheral 0x%06x", address);
-
-    uint32_t v = 0xababa;
-    switch(address) {
-    case 0xFFFFB3:
-        v = 0; // core->num_inst; // ??
-        break;
-    case 0xFFFFC5:
-        v = dsp->interrupts;
-        if (dsp->dma.eol) {
-            v |= INTERRUPT_DMA_EOL;
-        }
-        break;
-    case 0xFFFFD4:
-        v = dsp_dma_read(&dsp->dma, DMA_NEXT_BLOCK);
-        break;
-    case 0xFFFFD5:
-        v = dsp_dma_read(&dsp->dma, DMA_START_BLOCK);
-        break;
-    case 0xFFFFD6:
-        v = dsp_dma_read(&dsp->dma, DMA_CONTROL);
-        break;
-    case 0xFFFFD7:
-        v = dsp_dma_read(&dsp->dma, DMA_CONFIGURATION);
-        break;
-    }
-
-    DPRINTF(" -> 0x%06x\n", v);
-    return v;
-}
-
-static void write_peripheral(dsp_core_t* core, uint32_t address, uint32_t value) {
-    DSPState* dsp = container_of(core, DSPState, core);
-
-    DPRINTF("write_peripheral [0x%06x] = 0x%06x\n", address, value);
-
-    switch(address) {
-    case 0xFFFFC4:
-        if (value & 1) {
-            core->is_idle = true;
-        }
-        break;
-    case 0xFFFFC5:
-        dsp->interrupts &= ~value;
-        if (value & INTERRUPT_DMA_EOL) {
-            dsp->dma.eol = false;
-        }
-        break;
-    case 0xFFFFD4:
-        dsp_dma_write(&dsp->dma, DMA_NEXT_BLOCK, value);
-        break;
-    case 0xFFFFD5:
-        dsp_dma_write(&dsp->dma, DMA_START_BLOCK, value);
-        break;
-    case 0xFFFFD6:
-        dsp_dma_write(&dsp->dma, DMA_CONTROL, value);
-        break;
-    case 0xFFFFD7:
-        dsp_dma_write(&dsp->dma, DMA_CONFIGURATION, value);
-        break;
-    }
-}
-
-
-void dsp_step(DSPState* dsp)
-{
-    dsp56k_execute_instruction(&dsp->core);
-}
-
-void dsp_run(DSPState* dsp, int cycles)
-{
-    dsp->save_cycles += cycles;
-
-    if (dsp->save_cycles <= 0) return;
-
-    int dma_timer = 0;
-
-    while (dsp->save_cycles > 0)
-    {
-        dsp56k_execute_instruction(&dsp->core);
-        dsp->save_cycles -= dsp->core.instr_cycle;
-        dsp->core.cycle_count++;
-
-        if (dsp->dma.control & DMA_CONTROL_RUNNING) {
-            dma_timer++;
-        }
-
-        if (dma_timer > 2) {
-            dma_timer = 0;
-            dsp->dma.control &= ~DMA_CONTROL_RUNNING;
-            dsp->dma.control |= DMA_CONTROL_STOPPED;
-        }
-
-        if (dsp->core.is_idle) break;
-    }
-
-    /* FIXME: DMA timing be done cleaner. Xbox enables running
-     * then polls to make sure its running. But we complete DMA instantaneously,
-     * so when is it supposed to be signaled that it stopped? Maybe just wait at
-     * least one cycle? How long does hardware wait?
-     */
-}
-
-void dsp_bootstrap(DSPState* dsp)
-{
-    // scratch memory is dma'd in to pram by the bootrom
-    dsp->dma.scratch_rw(dsp->dma.rw_opaque,
-        (uint8_t*)dsp->core.pram, 0, 0x800*4, false);
-    for (int i = 0; i < 0x800; i++) {
-        if (dsp->core.pram[i] & 0xff000000) {
-            DPRINTF(stderr, "Bootstrap %04x: %08x\n", i, dsp->core.pram[i]);
-            dsp->core.pram[i] &= 0x00ffffff;
-        }
-    }
-    memset(dsp->core.pram_opcache, 0, sizeof(dsp->core.pram_opcache));
-}
-
-void dsp_start_frame(DSPState* dsp)
-{
-    dsp->interrupts |= INTERRUPT_START_FRAME;
-}
-
-/**
- * Disassemble DSP code between given addresses, return next PC address
- */
-uint32_t dsp_disasm_address(DSPState* dsp, FILE *out, uint32_t lowerAdr, uint32_t UpperAdr)
-{
-    uint32_t dsp_pc;
-
-    for (dsp_pc=lowerAdr; dsp_pc<=UpperAdr; dsp_pc++) {
-        dsp_pc += dsp56k_execute_one_disasm_instruction(&dsp->core, out, dsp_pc);
-    }
-    return dsp_pc;
-}
-
-uint32_t dsp_read_memory(DSPState* dsp, char space, uint32_t address)
-{
-    int space_id;
-
-    switch (space) {
-    case 'X':
-        space_id = DSP_SPACE_X;
-        break;
-    case 'Y':
-        space_id = DSP_SPACE_Y;
-        break;
-    case 'P':
-        space_id = DSP_SPACE_P;
-        break;
-    default:
-        assert(false);
-        return 0;
-    }
-
-    return dsp56k_read_memory(&dsp->core, space_id, address);
-}
-
-void dsp_write_memory(DSPState* dsp, char space, uint32_t address, uint32_t value)
-{
-    int space_id;
-
-    switch (space) {
-    case 'X':
-        space_id = DSP_SPACE_X;
-        break;
-    case 'Y':
-        space_id = DSP_SPACE_Y;
-        break;
-    case 'P':
-        space_id = DSP_SPACE_P;
-        break;
-    default:
-        assert(false);
-        return;
-    }
-
-    dsp56k_write_memory(&dsp->core, space_id, address, value);
-}
-
 /**
  * Output memory values between given addresses in given DSP address space.
  * Return next DSP address value.
@@ -525,3 +295,5 @@ bool dsp_disasm_set_register(DSPState* dsp, const char *arg, uint32_t value)
     }
     return false;
 }
+
+#endif

hw/xbox/mcpx/apu/dsp/debug.h

@@ -0,0 +1,37 @@
+/*
+ * MCPX DSP emulator
+
+ * Copyright (c) 2025 Matt Borgerson
+
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#ifndef HW_XBOX_MCPX_DSP_DEBUG_H
+#define HW_XBOX_MCPX_DSP_DEBUG_H
+
+#ifndef DEBUG_DSP
+#define DEBUG_DSP 0
+#endif
+
+#define TRACE_DSP_DISASM 0
+#define TRACE_DSP_DISASM_REG 0
+#define TRACE_DSP_DISASM_MEM 0
+
+#define DPRINTF(fmt, ...) \
+    do { \
+        if (DEBUG_DSP) fprintf(stderr, fmt, ## __VA_ARGS__); \
+    } while (0)
+
+#endif

hw/xbox/mcpx/apu/dsp/dsp.c

@@ -0,0 +1,242 @@
+/*
+ * MCPX DSP emulator
+ *
+ * Copyright (c) 2015 espes
+ * Copyright (c) 2020-2025 Matt Borgerson
+ *
+ * Adapted from Hatari DSP M56001 emulation
+ * (C) 2001-2008 ARAnyM developer team
+ * Adaption to Hatari (C) 2008 by Thomas Huth
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#include "qemu/osdep.h"
+#include "dsp_cpu.h"
+#include "dsp_dma.h"
+#include "dsp_state.h"
+#include "dsp.h"
+#include "debug.h"
+#include "trace.h"
+
+/* Defines */
+#define BITMASK(x)  ((1<<(x))-1)
+
+#define INTERRUPT_ABORT_FRAME (1 << 0)
+#define INTERRUPT_START_FRAME (1 << 1)
+#define INTERRUPT_DMA_EOL (1 << 7)
+
+static uint32_t read_peripheral(dsp_core_t* core, uint32_t address);
+static void write_peripheral(dsp_core_t* core, uint32_t address, uint32_t value);
+
+DSPState *dsp_init(void *rw_opaque,
+                   dsp_scratch_rw_func scratch_rw,
+                   dsp_fifo_rw_func fifo_rw)
+{
+    DPRINTF("dsp_init\n");
+
+    DSPState* dsp = (DSPState*)malloc(sizeof(DSPState));
+    memset(dsp, 0, sizeof(*dsp));
+
+    dsp->core.read_peripheral = read_peripheral;
+    dsp->core.write_peripheral = write_peripheral;
+
+    dsp->dma.core = &dsp->core;
+    dsp->dma.rw_opaque = rw_opaque;
+    dsp->dma.scratch_rw = scratch_rw;
+    dsp->dma.fifo_rw = fifo_rw;
+
+    dsp_reset(dsp);
+
+    return dsp;
+}
+
+void dsp_reset(DSPState* dsp)
+{
+    dsp56k_reset_cpu(&dsp->core);
+    dsp->save_cycles = 0;
+}
+
+void dsp_destroy(DSPState* dsp)
+{
+    free(dsp);
+}
+
+static uint32_t read_peripheral(dsp_core_t* core, uint32_t address) {
+    DSPState* dsp = container_of(core, DSPState, core);
+
+    uint32_t v = 0xababa;
+    switch(address) {
+    case 0xFFFFB3:
+        v = 0; // core->num_inst; // ??
+        break;
+    case 0xFFFFC5:
+        v = dsp->interrupts;
+        if (dsp->dma.eol) {
+            v |= INTERRUPT_DMA_EOL;
+        }
+        break;
+    case 0xFFFFD4:
+        v = dsp_dma_read(&dsp->dma, DMA_NEXT_BLOCK);
+        break;
+    case 0xFFFFD5:
+        v = dsp_dma_read(&dsp->dma, DMA_START_BLOCK);
+        break;
+    case 0xFFFFD6:
+        v = dsp_dma_read(&dsp->dma, DMA_CONTROL);
+        break;
+    case 0xFFFFD7:
+        v = dsp_dma_read(&dsp->dma, DMA_CONFIGURATION);
+        break;
+    }
+
+    trace_dsp_read_peripheral(address, v);
+    return v;
+}
+
+static void write_peripheral(dsp_core_t* core, uint32_t address, uint32_t value) {
+    DSPState* dsp = container_of(core, DSPState, core);
+
+    switch(address) {
+    case 0xFFFFC4:
+        if (value & 1) {
+            core->is_idle = true;
+        }
+        break;
+    case 0xFFFFC5:
+        dsp->interrupts &= ~value;
+        if (value & INTERRUPT_DMA_EOL) {
+            dsp->dma.eol = false;
+        }
+        break;
+    case 0xFFFFD4:
+        dsp_dma_write(&dsp->dma, DMA_NEXT_BLOCK, value);
+        break;
+    case 0xFFFFD5:
+        dsp_dma_write(&dsp->dma, DMA_START_BLOCK, value);
+        break;
+    case 0xFFFFD6:
+        dsp_dma_write(&dsp->dma, DMA_CONTROL, value);
+        break;
+    case 0xFFFFD7:
+        dsp_dma_write(&dsp->dma, DMA_CONFIGURATION, value);
+        break;
+    }
+
+    trace_dsp_write_peripheral(address, value);
+}
+
+
+void dsp_step(DSPState* dsp)
+{
+    dsp56k_execute_instruction(&dsp->core);
+}
+
+void dsp_run(DSPState* dsp, int cycles)
+{
+    dsp->save_cycles += cycles;
+
+    if (dsp->save_cycles <= 0) return;
+
+    int dma_timer = 0;
+
+    while (dsp->save_cycles > 0)
+    {
+        dsp56k_execute_instruction(&dsp->core);
+        dsp->save_cycles -= dsp->core.instr_cycle;
+        dsp->core.cycle_count++;
+
+        if (dsp->dma.control & DMA_CONTROL_RUNNING) {
+            dma_timer++;
+        }
+
+        if (dma_timer > 2) {
+            dma_timer = 0;
+            dsp->dma.control &= ~DMA_CONTROL_RUNNING;
+            dsp->dma.control |= DMA_CONTROL_STOPPED;
+        }
+
+        if (dsp->core.is_idle) break;
+    }
+
+    /* FIXME: DMA timing be done cleaner. Xbox enables running
+     * then polls to make sure its running. But we complete DMA instantaneously,
+     * so when is it supposed to be signaled that it stopped? Maybe just wait at
+     * least one cycle? How long does hardware wait?
+     */
+}
+
+void dsp_bootstrap(DSPState* dsp)
+{
+    // scratch memory is dma'd in to pram by the bootrom
+    dsp->dma.scratch_rw(dsp->dma.rw_opaque,
+        (uint8_t*)dsp->core.pram, 0, 0x800*4, false);
+    for (int i = 0; i < 0x800; i++) {
+        if (dsp->core.pram[i] & 0xff000000) {
+            DPRINTF("Bootstrap %04x: %08x\n", i, dsp->core.pram[i]);
+            dsp->core.pram[i] &= 0x00ffffff;
+        }
+    }
+    memset(dsp->core.pram_opcache, 0, sizeof(dsp->core.pram_opcache));
+}
+
+void dsp_start_frame(DSPState* dsp)
+{
+    dsp->interrupts |= INTERRUPT_START_FRAME;
+}
+
+uint32_t dsp_read_memory(DSPState* dsp, char space, uint32_t address)
+{
+    int space_id;
+
+    switch (space) {
+    case 'X':
+        space_id = DSP_SPACE_X;
+        break;
+    case 'Y':
+        space_id = DSP_SPACE_Y;
+        break;
+    case 'P':
+        space_id = DSP_SPACE_P;
+        break;
+    default:
+        assert(false);
+        return 0;
+    }
+
+    return dsp56k_read_memory(&dsp->core, space_id, address);
+}
+
+void dsp_write_memory(DSPState* dsp, char space, uint32_t address, uint32_t value)
+{
+    int space_id;
+
+    switch (space) {
+    case 'X':
+        space_id = DSP_SPACE_X;
+        break;
+    case 'Y':
+        space_id = DSP_SPACE_Y;
+        break;
+    case 'P':
+        space_id = DSP_SPACE_P;
+        break;
+    default:
+        assert(false);
+        return;
+    }
+
+    dsp56k_write_memory(&dsp->core, space_id, address, value);
+}

hw/xbox/mcpx/apu/dsp/dsp.h

@@ -36,7 +36,6 @@ typedef void (*dsp_scratch_rw_func)(
 typedef void (*dsp_fifo_rw_func)(
     void *opaque, uint8_t *ptr, unsigned int index, size_t len, bool dir);
 
-/* Dsp commands */
 DSPState *dsp_init(void *rw_opaque,
                    dsp_scratch_rw_func scratch_rw,
                    dsp_fifo_rw_func fifo_rw);
@@ -49,16 +48,14 @@ void dsp_run(DSPState* dsp, int cycles);
 void dsp_bootstrap(DSPState* dsp);
 void dsp_start_frame(DSPState* dsp);
 
-
-/* Dsp Debugger commands */
 uint32_t dsp_read_memory(DSPState* dsp, char space, uint32_t addr);
 void dsp_write_memory(DSPState* dsp, char space, uint32_t address, uint32_t value);
-uint32_t dsp_disasm_memory(DSPState* dsp, uint32_t dsp_memdump_addr, uint32_t dsp_memdump_upper, char space);
-uint32_t dsp_disasm_address(DSPState* dsp, FILE *out, uint32_t lowerAdr, uint32_t UpperAdr);
+
 void dsp_info(DSPState* dsp);
 void dsp_print_registers(DSPState* dsp);
 int dsp_get_register_address(DSPState* dsp, const char *arg, uint32_t **addr, uint32_t *mask);
+uint32_t dsp_disasm_memory(DSPState* dsp, uint32_t dsp_memdump_addr, uint32_t dsp_memdump_upper, char space);
+uint32_t dsp_disasm_address(DSPState* dsp, FILE *out, uint32_t lowerAdr, uint32_t UpperAdr);
 bool dsp_disasm_set_register(DSPState* dsp, const char *arg, uint32_t value);
 
-
 #endif /* DSP_H */

hw/xbox/mcpx/apu/dsp/dsp_cpu.c

@@ -2,6 +2,7 @@
  * DSP56300 emulator
  *
  * Copyright (c) 2015 espes
+ * Copyright (c) 2020-2025 Matt Borgerson
  *
  * Adapted from Hatari DSP M56001 emulation
  * (C) 2003-2008 ARAnyM developer team
@@ -23,20 +24,10 @@
  */
 
 #include "qemu/osdep.h"
-
-#include <stdbool.h>
-#include <string.h>
-#include <assert.h>
-
 #include "qemu/bswap.h"
-
 #include "dsp_cpu.h"
-
-#define TRACE_DSP_DISASM 0
-#define TRACE_DSP_DISASM_REG 0
-#define TRACE_DSP_DISASM_MEM 0
-
-#define DPRINTF(s, ...) printf(s, ## __VA_ARGS__)
+#include "debug.h"
+#include "trace.h"
 
 #define BITMASK(x)  ((1<<(x))-1)
 
@@ -80,18 +71,18 @@ static void dsp_rnd56(dsp_core_t* dsp, uint32_t *dest);
 static uint32_t dsp_signextend(int bits, uint32_t v);
 
 static const dsp_interrupt_t dsp_interrupt[12] = {
-    {DSP_INTER_RESET    ,   0x00, 0, "Reset"},
-    {DSP_INTER_ILLEGAL  ,   0x3e, 0, "Illegal"},
-    {DSP_INTER_STACK_ERROR  ,   0x02, 0, "Stack Error"},
-    {DSP_INTER_TRACE    ,   0x04, 0, "Trace"},
-    {DSP_INTER_SWI      ,   0x06, 0, "Swi"},
-    {DSP_INTER_HOST_COMMAND ,   0xff, 1, "Host Command"},
-    {DSP_INTER_HOST_RCV_DATA,   0x20, 1, "Host receive"},
-    {DSP_INTER_HOST_TRX_DATA,   0x22, 1, "Host transmit"},
-    {DSP_INTER_SSI_RCV_DATA_E,  0x0e, 2, "SSI receive with exception"},
-    {DSP_INTER_SSI_RCV_DATA ,   0x0c, 2, "SSI receive"},
-    {DSP_INTER_SSI_TRX_DATA_E,  0x12, 2, "SSI transmit with exception"},
-    {DSP_INTER_SSI_TRX_DATA ,   0x10, 2, "SSI tramsmit"}
+    { DSP_INTER_RESET, 0x00, 0, "Reset" },
+    { DSP_INTER_ILLEGAL, 0x3e, 0, "Illegal" },
+    { DSP_INTER_STACK_ERROR, 0x02, 0, "Stack Error" },
+    { DSP_INTER_TRACE, 0x04, 0, "Trace" },
+    { DSP_INTER_SWI, 0x06, 0, "Swi" },
+    { DSP_INTER_HOST_COMMAND, 0xff, 1, "Host Command" },
+    { DSP_INTER_HOST_RCV_DATA, 0x20, 1, "Host receive" },
+    { DSP_INTER_HOST_TRX_DATA, 0x22, 1, "Host transmit" },
+    { DSP_INTER_SSI_RCV_DATA_E, 0x0e, 2, "SSI receive with exception" },
+    { DSP_INTER_SSI_RCV_DATA, 0x0c, 2, "SSI receive" },
+    { DSP_INTER_SSI_TRX_DATA_E, 0x12, 2, "SSI transmit with exception" },
+    { DSP_INTER_SSI_TRX_DATA, 0x10, 2, "SSI transmit" }
 };
 
 static const int registers_tcc[16][2] = {
@@ -138,9 +129,8 @@ static const int registers_mask[64] = {
     16, 16, 16, 16
 };
 
-#include "dsp_emu.inl"
-
-#include "dsp_dis.inl"
+#include "dsp_emu.c.inc"
+#include "dsp_dis.c.inc"
 
 typedef bool (*match_func_t)(uint32_t op);
 
@@ -415,7 +405,6 @@ void dsp56k_reset_cpu(dsp_core_t* dsp)
 
     /* runtime shit */
 
-    dsp->executing_for_disasm = false;
     // start_time = SDL_GetTicks();
     dsp->num_inst = 0;
 
@@ -462,7 +451,7 @@ static uint16_t disasm_instruction(dsp_core_t* dsp, dsp_trace_disasm_t mode)
     if (mode == DSP_TRACE_MODE) {
         if (dsp->disasm_prev_inst_pc == dsp->pc) {
             if (!dsp->disasm_is_looping) {
-                printf( "Looping on DSP instruction at PC = $%04x\n", dsp->disasm_prev_inst_pc);
+                DPRINTF("Looping on DSP instruction at PC = $%04x\n", dsp->disasm_prev_inst_pc);
                 dsp->disasm_is_looping = true;
             }
             return 0;
@@ -520,7 +509,7 @@ static void disasm_reg_compare(dsp_core_t* dsp)
             case DSP_REG_X1:
             case DSP_REG_Y0:
             case DSP_REG_Y1:
-                printf("\tReg: %s  $%06x -> $%06x\n",
+                DPRINTF("\tReg: %s  $%06x -> $%06x\n",
                     registers_name[i], dsp->disasm_registers_save[i], dsp->registers[i]);
                 break;
             case DSP_REG_R0:
@@ -550,21 +539,21 @@ static void disasm_reg_compare(dsp_core_t* dsp)
             case DSP_REG_SR:
             case DSP_REG_LA:
             case DSP_REG_LC:
-                printf("\tReg: %s  $%04x -> $%04x\n",
+                DPRINTF("\tReg: %s  $%04x -> $%04x\n",
                     registers_name[i], dsp->disasm_registers_save[i], dsp->registers[i]);
                 break;
             case DSP_REG_OMR:
             case DSP_REG_SP:
             case DSP_REG_SSH:
             case DSP_REG_SSL:
-                printf("\tReg: %s  $%02x -> $%02x\n",
+                DPRINTF("\tReg: %s  $%02x -> $%02x\n",
                     registers_name[i], dsp->disasm_registers_save[i], dsp->registers[i]);
                 break;
             case DSP_REG_A0:
             case DSP_REG_A1:
             case DSP_REG_A2:
                 if (bRegA == false) {
-                    printf("\tReg: a   $%02x:%06x:%06x -> $%02x:%06x:%06x\n",
+                    DPRINTF("\tReg: a   $%02x:%06x:%06x -> $%02x:%06x:%06x\n",
                         dsp->disasm_registers_save[DSP_REG_A2], dsp->disasm_registers_save[DSP_REG_A1], dsp->disasm_registers_save[DSP_REG_A0],
                         dsp->registers[DSP_REG_A2], dsp->registers[DSP_REG_A1], dsp->registers[DSP_REG_A0]
                     );
@@ -575,7 +564,7 @@ static void disasm_reg_compare(dsp_core_t* dsp)
             case DSP_REG_B1:
             case DSP_REG_B2:
                 if (bRegB == false) {
-                    printf("\tReg: b   $%02x:%06x:%06x -> $%02x:%06x:%06x\n",
+                    DPRINTF("\tReg: b   $%02x:%06x:%06x -> $%02x:%06x:%06x\n",
                         dsp->disasm_registers_save[DSP_REG_B2], dsp->disasm_registers_save[DSP_REG_B1], dsp->disasm_registers_save[DSP_REG_B0],
                         dsp->registers[DSP_REG_B2], dsp->registers[DSP_REG_B1], dsp->registers[DSP_REG_B0]
                     );
@@ -587,70 +576,28 @@ static void disasm_reg_compare(dsp_core_t* dsp)
 
 #ifdef DSP_DISASM_REG_PC
     if (pc_save != dsp->pc) {
-        printf("\tReg: pc  $%04x -> $%04x\n", pc_save, dsp->pc);
+        DPRINTF("\tReg: pc  $%04x -> $%04x\n", pc_save, dsp->pc);
     }
 #endif
 }
 
 static const char* disasm_get_instruction_text(dsp_core_t* dsp)
 {
-    // const int len = sizeof(dsp->disasm_str_instr);
-    // uint64_t count, cycles;
-    // uint16_t cycle_diff;
-    // float percentage;
-    // int offset;
-
     if (dsp->disasm_is_looping) {
         dsp->disasm_str_instr2[0] = 0;
     }
-    // if (dsp->disasm_cur_inst_len == 1) {
-    //     offset = sprintf(dsp->disasm_str_instr2, "p:%04x  %06x         (%02d cyc)  %-*s\n", dsp->disasm_prev_inst_pc, dsp->disasm_cur_inst, dsp->instr_cycle, len, dsp->disasm_str_instr);
-    // } else {
-    //     offset = sprintf(dsp->disasm_str_instr2, "p:%04x  %06x %06x  (%02d cyc)  %-*s\n", dsp->disasm_prev_inst_pc, dsp->disasm_cur_inst, read_memory_p(dsp, dsp->disasm_prev_inst_pc + 1), dsp->instr_cycle, len, dsp->disasm_str_instr);
-    // }
-    // if (offset > 2 && Profile_DspAddressData(dsp->disasm_prev_inst_pc, &percentage, &count, &cycles, &cycle_diff)) {
-    //     offset -= 2;
-    //     sprintf(str_instr2+offset, "%5.2f%% (%"PRId64", %"PRId64", %d)\n",
-    //             percentage, count, cycles, cycle_diff);
-    // }
+    if (dsp->disasm_cur_inst_len == 1) {
+        snprintf(dsp->disasm_str_instr2, sizeof(dsp->disasm_str_instr2), "p:%04x  %06x         (%02d cyc)  %s", dsp->disasm_prev_inst_pc, dsp->disasm_cur_inst, dsp->instr_cycle, dsp->disasm_str_instr);
+    } else {
+        snprintf(dsp->disasm_str_instr2, sizeof(dsp->disasm_str_instr2), "p:%04x  %06x %06x  (%02d cyc)  %s", dsp->disasm_prev_inst_pc, dsp->disasm_cur_inst, read_memory_p(dsp, dsp->disasm_prev_inst_pc + 1), dsp->instr_cycle, dsp->disasm_str_instr);
+    }
     return dsp->disasm_str_instr2;
 }
 
-/**
- * Execute one instruction in trace mode at a given PC address.
- * */
-uint16_t dsp56k_execute_one_disasm_instruction(dsp_core_t* dsp, FILE *out, uint32_t pc)
-{
-    dsp_core_t dsp_core_save;
-
-    /* Set DSP in disasm mode */
-    dsp->executing_for_disasm = true;
-
-    /* Save DSP context before executing instruction */
-    memcpy(&dsp_core_save, dsp, sizeof(dsp_core_t));
-
-    /* execute and disasm instruction */
-    dsp->pc = pc;
-
-    /* Disasm instruction */
-    uint16_t instruction_length = disasm_instruction(dsp, DSP_DISASM_MODE) - 1;
-
-    /* Execute instruction at address given in parameter to get the number of cycles it takes */
-    dsp56k_execute_instruction(dsp);
-
-    fprintf(out, "%s", disasm_get_instruction_text(dsp));
-
-    /* Restore DSP context after executing instruction */
-    memcpy(dsp, &dsp_core_save, sizeof(dsp_core_t));
-
-    /* Unset DSP in disasm mode */
-    dsp->executing_for_disasm = false;
-
-    return instruction_length;
-}
-
 void dsp56k_execute_instruction(dsp_core_t* dsp)
 {
+    trace_dsp56k_execute_instruction(dsp->is_gp, dsp->pc);
+
     uint32_t disasm_return = 0;
     dsp->disasm_memory_ptr = 0;
 
@@ -661,17 +608,18 @@ void dsp56k_execute_instruction(dsp_core_t* dsp)
     dsp->cur_inst_len = 1;
     dsp->instr_cycle = 2;
 
-    /* Disasm current instruction ? (trace mode only) */
-    if (TRACE_DSP_DISASM) {
-        /* Call disasm_instruction only when DSP is called in trace mode */
-        if (!dsp->executing_for_disasm) {
-            disasm_return = disasm_instruction(dsp, DSP_TRACE_MODE);
+    bool tracing = TRACE_DSP_DISASM || trace_event_get_state(TRACE_DSP56K_EXECUTE_INSTRUCTION_DISASM);
 
-            if (disasm_return) {
-                printf( "%s", disasm_get_instruction_text(dsp));
+    /* Disasm current instruction ? (trace mode only) */
+    if (tracing) {
+        disasm_return = disasm_instruction(dsp, DSP_TRACE_MODE);
+        if (disasm_return) {
+            const char *text = disasm_get_instruction_text(dsp);
+            trace_dsp56k_execute_instruction_disasm(text);
+            if (TRACE_DSP_DISASM) {
+                DPRINTF("%s\n", text);
             }
-            if (disasm_return != 0 && TRACE_DSP_DISASM_REG) {
-                /* DSP regs trace enabled only if DSP DISASM is enabled */
+            if (TRACE_DSP_DISASM_REG) {
                 disasm_reg_save(dsp);
             }
         }
@@ -686,7 +634,7 @@ void dsp56k_execute_instruction(dsp_core_t* dsp)
         if (op->emu_func) {
             op->emu_func(dsp);
         } else {
-            printf("%x - %s\n", dsp->cur_inst, op->name);
+            DPRINTF("%x - %s\n", dsp->cur_inst, op->name);
             emu_undefined(dsp);
         }
     } else {
@@ -695,26 +643,18 @@ void dsp56k_execute_instruction(dsp_core_t* dsp)
     }
 
     /* Disasm current instruction ? (trace mode only) */
-    if (TRACE_DSP_DISASM) {
-        /* Display only when DSP is called in trace mode */
-        if (!dsp->executing_for_disasm) {
-            if (disasm_return != 0) {
-                // printf( "%s", disasm_get_instruction_text(dsp));
-
-                /* DSP regs trace enabled only if DSP DISASM is enabled */
-                if (TRACE_DSP_DISASM_REG)
-                    disasm_reg_compare(dsp);
-
-                if (TRACE_DSP_DISASM_MEM) {
-                    /* 1 memory change to display ? */
-                    if (dsp->disasm_memory_ptr == 1)
-                        printf( "\t%s\n", dsp->str_disasm_memory[0]);
-                    /* 2 memory changes to display ? */
-                    else if (dsp->disasm_memory_ptr == 2) {
-                        printf( "\t%s\n", dsp->str_disasm_memory[0]);
-                        printf( "\t%s\n", dsp->str_disasm_memory[1]);
-                    }
-                }
+    if (tracing && disasm_return) {
+        if (TRACE_DSP_DISASM_REG) {
+            disasm_reg_compare(dsp);
+        }
+        if (TRACE_DSP_DISASM_MEM) {
+            /* 1 memory change to display ? */
+            if (dsp->disasm_memory_ptr == 1)
+                DPRINTF("\t%s\n", dsp->str_disasm_memory[0]);
+            /* 2 memory changes to display ? */
+            else if (dsp->disasm_memory_ptr == 2) {
+                DPRINTF("\t%s\n", dsp->str_disasm_memory[0]);
+                DPRINTF("\t%s\n", dsp->str_disasm_memory[1]);
             }
         }
     }
@@ -734,7 +674,7 @@ void dsp56k_execute_instruction(dsp_core_t* dsp)
         /* Evaluate time after <N> instructions have been executed to avoid asking too frequently */
         uint32_t cur_time = SDL_GetTicks();
         if (cur_time-start_time>1000) {
-            printf( "Dsp: %d i/s\n", (dsp->num_inst*1000)/(cur_time-start_time));
+            DPRINTF("Dsp: %d i/s\n", (dsp->num_inst*1000)/(cur_time-start_time));
             start_time=cur_time;
             dsp->num_inst=0;
         }
@@ -1021,9 +961,6 @@ uint32_t dsp56k_read_memory(dsp_core_t* dsp, int space, uint32_t address)
 
 void dsp56k_write_memory(dsp_core_t* dsp, int space, uint32_t address, uint32_t value)
 {
-    assert((value & 0xFF000000) == 0);
-    assert((address & 0xFF000000) == 0);
-
     if (TRACE_DSP_DISASM_MEM)
         write_memory_disasm(dsp, space, address, value);
     else
@@ -1122,9 +1059,7 @@ static void dsp_write_reg(dsp_core_t* dsp, uint32_t numreg, uint32_t value)
                 /* Stack underflow or overflow detected, raise interrupt */
                 dsp56k_add_interrupt(dsp, DSP_INTER_STACK_ERROR);
                 dsp->registers[DSP_REG_SP] = value & (3<<DSP_SP_SE);
-                if (!dsp->executing_for_disasm) {
-                    printf( "Dsp: Stack Overflow or Underflow\n");
-                }
+                DPRINTF("Dsp: Stack Overflow or Underflow\n");
                 if (dsp->exception_debugging) {
                     assert(false);
                 }
@@ -1167,8 +1102,7 @@ static void dsp_stack_push(dsp_core_t* dsp, uint32_t curpc, uint32_t cursr, uint
     if ((stack_error==0) && (stack & (1<<DSP_SP_SE))) {
         /* Stack full, raise interrupt */
         dsp56k_add_interrupt(dsp, DSP_INTER_STACK_ERROR);
-        if (!dsp->executing_for_disasm)
-            printf("Dsp: Stack Overflow\n");
+        DPRINTF("Dsp: Stack Overflow\n");
         if (dsp->exception_debugging)
             assert(false);
     }
@@ -1204,8 +1138,7 @@ static void dsp_stack_pop(dsp_core_t* dsp, uint32_t *newpc, uint32_t *newsr)
     if ((stack_error==0) && (stack & (1<<DSP_SP_SE))) {
         /* Stack empty*/
         dsp56k_add_interrupt(dsp, DSP_INTER_STACK_ERROR);
-        if (!dsp->executing_for_disasm)
-            printf("Dsp: Stack underflow\n");
+        DPRINTF("Dsp: Stack underflow\n");
         if (dsp->exception_debugging)
             assert(false);
     }
@@ -1458,5 +1391,3 @@ static uint32_t dsp_signextend(int bits, uint32_t v) {
     assert(shift > 0);
     return (uint32_t)(((int32_t)v << shift) >> shift);
 }
-
-

hw/xbox/mcpx/apu/dsp/dsp_cpu.h

@@ -29,115 +29,7 @@
 #include <stdint.h>
 #include <stdbool.h>
 
-#define DSP_OMR_MA  0x00
-#define DSP_OMR_MB  0x01
-#define DSP_OMR_DE  0x02
-#define DSP_OMR_SD  0x06
-#define DSP_OMR_EA  0x07
-
-#define DSP_SR_C    0x00
-#define DSP_SR_V    0x01
-#define DSP_SR_Z    0x02
-#define DSP_SR_N    0x03
-#define DSP_SR_U    0x04
-#define DSP_SR_E    0x05
-#define DSP_SR_L    0x06
-
-#define DSP_SR_I0   0x08
-#define DSP_SR_I1   0x09
-#define DSP_SR_S0   0x0a
-#define DSP_SR_S1   0x0b
-#define DSP_SR_T    0x0d
-#define DSP_SR_LF   0x0f
-
-#define DSP_SP_SE   0x04
-#define DSP_SP_UF   0x05
-
-/* Registers numbers in dsp.registers[] */
-#define DSP_REG_X0  0x04
-#define DSP_REG_X1  0x05
-#define DSP_REG_Y0  0x06
-#define DSP_REG_Y1  0x07
-#define DSP_REG_A0  0x08
-#define DSP_REG_B0  0x09
-#define DSP_REG_A2  0x0a
-#define DSP_REG_B2  0x0b
-#define DSP_REG_A1  0x0c
-#define DSP_REG_B1  0x0d
-#define DSP_REG_A   0x0e
-#define DSP_REG_B   0x0f
-
-#define DSP_REG_R0  0x10
-#define DSP_REG_R1  0x11
-#define DSP_REG_R2  0x12
-#define DSP_REG_R3  0x13
-#define DSP_REG_R4  0x14
-#define DSP_REG_R5  0x15
-#define DSP_REG_R6  0x16
-#define DSP_REG_R7  0x17
-
-#define DSP_REG_N0  0x18
-#define DSP_REG_N1  0x19
-#define DSP_REG_N2  0x1a
-#define DSP_REG_N3  0x1b
-#define DSP_REG_N4  0x1c
-#define DSP_REG_N5  0x1d
-#define DSP_REG_N6  0x1e
-#define DSP_REG_N7  0x1f
-
-#define DSP_REG_M0  0x20
-#define DSP_REG_M1  0x21
-#define DSP_REG_M2  0x22
-#define DSP_REG_M3  0x23
-#define DSP_REG_M4  0x24
-#define DSP_REG_M5  0x25
-#define DSP_REG_M6  0x26
-#define DSP_REG_M7  0x27
-
-#define DSP_REG_SR  0x39
-#define DSP_REG_OMR 0x3a
-#define DSP_REG_SP  0x3b
-#define DSP_REG_SSH 0x3c
-#define DSP_REG_SSL 0x3d
-#define DSP_REG_LA  0x3e
-#define DSP_REG_LC  0x3f
-
-#define DSP_REG_NULL    0x00
-#define DSP_REG_LCSAVE  0x30
-
-#define DSP_REG_MAX 0x40
-
-/* Memory spaces for dsp.ram[], dsp.rom[] */
-#define DSP_SPACE_X 0x00
-#define DSP_SPACE_Y 0x01
-#define DSP_SPACE_P 0x02
-
-#define DSP_XRAM_SIZE 4096
-#define DSP_YRAM_SIZE 2048
-#define DSP_PRAM_SIZE 4096
-
-#define DSP_MIXBUFFER_BASE 0x001400
-#define DSP_MIXBUFFER_SIZE 1024
-
-#define DSP_PERIPH_BASE 0xFFFF80
-#define DSP_PERIPH_SIZE 128
-
-#define DSP_INTERRUPT_NONE      0x0
-#define DSP_INTERRUPT_DISABLED  0x1
-#define DSP_INTERRUPT_LONG      0x2
-
-#define DSP_INTER_RESET         0x0
-#define DSP_INTER_ILLEGAL       0x1
-#define DSP_INTER_STACK_ERROR       0x2
-#define DSP_INTER_TRACE         0x3
-#define DSP_INTER_SWI           0x4
-#define DSP_INTER_HOST_COMMAND      0x5
-#define DSP_INTER_HOST_RCV_DATA     0x6
-#define DSP_INTER_HOST_TRX_DATA     0x7
-#define DSP_INTER_SSI_RCV_DATA_E    0x8
-#define DSP_INTER_SSI_RCV_DATA      0x9
-#define DSP_INTER_SSI_TRX_DATA_E    0xa
-#define DSP_INTER_SSI_TRX_DATA      0xb
+#include "dsp_cpu_regs.h"
 
 typedef enum {
     DSP_TRACE_MODE,
@@ -209,10 +101,6 @@ struct dsp_core_s {
     /* Current instruction */
     uint32_t cur_inst;
 
-    /* DSP is in disasm mode ? */
-    /* If yes, stack overflow, underflow and illegal instructions messages are not displayed */
-    bool executing_for_disasm;
-
     char str_disasm_memory[2][50];     /* Buffer for memory change text in disasm mode */
     uint32_t disasm_memory_ptr;        /* Pointer for memory change in disasm mode */
 
@@ -250,7 +138,6 @@ struct dsp_core_s {
 /* Functions */
 void dsp56k_reset_cpu(dsp_core_t* dsp);		/* Set dsp_core to use */
 void dsp56k_execute_instruction(dsp_core_t* dsp);	/* Execute 1 instruction */
-uint16_t dsp56k_execute_one_disasm_instruction(dsp_core_t* dsp, FILE *out, uint32_t pc);	/* Execute 1 instruction in disasm mode */
 
 uint32_t dsp56k_read_memory(dsp_core_t* dsp, int space, uint32_t address);
 void dsp56k_write_memory(dsp_core_t* dsp, int space, uint32_t address, uint32_t value);

hw/xbox/mcpx/apu/dsp/dsp_cpu_regs.h

@@ -0,0 +1,138 @@
+/*
+ * DSP56300 emulator
+ *
+ * Copyright (c) 2015 espes
+ *
+ * Adapted from Hatari DSP M56001 emulation
+ * (C) 2003-2008 ARAnyM developer team
+ * Adaption to Hatari (C) 2008 by Thomas Huth
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#ifndef HW_XBOX_MCPX_DSP_DSP_CPU_REGS_H
+#define HW_XBOX_MCPX_DSP_DSP_CPU_REGS_H
+
+#define DSP_OMR_MA  0x00
+#define DSP_OMR_MB  0x01
+#define DSP_OMR_DE  0x02
+#define DSP_OMR_SD  0x06
+#define DSP_OMR_EA  0x07
+
+#define DSP_SR_C    0x00
+#define DSP_SR_V    0x01
+#define DSP_SR_Z    0x02
+#define DSP_SR_N    0x03
+#define DSP_SR_U    0x04
+#define DSP_SR_E    0x05
+#define DSP_SR_L    0x06
+
+#define DSP_SR_I0   0x08
+#define DSP_SR_I1   0x09
+#define DSP_SR_S0   0x0a
+#define DSP_SR_S1   0x0b
+#define DSP_SR_T    0x0d
+#define DSP_SR_LF   0x0f
+
+#define DSP_SP_SE   0x04
+#define DSP_SP_UF   0x05
+
+/* Registers numbers in dsp.registers[] */
+#define DSP_REG_X0  0x04
+#define DSP_REG_X1  0x05
+#define DSP_REG_Y0  0x06
+#define DSP_REG_Y1  0x07
+#define DSP_REG_A0  0x08
+#define DSP_REG_B0  0x09
+#define DSP_REG_A2  0x0a
+#define DSP_REG_B2  0x0b
+#define DSP_REG_A1  0x0c
+#define DSP_REG_B1  0x0d
+#define DSP_REG_A   0x0e
+#define DSP_REG_B   0x0f
+
+#define DSP_REG_R0  0x10
+#define DSP_REG_R1  0x11
+#define DSP_REG_R2  0x12
+#define DSP_REG_R3  0x13
+#define DSP_REG_R4  0x14
+#define DSP_REG_R5  0x15
+#define DSP_REG_R6  0x16
+#define DSP_REG_R7  0x17
+
+#define DSP_REG_N0  0x18
+#define DSP_REG_N1  0x19
+#define DSP_REG_N2  0x1a
+#define DSP_REG_N3  0x1b
+#define DSP_REG_N4  0x1c
+#define DSP_REG_N5  0x1d
+#define DSP_REG_N6  0x1e
+#define DSP_REG_N7  0x1f
+
+#define DSP_REG_M0  0x20
+#define DSP_REG_M1  0x21
+#define DSP_REG_M2  0x22
+#define DSP_REG_M3  0x23
+#define DSP_REG_M4  0x24
+#define DSP_REG_M5  0x25
+#define DSP_REG_M6  0x26
+#define DSP_REG_M7  0x27
+
+#define DSP_REG_SR  0x39
+#define DSP_REG_OMR 0x3a
+#define DSP_REG_SP  0x3b
+#define DSP_REG_SSH 0x3c
+#define DSP_REG_SSL 0x3d
+#define DSP_REG_LA  0x3e
+#define DSP_REG_LC  0x3f
+
+#define DSP_REG_NULL    0x00
+#define DSP_REG_LCSAVE  0x30
+
+#define DSP_REG_MAX 0x40
+
+/* Memory spaces for dsp.ram[], dsp.rom[] */
+#define DSP_SPACE_X 0x00
+#define DSP_SPACE_Y 0x01
+#define DSP_SPACE_P 0x02
+
+#define DSP_XRAM_SIZE 4096
+#define DSP_YRAM_SIZE 2048
+#define DSP_PRAM_SIZE 4096
+
+#define DSP_MIXBUFFER_BASE 0x001400
+#define DSP_MIXBUFFER_SIZE 1024
+
+#define DSP_PERIPH_BASE 0xFFFF80
+#define DSP_PERIPH_SIZE 128
+
+#define DSP_INTERRUPT_NONE      0x0
+#define DSP_INTERRUPT_DISABLED  0x1
+#define DSP_INTERRUPT_LONG      0x2
+
+#define DSP_INTER_RESET         0x0
+#define DSP_INTER_ILLEGAL       0x1
+#define DSP_INTER_STACK_ERROR       0x2
+#define DSP_INTER_TRACE         0x3
+#define DSP_INTER_SWI           0x4
+#define DSP_INTER_HOST_COMMAND      0x5
+#define DSP_INTER_HOST_RCV_DATA     0x6
+#define DSP_INTER_HOST_TRX_DATA     0x7
+#define DSP_INTER_SSI_RCV_DATA_E    0x8
+#define DSP_INTER_SSI_RCV_DATA      0x9
+#define DSP_INTER_SSI_TRX_DATA_E    0xa
+#define DSP_INTER_SSI_TRX_DATA      0xb
+
+#endif

hw/xbox/mcpx/apu/dsp/dsp_dma.c

@@ -18,48 +18,14 @@
  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  */
 
-#include <stdlib.h>
-#include <stdint.h>
-#include <stdbool.h>
-#include <assert.h>
-
-#include <stddef.h>
+#include "qemu/osdep.h"
 #include "qemu/compiler.h"
 #include "dsp_dma.h"
+#include "dsp_dma_regs.h"
 #include "dsp_state.h"
 
-
-#define DMA_CONFIGURATION_AUTOSTART (1 << 0)
-#define DMA_CONFIGURATION_AUTOREADY (1 << 1)
-#define DMA_CONFIGURATION_IOC_CLEAR (1 << 2)
-#define DMA_CONFIGURATION_EOL_CLEAR (1 << 3)
-#define DMA_CONFIGURATION_ERR_CLEAR (1 << 4)
-
-#define DMA_CONTROL_ACTION 0x7
-#define DMA_CONTROL_ACTION_NOP 0
-#define DMA_CONTROL_ACTION_START 1
-#define DMA_CONTROL_ACTION_STOP 2
-#define DMA_CONTROL_ACTION_FREEZE 3
-#define DMA_CONTROL_ACTION_UNFREEZE 4
-#define DMA_CONTROL_ACTION_ABORT 5
-#define DMA_CONTROL_FROZEN (1 << 3)
-#define DMA_CONTROL_RUNNING (1 << 4)
-#define DMA_CONTROL_STOPPED (1 << 5)
-
-#define NODE_POINTER_VAL 0x3fff
-#define NODE_POINTER_EOL (1 << 14)
-
-#define NODE_CONTROL_DIRECTION (1 << 1)
-
-
-// #define DEBUG
 #ifdef DEBUG
-# define DPRINTF(s, ...) printf(s, ## __VA_ARGS__)
-#else
-# define DPRINTF(s, ...) do { } while (0)
-#endif
 
-#ifdef DEBUG
 const char *buffer_names[] = {
     "fifo0",            /* 0x0 */
     "fifo1",            /* 0x1 */
@@ -98,8 +64,6 @@ const char *space_names[] = {
 
 #endif
 
-#define MIN(a,b) (((a)<(b))?(a):(b))
-
 static void scratch_circular_copy(
     DSPDMAState *s,
     uint32_t     scratch_base,
@@ -142,8 +106,6 @@ static void dsp_dma_run(DSPDMAState *s)
         return;
     }
 
-    // DSPState *dsp = container_of(s, DSPState, dma);
-
     while (!(s->next_block & NODE_POINTER_EOL)) {
         uint32_t addr = s->next_block & NODE_POINTER_VAL;
         uint32_t block_addr = 0;

hw/xbox/mcpx/apu/dsp/dsp_dma_regs.h

@@ -0,0 +1,46 @@
+/*
+ * MCPX DSP DMA
+ *
+ * Copyright (c) 2015 espes
+ * Copyright (c) 2020-2025 Matt Borgerson
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef HW_XBOX_MCPX_DSP_DSP_DMA_REGS_H
+#define HW_XBOX_MCPX_DSP_DSP_DMA_REGS_H
+
+#define DMA_CONFIGURATION_AUTOSTART (1 << 0)
+#define DMA_CONFIGURATION_AUTOREADY (1 << 1)
+#define DMA_CONFIGURATION_IOC_CLEAR (1 << 2)
+#define DMA_CONFIGURATION_EOL_CLEAR (1 << 3)
+#define DMA_CONFIGURATION_ERR_CLEAR (1 << 4)
+
+#define DMA_CONTROL_ACTION 0x7
+#define DMA_CONTROL_ACTION_NOP 0
+#define DMA_CONTROL_ACTION_START 1
+#define DMA_CONTROL_ACTION_STOP 2
+#define DMA_CONTROL_ACTION_FREEZE 3
+#define DMA_CONTROL_ACTION_UNFREEZE 4
+#define DMA_CONTROL_ACTION_ABORT 5
+#define DMA_CONTROL_FROZEN (1 << 3)
+#define DMA_CONTROL_RUNNING (1 << 4)
+#define DMA_CONTROL_STOPPED (1 << 5)
+
+#define NODE_POINTER_VAL 0x3fff
+#define NODE_POINTER_EOL (1 << 14)
+
+#define NODE_CONTROL_DIRECTION (1 << 1)
+
+#endif

hw/xbox/mcpx/apu/dsp/dsp_emu.c.inc

@@ -27,15 +27,10 @@ typedef void (*emu_func_t)(dsp_core_t* dsp);
 
 static void emu_undefined(dsp_core_t* dsp)
 {
-    if (!dsp->executing_for_disasm) {
-        dsp->cur_inst_len = 0;
-        printf("Dsp: 0x%04x: 0x%06x Illegal instruction\n",dsp->pc, dsp->cur_inst);
-        /* Add some artificial CPU cycles to avoid being stuck in an infinite loop */
-        dsp->instr_cycle += 100;
-    } else {
-        dsp->cur_inst_len = 1;
-        dsp->instr_cycle = 0;
-    }
+    dsp->cur_inst_len = 0;
+    printf("Dsp: 0x%04x: 0x%06x Illegal instruction\n",dsp->pc, dsp->cur_inst);
+    /* Add some artificial CPU cycles to avoid being stuck in an infinite loop */
+    dsp->instr_cycle += 100;
     if (dsp->exception_debugging) {
         assert(false);
     }

hw/xbox/mcpx/apu/dsp/gp_ep.c

@@ -0,0 +1,526 @@
+/*
+ * QEMU MCPX Audio Processing Unit implementation
+ *
+ * Copyright (c) 2012 espes
+ * Copyright (c) 2018-2019 Jannik Vogel
+ * Copyright (c) 2019-2025 Matt Borgerson
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "hw/xbox/mcpx/apu/apu_int.h"
+
+static const int16_t ep_silence[256][2] = { 0 };
+
+void mcpx_apu_update_dsp_preference(MCPXAPUState *d)
+{
+    static int last_known_preference = -1;
+
+    if (last_known_preference == (int)g_config.audio.use_dsp) {
+        return;
+    }
+
+    if (g_config.audio.use_dsp) {
+        d->monitor.point = MCPX_APU_DEBUG_MON_GP_OR_EP;
+        d->gp.realtime = true;
+        d->ep.realtime = true;
+    } else {
+        d->monitor.point = MCPX_APU_DEBUG_MON_VP;
+        d->gp.realtime = false;
+        d->ep.realtime = false;
+    }
+
+    last_known_preference = g_config.audio.use_dsp;
+}
+
+static void scatter_gather_rw(MCPXAPUState *d, hwaddr sge_base,
+                              unsigned int max_sge, uint8_t *ptr, uint32_t addr,
+                              size_t len, bool dir)
+{
+    unsigned int page_entry = addr / TARGET_PAGE_SIZE;
+    unsigned int offset_in_page = addr % TARGET_PAGE_SIZE;
+    unsigned int bytes_to_copy = TARGET_PAGE_SIZE - offset_in_page;
+
+    while (len > 0) {
+        assert(page_entry <= max_sge);
+
+        uint32_t prd_address = ldl_le_phys(&address_space_memory,
+                                           sge_base + page_entry * 8 + 0);
+        // uint32_t prd_control = ldl_le_phys(&address_space_memory,
+        //                                     sge_base + page_entry * 8 + 4);
+
+        hwaddr paddr = prd_address + offset_in_page;
+
+        if (bytes_to_copy > len) {
+            bytes_to_copy = len;
+        }
+
+        assert(paddr + bytes_to_copy < memory_region_size(d->ram));
+
+        if (dir) {
+            memcpy(&d->ram_ptr[paddr], ptr, bytes_to_copy);
+            memory_region_set_dirty(d->ram, paddr, bytes_to_copy);
+        } else {
+            memcpy(ptr, &d->ram_ptr[paddr], bytes_to_copy);
+        }
+
+        ptr += bytes_to_copy;
+        len -= bytes_to_copy;
+
+        /* After the first iteration, we are page aligned */
+        page_entry += 1;
+        bytes_to_copy = TARGET_PAGE_SIZE;
+        offset_in_page = 0;
+    }
+}
+
+static void gp_scratch_rw(void *opaque, uint8_t *ptr, uint32_t addr, size_t len,
+                          bool dir)
+{
+    MCPXAPUState *d = opaque;
+    // fprintf(stderr, "GP %s scratch 0x%x bytes (0x%x words) at %x (0x%x words)\n", dir ? "writing to" : "reading from", len, len/4, addr, addr/4);
+    scatter_gather_rw(d, d->regs[NV_PAPU_GPSADDR], d->regs[NV_PAPU_GPSMAXSGE],
+                      ptr, addr, len, dir);
+}
+
+static void ep_scratch_rw(void *opaque, uint8_t *ptr, uint32_t addr, size_t len,
+                          bool dir)
+{
+    MCPXAPUState *d = opaque;
+    // fprintf(stderr, "EP %s scratch 0x%x bytes (0x%x words) at %x (0x%x words)\n", dir ? "writing to" : "reading from", len, len/4, addr, addr/4);
+    scatter_gather_rw(d, d->regs[NV_PAPU_EPSADDR], d->regs[NV_PAPU_EPSMAXSGE],
+                      ptr, addr, len, dir);
+}
+
+static uint32_t circular_scatter_gather_rw(MCPXAPUState *d, hwaddr sge_base,
+                                           unsigned int max_sge, uint8_t *ptr,
+                                           uint32_t base, uint32_t end,
+                                           uint32_t cur, size_t len, bool dir)
+{
+    while (len > 0) {
+        unsigned int bytes_to_copy = end - cur;
+
+        if (bytes_to_copy > len) {
+            bytes_to_copy = len;
+        }
+
+        DPRINTF("circular scatter gather %s in range 0x%x - 0x%x at 0x%x of "
+                "length 0x%x / 0x%lx bytes\n",
+                dir ? "write" : "read", base, end, cur, bytes_to_copy, len);
+
+        assert((cur >= base) && ((cur + bytes_to_copy) <= end));
+        scatter_gather_rw(d, sge_base, max_sge, ptr, cur, bytes_to_copy, dir);
+
+        ptr += bytes_to_copy;
+        len -= bytes_to_copy;
+
+        /* After the first iteration we might have to wrap */
+        cur += bytes_to_copy;
+        if (cur >= end) {
+            assert(cur == end);
+            cur = base;
+        }
+    }
+
+    return cur;
+}
+
+static void gp_fifo_rw(void *opaque, uint8_t *ptr, unsigned int index,
+                       size_t len, bool dir)
+{
+    MCPXAPUState *d = opaque;
+    uint32_t base;
+    uint32_t end;
+    hwaddr cur_reg;
+    if (dir) {
+        assert(index < GP_OUTPUT_FIFO_COUNT);
+        base = GET_MASK(d->regs[NV_PAPU_GPOFBASE0 + 0x10 * index],
+                        NV_PAPU_GPOFBASE0_VALUE);
+        end = GET_MASK(d->regs[NV_PAPU_GPOFEND0 + 0x10 * index],
+                       NV_PAPU_GPOFEND0_VALUE);
+        cur_reg = NV_PAPU_GPOFCUR0 + 0x10 * index;
+    } else {
+        assert(index < GP_INPUT_FIFO_COUNT);
+        base = GET_MASK(d->regs[NV_PAPU_GPIFBASE0 + 0x10 * index],
+                        NV_PAPU_GPOFBASE0_VALUE);
+        end = GET_MASK(d->regs[NV_PAPU_GPIFEND0 + 0x10 * index],
+                       NV_PAPU_GPOFEND0_VALUE);
+        cur_reg = NV_PAPU_GPIFCUR0 + 0x10 * index;
+    }
+
+    uint32_t cur = GET_MASK(d->regs[cur_reg], NV_PAPU_GPOFCUR0_VALUE);
+
+    // fprintf(stderr, "GP %s fifo #%d, base = %x, end = %x, cur = %x, len = %x\n",
+    //     dir ? "writing to" : "reading from", index,
+    //     base, end, cur, len);
+
+    /* DSP hangs if current >= end; but forces current >= base */
+    assert(cur < end);
+    if (cur < base) {
+        cur = base;
+    }
+
+    cur = circular_scatter_gather_rw(d,
+        d->regs[NV_PAPU_GPFADDR], d->regs[NV_PAPU_GPFMAXSGE],
+        ptr, base, end, cur, len, dir);
+
+    SET_MASK(d->regs[cur_reg], NV_PAPU_GPOFCUR0_VALUE, cur);
+}
+
+static bool ep_sink_samples(MCPXAPUState *d, uint8_t *ptr, size_t len)
+{
+    if (d->monitor.point == MCPX_APU_DEBUG_MON_AC97) {
+        return false;
+    } else if ((d->monitor.point == MCPX_APU_DEBUG_MON_EP) ||
+        (d->monitor.point == MCPX_APU_DEBUG_MON_GP_OR_EP)) {
+        assert(len == sizeof(d->monitor.frame_buf));
+        memcpy(d->monitor.frame_buf, ptr, len);
+    }
+
+    return true;
+}
+
+static void ep_fifo_rw(void *opaque, uint8_t *ptr, unsigned int index,
+                       size_t len, bool dir)
+{
+    MCPXAPUState *d = opaque;
+    uint32_t base;
+    uint32_t end;
+    hwaddr cur_reg;
+    if (dir) {
+        assert(index < EP_OUTPUT_FIFO_COUNT);
+        base = GET_MASK(d->regs[NV_PAPU_EPOFBASE0 + 0x10 * index],
+                        NV_PAPU_GPOFBASE0_VALUE);
+        end = GET_MASK(d->regs[NV_PAPU_EPOFEND0 + 0x10 * index],
+                       NV_PAPU_GPOFEND0_VALUE);
+        cur_reg = NV_PAPU_EPOFCUR0 + 0x10 * index;
+    } else {
+        assert(index < EP_INPUT_FIFO_COUNT);
+        base = GET_MASK(d->regs[NV_PAPU_EPIFBASE0 + 0x10 * index],
+                        NV_PAPU_GPOFBASE0_VALUE);
+        end = GET_MASK(d->regs[NV_PAPU_EPIFEND0 + 0x10 * index],
+                       NV_PAPU_GPOFEND0_VALUE);
+        cur_reg = NV_PAPU_EPIFCUR0 + 0x10 * index;
+    }
+
+    uint32_t cur = GET_MASK(d->regs[cur_reg], NV_PAPU_GPOFCUR0_VALUE);
+
+    // fprintf(stderr, "EP %s fifo #%d, base = %x, end = %x, cur = %x, len = %x\n",
+    //     dir ? "writing to" : "reading from", index,
+    //     base, end, cur, len);
+
+    if (dir && index == 0) {
+        bool did_sink = ep_sink_samples(d, ptr, len);
+        if (did_sink) {
+            /* Since we are sinking, push silence out */
+            assert(len <= sizeof(ep_silence));
+            ptr = (uint8_t*)ep_silence;
+        }
+    }
+
+    /* DSP hangs if current >= end; but forces current >= base */
+    if (cur >= end) {
+        cur = cur % (end - base);
+    }
+    if (cur < base) {
+        cur = base;
+    }
+
+    cur = circular_scatter_gather_rw(d,
+        d->regs[NV_PAPU_EPFADDR], d->regs[NV_PAPU_EPFMAXSGE],
+        ptr, base, end, cur, len, dir);
+
+    SET_MASK(d->regs[cur_reg], NV_PAPU_GPOFCUR0_VALUE, cur);
+}
+
+static void proc_rst_write(DSPState *dsp, uint32_t oldval, uint32_t val)
+{
+    if (!(val & NV_PAPU_GPRST_GPRST) || !(val & NV_PAPU_GPRST_GPDSPRST)) {
+        dsp_reset(dsp);
+    } else if (
+        (!(oldval & NV_PAPU_GPRST_GPRST) || !(oldval & NV_PAPU_GPRST_GPDSPRST))
+        && ((val & NV_PAPU_GPRST_GPRST) && (val & NV_PAPU_GPRST_GPDSPRST))) {
+        dsp_bootstrap(dsp);
+    }
+}
+
+/* Global Processor - programmable DSP */
+static uint64_t gp_read(void *opaque, hwaddr addr, unsigned int size)
+{
+    MCPXAPUState *d = opaque;
+
+    assert(size == 4);
+    assert(addr % 4 == 0);
+
+    uint64_t r = 0;
+    switch (addr) {
+    case NV_PAPU_GPXMEM ... NV_PAPU_GPXMEM + 0x1000 * 4 - 1: {
+        uint32_t xaddr = (addr - NV_PAPU_GPXMEM) / 4;
+        r = dsp_read_memory(d->gp.dsp, 'X', xaddr);
+        // fprintf(stderr, "read GP NV_PAPU_GPXMEM [%x] -> %x\n", xaddr, r);
+        break;
+    }
+    case NV_PAPU_GPMIXBUF ... NV_PAPU_GPMIXBUF + 0x400 * 4 - 1: {
+        uint32_t xaddr = (addr - NV_PAPU_GPMIXBUF) / 4;
+        r = dsp_read_memory(d->gp.dsp, 'X', GP_DSP_MIXBUF_BASE + xaddr);
+        // fprintf(stderr, "read GP NV_PAPU_GPMIXBUF [%x] -> %x\n", xaddr, r);
+        break;
+    }
+    case NV_PAPU_GPYMEM ... NV_PAPU_GPYMEM + 0x800 * 4 - 1: {
+        uint32_t yaddr = (addr - NV_PAPU_GPYMEM) / 4;
+        r = dsp_read_memory(d->gp.dsp, 'Y', yaddr);
+        // fprintf(stderr, "read GP NV_PAPU_GPYMEM [%x] -> %x\n", yaddr, r);
+        break;
+    }
+    case NV_PAPU_GPPMEM ... NV_PAPU_GPPMEM + 0x1000 * 4 - 1: {
+        uint32_t paddr = (addr - NV_PAPU_GPPMEM) / 4;
+        r = dsp_read_memory(d->gp.dsp, 'P', paddr);
+        // fprintf(stderr, "read GP NV_PAPU_GPPMEM [%x] -> %x\n", paddr, r);
+        break;
+    }
+    default:
+        r = d->gp.regs[addr];
+        break;
+    }
+    DPRINTF("mcpx apu GP: read [0x%" HWADDR_PRIx "] -> 0x%lx\n", addr, r);
+
+    return r;
+}
+
+static void gp_write(void *opaque, hwaddr addr, uint64_t val, unsigned int size)
+{
+    MCPXAPUState *d = opaque;
+
+    qemu_mutex_lock(&d->lock);
+
+    assert(size == 4);
+    assert(addr % 4 == 0);
+
+    DPRINTF("mcpx apu GP: [0x%" HWADDR_PRIx "] = 0x%lx\n", addr, val);
+
+    switch (addr) {
+    case NV_PAPU_GPXMEM ... NV_PAPU_GPXMEM + 0x1000 * 4 - 1: {
+        uint32_t xaddr = (addr - NV_PAPU_GPXMEM) / 4;
+        // fprintf(stderr, "gp write xmem %x = %x\n", xaddr, val);
+        dsp_write_memory(d->gp.dsp, 'X', xaddr, val);
+        break;
+    }
+    case NV_PAPU_GPMIXBUF ... NV_PAPU_GPMIXBUF + 0x400 * 4 - 1: {
+        uint32_t xaddr = (addr - NV_PAPU_GPMIXBUF) / 4;
+        // fprintf(stderr, "gp write xmixbuf %x = %x\n", xaddr, val);
+        dsp_write_memory(d->gp.dsp, 'X', GP_DSP_MIXBUF_BASE + xaddr, val);
+        break;
+    }
+    case NV_PAPU_GPYMEM ... NV_PAPU_GPYMEM + 0x800 * 4 - 1: {
+        uint32_t yaddr = (addr - NV_PAPU_GPYMEM) / 4;
+        // fprintf(stderr, "gp write ymem %x = %x\n", yaddr, val);
+        dsp_write_memory(d->gp.dsp, 'Y', yaddr, val);
+        break;
+    }
+    case NV_PAPU_GPPMEM ... NV_PAPU_GPPMEM + 0x1000 * 4 - 1: {
+        uint32_t paddr = (addr - NV_PAPU_GPPMEM) / 4;
+        // fprintf(stderr, "gp write pmem %x = %x\n", paddr, val);
+        dsp_write_memory(d->gp.dsp, 'P', paddr, val);
+        break;
+    }
+    case NV_PAPU_GPRST:
+        proc_rst_write(d->gp.dsp, d->gp.regs[NV_PAPU_GPRST], val);
+        d->gp.regs[NV_PAPU_GPRST] = val;
+        break;
+    default:
+        d->gp.regs[addr] = val;
+        break;
+    }
+
+    qemu_mutex_unlock(&d->lock);
+}
+
+const MemoryRegionOps gp_ops = {
+    .read = gp_read,
+    .write = gp_write,
+};
+
+/* Encode Processor - encoding DSP */
+static uint64_t ep_read(void *opaque, hwaddr addr, unsigned int size)
+{
+    MCPXAPUState *d = opaque;
+
+    assert(size == 4);
+    assert(addr % 4 == 0);
+
+    uint64_t r = 0;
+    switch (addr) {
+    case NV_PAPU_EPXMEM ... NV_PAPU_EPXMEM + 0xC00 * 4 - 1: {
+        uint32_t xaddr = (addr - NV_PAPU_EPXMEM) / 4;
+        r = dsp_read_memory(d->ep.dsp, 'X', xaddr);
+        // fprintf(stderr, "read EP  NV_PAPU_EPXMEM [%x] -> %x\n", xaddr, r);
+        break;
+    }
+    case NV_PAPU_EPYMEM ... NV_PAPU_EPYMEM + 0x100 * 4 - 1: {
+        uint32_t yaddr = (addr - NV_PAPU_EPYMEM) / 4;
+        r = dsp_read_memory(d->ep.dsp, 'Y', yaddr);
+        // fprintf(stderr, "read EP  NV_PAPU_EPYMEM [%x] -> %x\n", yaddr, r);
+        break;
+    }
+    case NV_PAPU_EPPMEM ... NV_PAPU_EPPMEM + 0x1000 * 4 - 1: {
+        uint32_t paddr = (addr - NV_PAPU_EPPMEM) / 4;
+        r = dsp_read_memory(d->ep.dsp, 'P', paddr);
+        // fprintf(stderr, "read EP  NV_PAPU_EPPMEM [%x] -> %x\n", paddr, r);
+        break;
+    }
+    default:
+        r = d->ep.regs[addr];
+        break;
+    }
+    DPRINTF("mcpx apu EP: read [0x%" HWADDR_PRIx "] -> 0x%lx\n", addr, r);
+
+    return r;
+}
+
+static void ep_write(void *opaque, hwaddr addr, uint64_t val, unsigned int size)
+{
+    MCPXAPUState *d = opaque;
+
+    qemu_mutex_lock(&d->lock);
+
+    assert(size == 4);
+    assert(addr % 4 == 0);
+
+    DPRINTF("mcpx apu EP: [0x%" HWADDR_PRIx "] = 0x%lx\n", addr, val);
+
+    switch (addr) {
+    case NV_PAPU_EPXMEM ... NV_PAPU_EPXMEM + 0xC00 * 4 - 1: {
+        uint32_t xaddr = (addr - NV_PAPU_EPXMEM) / 4;
+        dsp_write_memory(d->ep.dsp, 'X', xaddr, val);
+        // fprintf(stderr, "ep write xmem %x = %x\n", xaddr, val);
+        break;
+    }
+    case NV_PAPU_EPYMEM ... NV_PAPU_EPYMEM + 0x100 * 4 - 1: {
+        uint32_t yaddr = (addr - NV_PAPU_EPYMEM) / 4;
+        dsp_write_memory(d->ep.dsp, 'Y', yaddr, val);
+        // fprintf(stderr, "ep write ymem %x = %x\n", yaddr, val);
+        break;
+    }
+    case NV_PAPU_EPPMEM ... NV_PAPU_EPPMEM + 0x1000 * 4 - 1: {
+        uint32_t paddr = (addr - NV_PAPU_EPPMEM) / 4;
+        // fprintf(stderr, "ep write pmem %x = %x\n", paddr, val);
+        dsp_write_memory(d->ep.dsp, 'P', paddr, val);
+        break;
+    }
+    case NV_PAPU_EPRST:
+        proc_rst_write(d->ep.dsp, d->ep.regs[NV_PAPU_EPRST], val);
+        d->ep.regs[NV_PAPU_EPRST] = val;
+        d->ep_frame_div = 0; /* FIXME: Still unsure about frame sync */
+        break;
+    default:
+        d->ep.regs[addr] = val;
+        break;
+    }
+
+    qemu_mutex_unlock(&d->lock);
+}
+
+const MemoryRegionOps ep_ops = {
+    .read = ep_read,
+    .write = ep_write,
+};
+
+void mcpx_apu_dsp_frame(MCPXAPUState *d, float mixbins[NUM_MIXBINS][NUM_SAMPLES_PER_FRAME])
+{
+    /* Write VP results to the GP DSP MIXBUF */
+    for (int mixbin = 0; mixbin < NUM_MIXBINS; mixbin++) {
+        uint32_t base = GP_DSP_MIXBUF_BASE + mixbin * NUM_SAMPLES_PER_FRAME;
+        for (int sample = 0; sample < NUM_SAMPLES_PER_FRAME; sample++) {
+            dsp_write_memory(d->gp.dsp, 'X', base + sample,
+                             float_to_24b(mixbins[mixbin][sample]));
+        }
+    }
+
+    bool ep_enabled = (d->ep.regs[NV_PAPU_EPRST] & NV_PAPU_GPRST_GPRST) &&
+                      (d->ep.regs[NV_PAPU_EPRST] & NV_PAPU_GPRST_GPDSPRST);
+
+    /* Run GP */
+    if ((d->gp.regs[NV_PAPU_GPRST] & NV_PAPU_GPRST_GPRST) &&
+        (d->gp.regs[NV_PAPU_GPRST] & NV_PAPU_GPRST_GPDSPRST)) {
+        dsp_start_frame(d->gp.dsp);
+        d->gp.dsp->core.is_idle = false;
+        d->gp.dsp->core.cycle_count = 0;
+        do {
+            dsp_run(d->gp.dsp, 1000);
+        } while (!d->gp.dsp->core.is_idle && d->gp.realtime);
+        g_dbg.gp.cycles = d->gp.dsp->core.cycle_count;
+
+        if ((d->monitor.point == MCPX_APU_DEBUG_MON_GP) ||
+            (d->monitor.point == MCPX_APU_DEBUG_MON_GP_OR_EP && !ep_enabled)) {
+            int off = (d->ep_frame_div % 8) * NUM_SAMPLES_PER_FRAME;
+            for (int i = 0; i < NUM_SAMPLES_PER_FRAME; i++) {
+                uint32_t l = dsp_read_memory(d->gp.dsp, 'X', 0x1400 + i);
+                d->monitor.frame_buf[off + i][0] = l >> 8;
+                uint32_t r =
+                    dsp_read_memory(d->gp.dsp, 'X', 0x1400 + 1 * 0x20 + i);
+                d->monitor.frame_buf[off + i][1] = r >> 8;
+            }
+        }
+    }
+
+    /* Run EP */
+    if ((d->ep.regs[NV_PAPU_EPRST] & NV_PAPU_GPRST_GPRST) &&
+        (d->ep.regs[NV_PAPU_EPRST] & NV_PAPU_GPRST_GPDSPRST)) {
+        if (d->ep_frame_div % 8 == 0) {
+            dsp_start_frame(d->ep.dsp);
+            d->ep.dsp->core.is_idle = false;
+            d->ep.dsp->core.cycle_count = 0;
+            do {
+                dsp_run(d->ep.dsp, 1000);
+            } while (!d->ep.dsp->core.is_idle && d->ep.realtime);
+            g_dbg.ep.cycles = d->ep.dsp->core.cycle_count;
+        }
+    }
+}
+
+void mcpx_apu_dsp_init(MCPXAPUState *d)
+{
+    d->gp.dsp = dsp_init(d, gp_scratch_rw, gp_fifo_rw);
+    for (int i = 0; i < DSP_PRAM_SIZE; i++) {
+        d->gp.dsp->core.pram[i] = 0xCACACACA;
+    }
+    memset(d->gp.dsp->core.pram_opcache, 0,
+           sizeof(d->gp.dsp->core.pram_opcache));
+    d->gp.dsp->is_gp = true;
+    d->gp.dsp->core.is_gp = true;
+    d->gp.dsp->core.is_idle = false;
+    d->gp.dsp->core.cycle_count = 0;
+
+    d->ep.dsp = dsp_init(d, ep_scratch_rw, ep_fifo_rw);
+    for (int i = 0; i < DSP_PRAM_SIZE; i++) {
+        d->ep.dsp->core.pram[i] = 0xCACACACA;
+    }
+    memset(d->ep.dsp->core.pram_opcache, 0,
+           sizeof(d->ep.dsp->core.pram_opcache));
+    for (int i = 0; i < DSP_XRAM_SIZE; i++) {
+        d->ep.dsp->core.xram[i] = 0xCACACACA;
+    }
+    for (int i = 0; i < DSP_YRAM_SIZE; i++) {
+        d->ep.dsp->core.yram[i] = 0xCACACACA;
+    }
+    d->ep.dsp->is_gp = false;
+    d->ep.dsp->core.is_gp = false;
+    d->ep.dsp->core.is_idle = false;
+    d->ep.dsp->core.cycle_count = 0;
+
+    /* Until DSP is more performant, a switch to decide whether or not we should
+     * use the full audio pipeline or not.
+     */
+    mcpx_apu_update_dsp_preference(d);
+}

hw/xbox/mcpx/apu/dsp/gp_ep.h

@@ -0,0 +1,57 @@
+/*
+ * QEMU MCPX Audio Processing Unit implementation
+ *
+ * Copyright (c) 2012 espes
+ * Copyright (c) 2018-2019 Jannik Vogel
+ * Copyright (c) 2019-2025 Matt Borgerson
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, see <http://www.gnu.org/licenses/>.
+ */
+#ifndef HW_XBOX_MCPX_APU_GP_EP_H
+#define HW_XBOX_MCPX_APU_GP_EP_H
+
+#include "qemu/osdep.h"
+#include "hw/hw.h"
+#include "hw/pci/pci.h"
+#include "hw/xbox/mcpx/apu/apu_regs.h"
+
+#include "dsp.h"
+#include "dsp_dma.h"
+#include "dsp_cpu.h"
+#include "dsp_state.h"
+
+typedef struct MCPXAPUState MCPXAPUState;
+
+typedef struct MCPXAPUGPState {
+    bool realtime;
+    MemoryRegion mmio;
+    DSPState *dsp;
+    uint32_t regs[0x10000];
+} MCPXAPUGPState;
+
+typedef struct MCPXAPUEPState {
+    bool realtime;
+    MemoryRegion mmio;
+    DSPState *dsp;
+    uint32_t regs[0x10000];
+} MCPXAPUEPState;
+
+extern const MemoryRegionOps gp_ops;
+extern const MemoryRegionOps ep_ops;
+
+void mcpx_apu_dsp_init(MCPXAPUState *d);
+void mcpx_apu_update_dsp_preference(MCPXAPUState *d);
+void mcpx_apu_dsp_frame(MCPXAPUState *d, float mixbins[NUM_MIXBINS][NUM_SAMPLES_PER_FRAME]);
+
+#endif

hw/xbox/mcpx/apu/dsp/meson.build

@@ -0,0 +1,4 @@
+libdsp = static_library('dsp', files(['debug.c', 'dsp.c', 'dsp_cpu.c', 'dsp_dma.c']) + genh)
+dsp = declare_dependency(objects: libdsp.extract_all_objects(recursive: false))
+
+mcpx_ss.add(dsp, files('gp_ep.c'))

hw/xbox/mcpx/apu/dsp/trace-events

@@ -0,0 +1,9 @@
+# See docs/devel/tracing.rst for syntax documentation.
+
+# dsp.c
+dsp_read_peripheral(uint32_t addr, uint32_t val) "addr 0x%"PRIx32" val 0x%"PRIx32
+dsp_write_peripheral(uint32_t addr, uint32_t val) "addr 0x%"PRIx32" val 0x%"PRIx32
+
+# dsp_cpu.c
+dsp56k_execute_instruction(uint32_t id, uint32_t pc) "[gp=%d]: pc=0x%"PRIx32
+dsp56k_execute_instruction_disasm(const char *disasm) "%s"

hw/xbox/mcpx/apu/dsp/trace.h

@@ -0,0 +1 @@
+#include "trace/trace-hw_xbox_mcpx_apu_dsp.h"

hw/xbox/mcpx/apu/fpconv.h

@@ -1,7 +1,7 @@
 /*
  * Helper FP conversions
  *
- * Copyright (c) 2020-2021 Matt Borgerson
+ * Copyright (c) 2020-2025 Matt Borgerson
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
@@ -21,27 +21,39 @@
 #ifndef FLOATCONV_H
 #define FLOATCONV_H
 
-static float uint8_to_float(uint8_t value)
+#include <stdint.h>
+
+static inline float int8_to_float(int8_t x)
+{
+    return x / 128.0f;
+}
+
+static inline float uint8_to_float(uint8_t value)
 {
     return ((int)value - 0x80) / (1.0 * 0x80);
 }
 
-static float int16_to_float(int16_t value)
+static inline float int16_to_float(int16_t value)
 {
     return value / (1.0 * 0x8000);
 }
 
-static float int32_to_float(int32_t value)
+static inline float s6p9_to_float(int16_t value)
+{
+    return value / 512.0f;
+}
+
+static inline float int32_to_float(int32_t value)
 {
     return value / (1.0 * 0x80000000);
 }
 
-static float int24_to_float(int32_t value)
+static inline float int24_to_float(int32_t value)
 {
     return int32_to_float((uint32_t)value << 8);
 }
 
-static uint32_t float_to_24b(float value)
+static inline uint32_t float_to_24b(float value)
 {
     double scaled_value = value * (8.0 * 0x100000);
     int int24;

hw/xbox/mcpx/apu/meson.build

@@ -0,0 +1,7 @@
+mcpx_ss.add(sdl, files(
+	'apu.c',
+	'debug.c',
+	))
+
+subdir('vp')
+subdir('dsp')

hw/xbox/mcpx/apu/trace.h

@@ -0,0 +1 @@
+#include "trace/trace-hw_xbox_mcpx_apu.h"

hw/xbox/mcpx/apu/vp/hrtf.h

@@ -0,0 +1,137 @@
+/*
+ * HRTF Filter
+ *
+ * Copyright (c) 2025 Matt Borgerson
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef HW_XBOX_MCPX_HRTF_H
+#define HW_XBOX_MCPX_HRTF_H
+
+#include <string.h>
+#include <stddef.h>
+#include <math.h>
+
+#include "hw/xbox/mcpx/apu/apu_regs.h"
+
+#define HRTF_SAMPLES_PER_FRAME  NUM_SAMPLES_PER_FRAME
+#define HRTF_NUM_TAPS           31
+#define HRTF_MAX_DELAY_SAMPLES  42
+#define HRTF_BUFLEN             (HRTF_NUM_TAPS + HRTF_MAX_DELAY_SAMPLES)
+#define HRTF_PARAM_SMOOTH_ALPHA 0.01f
+
+typedef struct {
+    int buf_pos;
+    struct {
+        float buf[HRTF_BUFLEN];
+        float hrir_coeff_cur[HRTF_NUM_TAPS];
+        float hrir_coeff_tar[HRTF_NUM_TAPS];
+    } ch[2];
+    float itd_cur;
+    float itd_tar;
+} HrtfFilter;
+
+static inline void hrtf_filter_init(HrtfFilter *f)
+{
+    memset(f, 0, sizeof(*f));
+}
+
+static inline void
+hrtf_filter_set_target_params(HrtfFilter *f, float hrir_coeff[2][HRTF_NUM_TAPS],
+                              float itd)
+{
+    f->itd_tar =
+        fmaxf(-HRTF_MAX_DELAY_SAMPLES, fminf(itd, HRTF_MAX_DELAY_SAMPLES));
+
+    for (int ch = 0; ch < 2; ch++) {
+        float *coeff = f->ch[ch].hrir_coeff_tar;
+        memcpy(coeff, hrir_coeff[ch], sizeof(f->ch[ch].hrir_coeff_tar));
+
+        // Normalize coefficients for unity filter gain
+        float s = 0.0f;
+        for (int k = 0; k < HRTF_NUM_TAPS; k++) {
+            s += fabsf(coeff[k]);
+        }
+        if (s == 0.0f || s == 1.0f) {
+            break;
+        }
+        for (int k = 0; k < HRTF_NUM_TAPS; k++) {
+            coeff[k] /= s;
+        }
+    }
+}
+
+static inline float hrtf_filter_smooth_param(float cur, float tar)
+{
+    // FIXME: Match hardware parameter transition
+    return cur + HRTF_PARAM_SMOOTH_ALPHA * (tar - cur);
+}
+
+static inline void hrtf_filter_step_parameters(HrtfFilter *f)
+{
+    for (int ch = 0; ch < 2; ch++) {
+        float *coeff_cur = f->ch[ch].hrir_coeff_cur;
+        float *coeff_tar = f->ch[ch].hrir_coeff_tar;
+        for (int k = 0; k < HRTF_NUM_TAPS; k++) {
+            coeff_cur[k] = hrtf_filter_smooth_param(coeff_cur[k], coeff_tar[k]);
+        }
+    }
+    f->itd_cur = hrtf_filter_smooth_param(f->itd_cur, f->itd_tar);
+}
+
+static inline void hrtf_filter_process(HrtfFilter *f,
+                                       float in[HRTF_SAMPLES_PER_FRAME][2],
+                                       float out[HRTF_SAMPLES_PER_FRAME][2])
+{
+    for (int n = 0; n < HRTF_SAMPLES_PER_FRAME; n++) {
+        hrtf_filter_step_parameters(f);
+
+        for (int ch = 0; ch < 2; ch++) {
+            float *buf = f->ch[ch].buf;
+            float *coeff = f->ch[ch].hrir_coeff_cur;
+
+            // Push new sample
+            buf[f->buf_pos] = in[n][ch];
+
+            // Interaural time difference (channel delay)
+            float d = f->itd_cur * (ch == 0 ? +1.0f : -1.0f);
+            if (d < 0.0f) {
+                d = 0.0f;
+            }
+            int di = d;
+            float dfrac = d - di;
+
+            // HRIR Convolution
+            float acc = 0.0f;
+            for (int k = 0; k < HRTF_NUM_TAPS; k++) {
+                int idx1 = (f->buf_pos - di - k + HRTF_BUFLEN) % HRTF_BUFLEN;
+                float s = buf[idx1];
+
+                // Linear interpolation for fractional part
+                if (dfrac > 0.0f) {
+                    int idx2 = (idx1 - 1 + HRTF_BUFLEN) % HRTF_BUFLEN;
+                    s = s * (1 - dfrac) + buf[idx2] * dfrac;
+                }
+                acc += coeff[k] * s;
+            }
+
+            out[n][ch] = acc;
+        }
+
+        f->buf_pos = (f->buf_pos + 1) % HRTF_BUFLEN;
+    }
+}
+
+#endif

hw/xbox/mcpx/apu/vp/meson.build

@@ -0,0 +1,3 @@
+mcpx_ss.add(libsamplerate, files(
+	'vp.c'
+	))

hw/xbox/mcpx/apu/vp/vp.h

@@ -0,0 +1,112 @@
+/*
+ * QEMU MCPX Audio Processing Unit implementation
+ *
+ * Copyright (c) 2012 espes
+ * Copyright (c) 2018-2019 Jannik Vogel
+ * Copyright (c) 2019-2025 Matt Borgerson
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, see <http://www.gnu.org/licenses/>.
+ */
+#ifndef HW_XBOX_MCPX_VP_H
+#define HW_XBOX_MCPX_VP_H
+
+#include <samplerate.h>
+
+#include "qemu/osdep.h"
+#include "qemu/thread.h"
+#include "hw/hw.h"
+#include "hw/pci/pci.h"
+#include "hw/xbox/mcpx/apu/apu_regs.h"
+#include "svf.h"
+#include "hrtf.h"
+
+typedef struct MCPXAPUState MCPXAPUState;
+
+typedef struct MCPXAPUVPSSLData {
+    uint32_t base[MCPX_HW_SSLS_PER_VOICE];
+    uint8_t count[MCPX_HW_SSLS_PER_VOICE];
+    int ssl_index;
+    int ssl_seg;
+} MCPXAPUVPSSLData;
+
+typedef struct MCPXAPUVoiceFilter {
+    uint16_t voice;
+    float resample_buf[NUM_SAMPLES_PER_FRAME * 2];
+    SRC_STATE *resampler;
+    sv_filter svf[2];
+    HrtfFilter hrtf;
+} MCPXAPUVoiceFilter;
+
+typedef struct VoiceWorkItem {
+    int voice;
+    int list;
+} VoiceWorkItem;
+
+typedef struct VoiceWorker {
+    QemuThread thread;
+    float mixbins[NUM_MIXBINS][NUM_SAMPLES_PER_FRAME];
+    float sample_buf[NUM_SAMPLES_PER_FRAME][2];
+    VoiceWorkItem queue[MCPX_HW_MAX_VOICES];
+    int queue_len;
+} VoiceWorker;
+
+typedef struct VoiceWorkDispatch {
+    QemuMutex lock;
+    int num_workers;
+    VoiceWorker *workers;
+    bool workers_should_exit;
+    QemuCond work_pending;
+    uint64_t workers_pending;
+    QemuCond work_finished;
+    float mixbins[NUM_MIXBINS][NUM_SAMPLES_PER_FRAME];
+    VoiceWorkItem queue[MCPX_HW_MAX_VOICES];
+    int queue_len;
+} VoiceWorkDispatch;
+
+typedef struct {
+    MemoryRegion mmio;
+    VoiceWorkDispatch voice_work_dispatch;
+    MCPXAPUVoiceFilter filters[MCPX_HW_MAX_VOICES];
+
+    // FIXME: Where are these stored?
+    int ssl_base_page;
+    MCPXAPUVPSSLData ssl[MCPX_HW_MAX_VOICES];
+    uint8_t hrtf_headroom;
+    uint8_t hrtf_submix[4];
+    uint8_t submix_headroom[NUM_MIXBINS];
+    float sample_buf[NUM_SAMPLES_PER_FRAME][2];
+    uint64_t voice_locked[4];
+    QemuSpin voice_spinlocks[MCPX_HW_MAX_VOICES];
+
+    struct {
+        int current_entry;
+        // FIXME: Stored in RAM
+        struct {
+            float hrir[2][HRTF_NUM_TAPS];
+            float itd;
+        } entries[HRTF_ENTRY_COUNT];
+    } hrtf;
+
+    uint32_t inbuf_sge_handle; //FIXME: Where is this stored?
+    uint32_t outbuf_sge_handle; //FIXME: Where is this stored?
+} MCPXAPUVPState;
+
+extern const MemoryRegionOps vp_ops;
+
+void mcpx_apu_vp_init(MCPXAPUState *d);
+void mcpx_apu_vp_finalize(MCPXAPUState *d);
+void mcpx_apu_vp_frame(MCPXAPUState *d, float mixbins[NUM_MIXBINS][NUM_SAMPLES_PER_FRAME]);
+void mcpx_apu_vp_reset(MCPXAPUState *d);
+
+#endif

hw/xbox/mcpx/meson.build

@@ -1,10 +1,8 @@
-mcpx_ss = ss.source_set()
-mcpx_ss.add(sdl, libsamplerate, files(
-	'apu.c',
-	'aci.c',
-	'dsp/dsp.c',
-	'dsp/dsp_cpu.c',
-	'dsp/dsp_dma.c',
-	))
 
-specific_ss.add_all(mcpx_ss)
+mcpx_ss = ss.source_set()
+mcpx_ss.add(files('aci.c'))
+
+subdir('apu')
+subdir('nvnet')
+
+specific_ss.add_all(mcpx_ss)

hw/xbox/mcpx/nvnet/meson.build

@@ -0,0 +1 @@
+mcpx_ss.add(files('nvnet.c'))

hw/xbox/mcpx/nvnet/nvnet_regs.h

@@ -0,0 +1,267 @@
+/*
+ * QEMU nForce Ethernet Controller register definitions
+ *
+ * Copyright (c) 2013 espes
+ * Copyright (c) 2015-2025 Matt Borgerson
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, see <http://www.gnu.org/licenses/>.
+ *
+ * --
+ *
+ * Most definitions are based on forcedeth.c, taken from cromwell project.
+ * Original forcedeth.c license follows:
+ *
+ * --
+ *    forcedeth.c -- Etherboot device driver for the NVIDIA nForce
+ *           media access controllers.
+ *
+ * Note: This driver is based on the Linux driver that was based on
+ *      a cleanroom reimplementation which was based on reverse
+ *      engineered documentation written by Carl-Daniel Hailfinger
+ *      and Andrew de Quincey. It's neither supported nor endorsed
+ *      by NVIDIA Corp. Use at your own risk.
+ *
+ *    Written 2004 by Timothy Legge <tlegge@rogers.com>
+ *
+ *    This program is free software; you can redistribute it and/or modify
+ *    it under the terms of the GNU General Public License as published by
+ *    the Free Software Foundation; either version 2 of the License, or
+ *    (at your option) any later version.
+ *
+ *    This program is distributed in the hope that it will be useful,
+ *    but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *    GNU General Public License for more details.
+ *
+ *    You should have received a copy of the GNU General Public License
+ *    along with this program; if not, write to the Free Software
+ *    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ *
+ *    Portions of this code based on:
+ *       forcedeth: Ethernet driver for NVIDIA nForce media access controllers:
+ *
+ *   (C) 2003 Manfred Spraul
+ *       See Linux Driver for full information
+ *
+ *   Linux Driver Version 0.22, 19 Jan 2004
+ *
+ *
+ *    REVISION HISTORY:
+ *    ================
+ *    v1.0   01-31-2004  timlegge    Initial port of Linux driver
+ *    v1.1   02-03-2004  timlegge    Large Clean up, first release
+ *
+ *    Indent Options: indent -kr -i8
+ ***************************************************************************/
+
+#ifndef HW_NVNET_REGS_H
+#define HW_NVNET_REGS_H
+
+// clang-format off
+
+#define NVNET_IRQ_STATUS                         0x000
+#  define NVNET_IRQ_STATUS_RX                      0x00000002
+#  define NVNET_IRQ_STATUS_RX_NOBUF                0x00000004
+#  define NVNET_IRQ_STATUS_TX_ERR                  0x00000008
+#  define NVNET_IRQ_STATUS_TX                      0x00000010
+#  define NVNET_IRQ_STATUS_TIMER                   0x00000020
+#  define NVNET_IRQ_STATUS_MIIEVENT                0x00000040
+#define NVNET_IRQ_MASK                           0x004
+#define NVNET_UNKNOWN_SETUP_REG6                 0x008
+#  define NVNET_UNKNOWN_SETUP_REG6_VAL             3
+/*
+ * NVNET_POLLING_INTERVAL_DEFAULT is the interval length of the timer source on the nic
+ * NVNET_POLLING_INTERVAL_DEFAULT=97 would result in an interval length of 1 ms
+ */
+#define NVNET_POLLING_INTERVAL                   0x00C
+#  define NVNET_POLLING_INTERVAL_DEFAULT           970
+#define NVNET_MISC1                              0x080
+#  define NVNET_MISC1_HD                           0x00000002
+#  define NVNET_MISC1_FORCE                        0x003B0F3C
+#define NVNET_TRANSMITTER_CONTROL                0x084
+#  define NVNET_TRANSMITTER_CONTROL_START          0x00000001
+#define NVNET_TRANSMITTER_STATUS                 0x088
+#  define NVNET_TRANSMITTER_STATUS_BUSY            0x00000001
+#define NVNET_PACKET_FILTER                      0x08C
+#  define NVNET_PACKET_FILTER_ALWAYS               0x007F0008
+#  define NVNET_PACKET_FILTER_PROMISC              0x00000080
+#  define NVNET_PACKET_FILTER_MYADDR               0x00000020
+#define NVNET_OFFLOAD                            0x090
+#  define NVNET_OFFLOAD_HOMEPHY                    0x00000601
+#  define NVNET_OFFLOAD_NORMAL                     0x000005EE
+#define NVNET_RECEIVER_CONTROL                   0x094
+#  define NVNET_RECEIVER_CONTROL_START             0x00000001
+#define NVNET_RECEIVER_STATUS                    0x098
+#  define NVNET_RECEIVER_STATUS_BUSY               0x00000001
+#define NVNET_RANDOM_SEED                        0x09C
+#  define NVNET_RANDOM_SEED_MASK                   0x000000FF
+#  define NVNET_RANDOM_SEED_FORCE                  0x00007F00
+#define NVNET_UNKNOWN_SETUP_REG1                 0x0A0
+#  define NVNET_UNKNOWN_SETUP_REG1_VAL             0x0016070F
+#define NVNET_UNKNOWN_SETUP_REG2                 0x0A4
+#  define NVNET_UNKNOWN_SETUP_REG2_VAL             0x00000016
+#define NVNET_MAC_ADDR_A                         0x0A8
+#define NVNET_MAC_ADDR_B                         0x0AC
+#define NVNET_MULTICAST_ADDR_A                   0x0B0
+#  define NVNET_MULTICAST_ADDR_A_FORCE             0x00000001
+#define NVNET_MULTICAST_ADDR_B                   0x0B4
+#define NVNET_MULTICAST_MASK_A                   0x0B8
+#define NVNET_MULTICAST_MASK_B                   0x0BC
+#define NVNET_TX_RING_PHYS_ADDR                  0x100
+#define NVNET_RX_RING_PHYS_ADDR                  0x104
+#define NVNET_RING_SIZE                          0x108
+#  define NVNET_RING_SIZE_TX                       0x0000FFFF
+#  define NVNET_RING_SIZE_RX                       0xFFFF0000
+#define NVNET_UNKNOWN_TRANSMITTER_REG            0x10C
+#define NVNET_LINKSPEED                          0x110
+#  define NVNET_LINKSPEED_FORCE                    0x00010000
+#  define NVNET_LINKSPEED_10                       10
+#  define NVNET_LINKSPEED_100                      100
+#  define NVNET_LINKSPEED_1000                     1000
+#define NVNET_TX_RING_CURRENT_DESC_PHYS_ADDR     0x11C
+#define NVNET_RX_RING_CURRENT_DESC_PHYS_ADDR     0x120
+#define NVNET_TX_CURRENT_BUFFER_PHYS_ADDR        0x124
+#define NVNET_RX_CURRENT_BUFFER_PHYS_ADDR        0x12C
+#define NVNET_UNKNOWN_SETUP_REG5                 0x130
+#  define NVNET_UNKNOWN_SETUP_REG5_BIT31           (1 << 31)
+#define NVNET_TX_RING_NEXT_DESC_PHYS_ADDR        0x134
+#define NVNET_RX_RING_NEXT_DESC_PHYS_ADDR        0x138
+#define NVNET_UNKNOWN_SETUP_REG8                 0x13C
+#  define NVNET_UNKNOWN_SETUP_REG8_VAL1            0x00300010
+#define NVNET_UNKNOWN_SETUP_REG7                 0x140
+#  define NVNET_UNKNOWN_SETUP_REG7_VAL             0x00300010
+#define NVNET_TX_RX_CONTROL                      0x144
+#  define NVNET_TX_RX_CONTROL_KICK                 0x00000001
+#  define NVNET_TX_RX_CONTROL_BIT1                 0x00000002
+#  define NVNET_TX_RX_CONTROL_BIT2                 0x00000004
+#  define NVNET_TX_RX_CONTROL_IDLE                 0x00000008
+#  define NVNET_TX_RX_CONTROL_RESET                0x00000010
+#define NVNET_MII_STATUS                         0x180
+#  define NVNET_MII_STATUS_ERROR                   0x00000001
+#  define NVNET_MII_STATUS_LINKCHANGE              0x00000008
+#define NVNET_UNKNOWN_SETUP_REG4                 0x184
+#  define NVNET_UNKNOWN_SETUP_REG4_VAL             8
+#define NVNET_ADAPTER_CONTROL                    0x188
+#  define NVNET_ADAPTER_CONTROL_START              0x00000002
+#  define NVNET_ADAPTER_CONTROL_LINKUP             0x00000004
+#  define NVNET_ADAPTER_CONTROL_PHYVALID           0x00004000
+#  define NVNET_ADAPTER_CONTROL_RUNNING            0x00100000
+#  define NVNET_ADAPTER_CONTROL_PHYSHIFT           24
+#define NVNET_MII_SPEED                          0x18C
+#  define NVNET_MII_SPEED_BIT8                     (1 << 8)
+#  define NVNET_MII_SPEED_DELAY                    5
+#define NVNET_MDIO_ADDR                          0x190
+#  define NVNET_MDIO_ADDR_INUSE                    0x00008000
+#  define NVNET_MDIO_ADDR_WRITE                    0x00000400
+#  define NVNET_MDIO_ADDR_PHYADDR                  0x000003E0
+#  define NVNET_MDIO_ADDR_PHYREG                   0x0000001F
+#define NVNET_MDIO_DATA                          0x194
+#define NVNET_WAKEUPFLAGS                        0x200
+#  define NVNET_WAKEUPFLAGS_VAL                    0x00007770
+#  define NVNET_WAKEUPFLAGS_BUSYSHIFT              24
+#  define NVNET_WAKEUPFLAGS_ENABLESHIFT            16
+#  define NVNET_WAKEUPFLAGS_D3SHIFT                12
+#  define NVNET_WAKEUPFLAGS_D2SHIFT                8
+#  define NVNET_WAKEUPFLAGS_D1SHIFT                4
+#  define NVNET_WAKEUPFLAGS_D0SHIFT                0
+#  define NVNET_WAKEUPFLAGS_ACCEPT_MAGPAT          0x00000001
+#  define NVNET_WAKEUPFLAGS_ACCEPT_WAKEUPPAT       0x00000002
+#  define NVNET_WAKEUPFLAGS_ACCEPT_LINKCHANGE      0x00000004
+#define NVNET_PATTERN_CRC                        0x204
+#define NVNET_PATTERN_MASK                       0x208
+#define NVNET_POWERCAP                           0x268
+#  define NVNET_POWERCAP_D3SUPP                    (1 << 30)
+#  define NVNET_POWERCAP_D2SUPP                    (1 << 26)
+#  define NVNET_POWERCAP_D1SUPP                    (1 << 25)
+#define NVNET_POWERSTATE                         0x26C
+#  define NVNET_POWERSTATE_POWEREDUP               0x00008000
+#  define NVNET_POWERSTATE_VALID                   0x00000100
+#  define NVNET_POWERSTATE_MASK                    0x00000003
+#  define NVNET_POWERSTATE_D0                      0x00000000
+#  define NVNET_POWERSTATE_D1                      0x00000001
+#  define NVNET_POWERSTATE_D2                      0x00000002
+#  define NVNET_POWERSTATE_D3                      0x00000003
+
+#define NV_TX_LASTPACKET      (1 << 0)
+#define NV_TX_RETRYERROR      (1 << 3)
+#define NV_TX_LASTPACKET1     (1 << 8)
+#define NV_TX_DEFERRED        (1 << 10)
+#define NV_TX_CARRIERLOST     (1 << 11)
+#define NV_TX_LATECOLLISION   (1 << 12)
+#define NV_TX_UNDERFLOW       (1 << 13)
+#define NV_TX_ERROR           (1 << 14)
+#define NV_TX_VALID           (1 << 15)
+#define NV_RX_DESCRIPTORVALID (1 << 0)
+#define NV_RX_MISSEDFRAME     (1 << 1)
+#define NV_RX_SUBSTRACT1      (1 << 3)
+#define NV_RX_BIT4            (1 << 4)
+#define NV_RX_ERROR1          (1 << 7)
+#define NV_RX_ERROR2          (1 << 8)
+#define NV_RX_ERROR3          (1 << 9)
+#define NV_RX_ERROR4          (1 << 10)
+#define NV_RX_CRCERR          (1 << 11)
+#define NV_RX_OVERFLOW        (1 << 12)
+#define NV_RX_FRAMINGERR      (1 << 13)
+#define NV_RX_ERROR           (1 << 14)
+#define NV_RX_AVAIL           (1 << 15)
+
+/* Miscelaneous hardware related defines: */
+#define NV_PCI_REGSZ          0x270
+
+/* various timeout delays: all in usec */
+#define NV_TXRX_RESET_DELAY   4
+#define NV_TXSTOP_DELAY1      10
+#define NV_TXSTOP_DELAY1MAX   500000
+#define NV_TXSTOP_DELAY2      100
+#define NV_RXSTOP_DELAY1      10
+#define NV_RXSTOP_DELAY1MAX   500000
+#define NV_RXSTOP_DELAY2      100
+#define NV_SETUP5_DELAY       5
+#define NV_SETUP5_DELAYMAX    50000
+#define NV_POWERUP_DELAY      5
+#define NV_POWERUP_DELAYMAX   5000
+#define NV_MIIBUSY_DELAY      50
+#define NV_MIIPHY_DELAY       10
+#define NV_MIIPHY_DELAYMAX    10000
+#define NV_WAKEUPPATTERNS     5
+#define NV_WAKEUPMASKENTRIES  4
+
+/* General driver defaults */
+#define NV_WATCHDOG_TIMEO     (2 * HZ)
+#define DEFAULT_MTU           1500
+
+#define RX_RING               4
+#define TX_RING               2
+/* limited to 1 packet until we understand NV_TX_LASTPACKET */
+#define TX_LIMIT_STOP         10
+#define TX_LIMIT_START        5
+
+/* rx / tx mac addr + type + vlan + align + slack*/
+#define RX_NIC_BUFSIZE        (DEFAULT_MTU + 64)
+/* even more slack */
+#define RX_ALLOC_BUFSIZE      (DEFAULT_MTU + 128)
+#define TX_ALLOC_BUFSIZE      (DEFAULT_MTU + 128)
+
+#define OOM_REFILL            (1 + HZ / 20)
+#define POLL_WAIT             (1 + HZ / 100)
+
+/* Link partner ability register. */
+#define LPA_SLCT     0x001F  /* Same as advertise selector  */
+#define LPA_RESV     0x1C00  /* Unused...                   */
+#define LPA_RFAULT   0x2000  /* Link partner faulted        */
+#define LPA_NPAGE    0x8000  /* Next page bit               */
+
+// clang-format on
+
+#endif /* HW_NVNET_REGS_H */

hw/xbox/mcpx/nvnet/trace-events

@@ -1,8 +1,8 @@
 # See docs/devel/tracing.rst for syntax documentation.
 
 # nvnet.c
-nvnet_mii_read(unsigned int phy_addr, uint32_t addr, const char *name, uint64_t val) "phy %d addr 0x%"PRIx32" %s val 0x%"PRIx64
-nvnet_mii_write(unsigned int phy_addr, uint32_t addr, const char *name, uint64_t val) "phy %d addr 0x%"PRIx32" %s val 0x%"PRIx64
+nvnet_phy_reg_read(unsigned int phy_addr, uint8_t addr, const char *name, uint16_t val) "phy %d addr 0x%"PRIx8" %s val 0x%"PRIx16
+nvnet_phy_reg_write(unsigned int phy_addr, uint8_t addr, const char *name, uint16_t val) "phy %d addr 0x%"PRIx8" %s val 0x%"PRIx16
 nvnet_reg_read(uint32_t addr, const char *name, unsigned int size, uint64_t val) "addr 0x%"PRIx32" %s size %d val 0x%"PRIx64
 nvnet_reg_write(uint32_t addr, const char *name, unsigned int size, uint64_t val) "addr 0x%"PRIx32" %s size %d val 0x%"PRIx64
 nvnet_io_read(uint32_t addr, unsigned int size, uint64_t val) "addr 0x%"PRIx32" size %d val 0x%"PRIx64
@@ -14,3 +14,13 @@ nvnet_rx_filter_ucast_match(uint8_t b0, uint8_t b1, uint8_t b2, uint8_t b3, uint
 nvnet_rx_filter_ucast_mismatch(uint8_t b0, uint8_t b1, uint8_t b2, uint8_t b3, uint8_t b4, uint8_t b5) "unicast mismatch: %02x:%02x:%02x:%02x:%02x:%02x"
 nvnet_rx_oversized(size_t size) "Received packet dropped because it was oversized (%zu bytes)"
 nvnet_rx_filter_dropped(void) "Received packet dropped by RX filter"
+nvnet_link_negotiation_start(void) "Start link auto negotiation"
+nvnet_link_negotiation_done(void) "Auto negotiation is completed"
+nvnet_link_status_changed(bool status) "New link status: %d"
+nvnet_cant_rx(bool rx_en, bool dma_en, bool link_up, bool buf_avail) "rx_en:%d dma_en:%d link_up:%d buf_avail:%d"
+nvnet_cant_tx(bool tx_en, bool dma_en, bool link_up) "tx_en:%d dma_en:%d link_up:%d"
+nvnet_update_irq(uint32_t status, uint32_t mask) "status 0x%"PRIx32" mask 0x%"PRIx32
+nvnet_desc_store(uint32_t desc_addr, uint32_t buf_addr, uint16_t length, uint16_t flags) "addr 0x%"PRIx32" buf 0x%"PRIx32" length 0x%"PRIx16" flags 0x%"PRIx16
+nvnet_rx_dma(uint32_t addr, size_t size) "addr 0x%"PRIx32" size 0x%zx"
+nvnet_tx_dma(uint32_t addr, size_t size) "addr 0x%"PRIx32" size 0x%zx"
+nvnet_packet_tx(size_t size) "size 0x%zx"

hw/xbox/mcpx/nvnet/trace.h

@@ -0,0 +1 @@
+#include "trace/trace-hw_xbox_mcpx_nvnet.h"

hw/xbox/mcpx/trace.h

@@ -1 +0,0 @@
-#include "trace/trace-hw_xbox_mcpx.h"

hw/xbox/meson.build

@@ -5,7 +5,6 @@ specific_ss.add(files(
 	# 'chihiro.c',
 	'eeprom_generation.c',
 	'lpc47m157.c',
-	'nvnet.c',
 	'smbus_adm1032.c',
 	'smbus_cx25871.c',
 	'smbus_fs454.c',

hw/xbox/nv2a/debug.h

@@ -155,8 +155,9 @@ static inline void nv2a_profile_inc_counter(enum NV2A_PROF_COUNTERS_ENUM cnt)
 void nv2a_dbg_renderdoc_init(void);
 void *nv2a_dbg_renderdoc_get_api(void);
 bool nv2a_dbg_renderdoc_available(void);
-void nv2a_dbg_renderdoc_capture_frames(int num_frames);
+void nv2a_dbg_renderdoc_capture_frames(int num_frames, bool trace);
 extern int renderdoc_capture_frames;
+extern bool renderdoc_trace_frames;
 #endif
 
 #ifdef __cplusplus

hw/xbox/nv2a/nv2a.c

@@ -423,7 +423,7 @@ const VMStateDescription vmstate_nv2a_pgraph_vertex_attributes = {
 
 static const VMStateDescription vmstate_nv2a = {
     .name = "nv2a",
-    .version_id = 2,
+    .version_id = 3,
     .minimum_version_id = 1,
     .post_save = nv2a_post_save,
     .post_load = nv2a_post_load,
@@ -507,9 +507,11 @@ static const VMStateDescription vmstate_nv2a = {
         VMSTATE_BOOL_ARRAY(pgraph.ltc1_dirty, NV2AState, NV2A_LTC1_COUNT),
         VMSTATE_STRUCT_ARRAY(pgraph.vertex_attributes, NV2AState, NV2A_VERTEXSHADER_ATTRIBUTES, 1, vmstate_nv2a_pgraph_vertex_attributes, VertexAttribute),
         VMSTATE_UINT32(pgraph.inline_array_length, NV2AState),
-        VMSTATE_UINT32_ARRAY(pgraph.inline_array, NV2AState, NV2A_MAX_BATCH_LENGTH),
+        VMSTATE_UINT32_SUB_ARRAY(pgraph.inline_array, NV2AState, 0, NV2A_MAX_BATCH_LENGTH_V2),
+        VMSTATE_UINT32_SUB_ARRAY_V(pgraph.inline_array, NV2AState, NV2A_MAX_BATCH_LENGTH_V2, NV2A_MAX_BATCH_LENGTH - NV2A_MAX_BATCH_LENGTH_V2, 3),
         VMSTATE_UINT32(pgraph.inline_elements_length, NV2AState), // fixme
-        VMSTATE_UINT32_ARRAY(pgraph.inline_elements, NV2AState, NV2A_MAX_BATCH_LENGTH),
+        VMSTATE_UINT32_SUB_ARRAY(pgraph.inline_elements, NV2AState, 0, NV2A_MAX_BATCH_LENGTH_V2),
+        VMSTATE_UINT32_SUB_ARRAY_V(pgraph.inline_elements, NV2AState, NV2A_MAX_BATCH_LENGTH_V2, NV2A_MAX_BATCH_LENGTH - NV2A_MAX_BATCH_LENGTH_V2, 3),
         VMSTATE_UINT32(pgraph.inline_buffer_length, NV2AState), // fixme
         VMSTATE_UINT32(pgraph.draw_arrays_length, NV2AState),
         VMSTATE_UINT32(pgraph.draw_arrays_max_count, NV2AState),

hw/xbox/nv2a/nv2a_regs.h

@@ -315,11 +315,14 @@
 #define NV_PGRAPH_CSV0_C                                 0x00000FB8
 #   define NV_PGRAPH_CSV0_C_CHEOPS_PROGRAM_START                0x0000FF00
 #   define NV_PGRAPH_CSV0_C_SPECULAR_ENABLE                     (1 << 16)
+#   define NV_PGRAPH_CSV0_C_ALPHA_FROM_MATERIAL_SPECULAR        (1 << 17)
+#   define NV_PGRAPH_CSV0_C_SEPARATE_SPECULAR                   (1 << 18)
 #   define NV_PGRAPH_CSV0_C_SPECULAR                            (3 << 19)
 #   define NV_PGRAPH_CSV0_C_DIFFUSE                             (3 << 21)
 #   define NV_PGRAPH_CSV0_C_AMBIENT                             (3 << 23)
 #   define NV_PGRAPH_CSV0_C_EMISSION                            (3 << 25)
 #   define NV_PGRAPH_CSV0_C_NORMALIZATION_ENABLE                (1 << 27)
+#   define NV_PGRAPH_CSV0_C_LOCALEYE                            (1 << 30)
 #   define NV_PGRAPH_CSV0_C_LIGHTING                            (1 << 31)
 #define NV_PGRAPH_CSV1_B                                 0x00000FBC
 #define NV_PGRAPH_CSV1_A                                 0x00000FC0
@@ -406,6 +409,10 @@
 #       define NV_PGRAPH_CLEARRECTY_YMIN                          0x00000FFF
 #       define NV_PGRAPH_CLEARRECTY_YMAX                          0x0FFF0000
 #define NV_PGRAPH_COLORCLEARVALUE                        0x0000186C
+#define NV_PGRAPH_COLORKEYCOLOR0                         0x00001870
+#define NV_PGRAPH_COLORKEYCOLOR1                         0x00001874
+#define NV_PGRAPH_COLORKEYCOLOR2                         0x00001878
+#define NV_PGRAPH_COLORKEYCOLOR3                         0x0000187C
 #define NV_PGRAPH_COMBINEFACTOR0                         0x00001880
 #define NV_PGRAPH_COMBINEFACTOR1                         0x000018A0
 #define NV_PGRAPH_COMBINEALPHAI0                         0x000018C0
@@ -530,6 +537,7 @@
 #define NV_PGRAPH_TEXADDRESS2                            0x000019C4
 #define NV_PGRAPH_TEXADDRESS3                            0x000019C8
 #define NV_PGRAPH_TEXCTL0_0                              0x000019CC
+#   define NV_PGRAPH_TEXCTL0_0_COLORKEYMODE                     0x03
 #   define NV_PGRAPH_TEXCTL0_0_ALPHAKILLEN                      (1 << 2)
 #   define NV_PGRAPH_TEXCTL0_0_MAX_LOD_CLAMP                    0x0003FFC0
 #   define NV_PGRAPH_TEXCTL0_0_MIN_LOD_CLAMP                    0x3FFC0000
@@ -882,6 +890,10 @@
 #       define NV097_SET_CONTROL0_STENCIL_WRITE_ENABLE            (1 << 0)
 #       define NV097_SET_CONTROL0_Z_FORMAT                        (1 << 12)
 #       define NV097_SET_CONTROL0_Z_PERSPECTIVE_ENABLE            (1 << 16)
+#   define NV097_SET_LIGHT_CONTROL                            0x00000294
+#       define NV097_SET_LIGHT_CONTROL_SEPARATE_SPECULAR          1
+#       define NV097_SET_LIGHT_CONTROL_LOCALEYE                   (1 << 16)
+#       define NV097_SET_LIGHT_CONTROL_ALPHA_FROM_MATERIAL_SPECULAR (1 << 17)
 #   define NV097_SET_COLOR_MATERIAL                           0x00000298
 #   define NV097_SET_FOG_MODE                                 0x0000029C
 #       define NV097_SET_FOG_MODE_V_LINEAR                        0x2601
@@ -1049,6 +1061,7 @@
 #       define NV097_SET_TEXGEN_VIEW_MODEL_LOCAL_VIEWER           0
 #       define NV097_SET_TEXGEN_VIEW_MODEL_INFINITE_VIEWER        1
 #   define NV097_SET_FOG_PLANE                                0x000009D0
+#   define NV097_SET_SPECULAR_PARAMS                          0x000009E0
 #   define NV097_SET_SCENE_AMBIENT_COLOR                      0x00000A10
 #   define NV097_SET_VIEWPORT_OFFSET                          0x00000A20
 #   define NV097_SET_POINT_PARAMS                             0x00000A30
@@ -1057,6 +1070,7 @@
 #   define NV097_SET_COMBINER_FACTOR1                         0x00000A80
 #   define NV097_SET_COMBINER_ALPHA_OCW                       0x00000AA0
 #   define NV097_SET_COMBINER_COLOR_ICW                       0x00000AC0
+#   define NV097_SET_COLOR_KEY_COLOR                          0x00000AE0
 #   define NV097_SET_VIEWPORT_SCALE                           0x00000AF0
 #   define NV097_SET_TRANSFORM_PROGRAM                        0x00000B00
 #   define NV097_SET_TRANSFORM_CONSTANT                       0x00000B80
@@ -1099,6 +1113,11 @@
 #   define NV097_SET_TEXCOORD3_4F                             0x00001620
 #   define NV097_SET_TEXCOORD3_2S                             0x00001610
 #   define NV097_SET_TEXCOORD3_4S                             0x00001630
+#   define NV097_SET_FOG_COORD                                0x00001698
+#   define NV097_SET_WEIGHT1F                                 0x0000169C
+#   define NV097_SET_WEIGHT2F                                 0x000016A0
+#   define NV097_SET_WEIGHT3F                                 0x000016B0
+#   define NV097_SET_WEIGHT4F                                 0x000016C0
 #   define NV097_SET_VERTEX_DATA_ARRAY_OFFSET                 0x00001720
 #   define NV097_SET_VERTEX_DATA_ARRAY_FORMAT                 0x00001760
 #       define NV097_SET_VERTEX_DATA_ARRAY_FORMAT_TYPE            0x0000000F
@@ -1252,6 +1271,7 @@
 #   define NV097_SET_CLEAR_RECT_HORIZONTAL                    0x00001D98
 #   define NV097_SET_CLEAR_RECT_VERTICAL                      0x00001D9C
 #   define NV097_SET_SPECULAR_FOG_FACTOR                      0x00001E20
+#   define NV097_SET_SPECULAR_PARAMS_BACK                     0x00001E28
 #   define NV097_SET_COMBINER_COLOR_OCW                       0x00001E40
 #   define NV097_SET_COMBINER_CONTROL                         0x00001E60
 #   define NV097_SET_SHADOW_ZSLOPE_THRESHOLD                  0x00001E68
@@ -1453,7 +1473,22 @@
 #define NV2A_NUM_SUBCHANNELS 8
 #define NV2A_CACHE1_SIZE 128
 
-#define NV2A_MAX_BATCH_LENGTH 0x1FFFF
+/* This is a multi-use limit. Testing on an Xbox 1.0, it is possible to send
+ * arrays of at least 0x0FFFFF elements without issue, however sending
+ * NV097_DRAW_ARRAYS with a start value > 0xFFFF raises an exception implying
+ * that there may be a vertex limit. Since xemu uses batch length for vertex
+ * elements in NV097_INLINE_ARRAY the size should ideally be high enough to
+ * accommodate 0xFFFF vertices with maximum attributes specified.
+ *
+ * Retail games are known to send at least 0x410FA elements in a single draw, so
+ * a somewhat larger value is selected to balance memory use with real-world
+ * limits.
+ *
+ * NV2A_MAX_BATCH_LENGTH_V2 is the previous limit, for migration.
+ * FIXME: Remove NV2A_MAX_BATCH_LENGTH_V2 at some point in the future.
+ */
+#define NV2A_MAX_BATCH_LENGTH 0x07FFFF
+#define NV2A_MAX_BATCH_LENGTH_V2 0x1FFFF
 #define NV2A_VERTEXSHADER_ATTRIBUTES 16
 #define NV2A_MAX_TEXTURES 4
 

hw/xbox/nv2a/pgraph/debug_renderdoc.c

@@ -36,6 +36,7 @@
 static RENDERDOC_API_1_6_0 *rdoc_api = NULL;
 
 int renderdoc_capture_frames = 0;
+bool renderdoc_trace_frames = false;
 
 void nv2a_dbg_renderdoc_init(void)
 {
@@ -89,7 +90,8 @@ bool nv2a_dbg_renderdoc_available(void)
     return rdoc_api != NULL;
 }
 
-void nv2a_dbg_renderdoc_capture_frames(int num_frames)
+void nv2a_dbg_renderdoc_capture_frames(int num_frames, bool trace)
 {
     renderdoc_capture_frames += num_frames;
+    renderdoc_trace_frames = trace;
 }

hw/xbox/nv2a/pgraph/gl/constants.h

@@ -298,7 +298,7 @@ static const SurfaceFormatInfo kelvin_surface_color_format_gl_map[] = {
     [NV097_SET_SURFACE_FORMAT_COLOR_LE_B8] =
         {1, GL_R8, GL_RED, GL_UNSIGNED_BYTE, GL_COLOR_ATTACHMENT0},
     [NV097_SET_SURFACE_FORMAT_COLOR_LE_G8B8] =
-        {2, GL_RG8, GL_RG, GL_UNSIGNED_SHORT, GL_COLOR_ATTACHMENT0},
+        {2, GL_RG8, GL_RG, GL_UNSIGNED_BYTE, GL_COLOR_ATTACHMENT0},
 };
 
 static const SurfaceFormatInfo kelvin_surface_zeta_float_format_gl_map[] = {

hw/xbox/nv2a/pgraph/gl/debug.c

@@ -29,6 +29,8 @@
 #include <assert.h>
 
 #ifdef CONFIG_RENDERDOC
+#include "trace/control.h"
+
 #pragma GCC diagnostic ignored "-Wstrict-prototypes"
 #include "thirdparty/renderdoc_app.h"
 #endif
@@ -154,7 +156,8 @@ void gl_debug_frame_terminator(void)
         RENDERDOC_API_1_6_0 *rdoc_api = nv2a_dbg_renderdoc_get_api();
 
         if (rdoc_api->IsTargetControlConnected()) {
-            if (rdoc_api->IsFrameCapturing()) {
+            bool capturing = rdoc_api->IsFrameCapturing();
+            if (capturing && renderdoc_capture_frames == 0) {
                 rdoc_api->EndFrameCapture(NULL, NULL);
                 GLenum error = glGetError();
                 if (error != GL_NO_ERROR) {
@@ -162,14 +165,23 @@ void gl_debug_frame_terminator(void)
                             "Renderdoc EndFrameCapture triggered GL error 0x%X - ignoring\n",
                             error);
                 }
+                if (renderdoc_trace_frames) {
+                    trace_enable_events("-nv2a_pgraph_*");
+                    renderdoc_trace_frames = false;
+                }
             }
             if (renderdoc_capture_frames > 0) {
-                rdoc_api->StartFrameCapture(NULL, NULL);
-                GLenum error = glGetError();
-                if (error != GL_NO_ERROR) {
-                    fprintf(stderr,
-                            "Renderdoc StartFrameCapture triggered GL error 0x%X - ignoring\n",
-                            error);
+                if (!capturing) {
+                    if (renderdoc_trace_frames) {
+                        trace_enable_events("nv2a_pgraph_*");
+                    }
+                    rdoc_api->StartFrameCapture(NULL, NULL);
+                    GLenum error = glGetError();
+                    if (error != GL_NO_ERROR) {
+                        fprintf(stderr,
+                                "Renderdoc StartFrameCapture triggered GL error 0x%X - ignoring\n",
+                                error);
+                    }
                 }
                 --renderdoc_capture_frames;
             }

hw/xbox/nv2a/pgraph/gl/debug.h

@@ -32,10 +32,10 @@
 #include "config-host.h"
 
 void gl_debug_initialize(void);
-void gl_debug_message(bool cc, const char *fmt, ...);
-void gl_debug_group_begin(const char *fmt, ...);
+void gl_debug_message(bool cc, const char *fmt, ...) __attribute__ ((format (printf, 2, 3)));
+void gl_debug_group_begin(const char *fmt, ...) __attribute__ ((format (printf, 1, 2)));
 void gl_debug_group_end(void);
-void gl_debug_label(GLenum target, GLuint name, const char *fmt, ...);
+void gl_debug_label(GLenum target, GLuint name, const char *fmt, ...) __attribute__ ((format (printf, 3, 4)));
 void gl_debug_frame_terminator(void);
 
 # define NV2A_GL_DPRINTF(cc, format, ...) \

hw/xbox/nv2a/pgraph/gl/display.c

@@ -68,7 +68,7 @@ void pgraph_gl_init_display(NV2AState *d)
         "{\n"
         "    vec2 texCoord = gl_FragCoord.xy/display_size;\n"
         "    float rel = display_size.y/textureSize(tex, 0).y/line_offset;\n"
-        "    texCoord.y = 1 + rel*(texCoord.y - 1);"
+        "    texCoord.y = rel*(1.0f - texCoord.y);"
         "    out_Color.rgba = texture(tex, texCoord);\n"
         "    if (pvideo_enable) {\n"
         "        vec2 screenCoord = gl_FragCoord.xy - 0.5;\n"

hw/xbox/nv2a/pgraph/gl/draw.c

@@ -92,7 +92,6 @@ void pgraph_gl_clear_surface(NV2AState *d, uint32_t parameter)
                  scissor_height = ymax - ymin + 1;
     pgraph_apply_anti_aliasing_factor(pg, &xmin, &ymin);
     pgraph_apply_anti_aliasing_factor(pg, &scissor_width, &scissor_height);
-    ymin = pg->surface_binding_dim.height - (ymin + scissor_height);
 
     NV2A_DPRINTF("Translated clear rect to %d,%d - %d,%d\n", xmin, ymin,
                  xmin + scissor_width - 1, ymin + scissor_height - 1);
@@ -204,9 +203,10 @@ void pgraph_gl_draw_begin(NV2AState *d)
     }
 
     /* Front-face select */
+    /* Winding is reverse here because clip-space y-coordinates are inverted */
     glFrontFace(pgraph_reg_r(pg, NV_PGRAPH_SETUPRASTER)
                     & NV_PGRAPH_SETUPRASTER_FRONTFACE
-                        ? GL_CCW : GL_CW);
+                        ? GL_CW : GL_CCW);
 
     /* Polygon offset */
     /* FIXME: GL implementation-specific, maybe do this in VS? */
@@ -340,7 +340,6 @@ void pgraph_gl_draw_begin(NV2AState *d)
 
     pgraph_apply_anti_aliasing_factor(pg, &xmin, &ymin);
     pgraph_apply_anti_aliasing_factor(pg, &scissor_width, &scissor_height);
-    ymin = pg->surface_binding_dim.height - (ymin + scissor_height);
     pgraph_apply_scaling_factor(pg, &xmin, &ymin);
     pgraph_apply_scaling_factor(pg, &scissor_width, &scissor_height);
 

hw/xbox/nv2a/pgraph/gl/renderer.h

@@ -33,7 +33,7 @@
 #include "hw/xbox/nv2a/nv2a_regs.h"
 #include "hw/xbox/nv2a/pgraph/surface.h"
 #include "hw/xbox/nv2a/pgraph/texture.h"
-#include "hw/xbox/nv2a/pgraph/shaders.h"
+#include "hw/xbox/nv2a/pgraph/glsl/shaders.h"
 
 #include "gloffscreen.h"
 #include "constants.h"
@@ -82,6 +82,30 @@ typedef struct TextureBinding {
     GLuint gl_texture;
 } TextureBinding;
 
+typedef struct ShaderModuleCacheKey {
+    GLenum kind;
+    union {
+        struct {
+            VshState state;
+            GenVshGlslOptions glsl_opts;
+        } vsh;
+        struct {
+            GeomState state;
+            GenGeomGlslOptions glsl_opts;
+        } geom;
+        struct {
+            PshState state;
+            GenPshGlslOptions glsl_opts;
+        } psh;
+    };
+} ShaderModuleCacheKey;
+
+typedef struct ShaderModuleCacheEntry {
+    LruNode node;
+    ShaderModuleCacheKey key;
+    GLuint gl_shader;
+} ShaderModuleCacheEntry;
+
 typedef struct ShaderBinding {
     LruNode node;
     bool initialized;
@@ -96,36 +120,10 @@ typedef struct ShaderBinding {
     GLuint gl_program;
     GLenum gl_primitive_mode;
 
-    GLint psh_constant_loc[9][2];
-    GLint alpha_ref_loc;
-
-    GLint bump_mat_loc[NV2A_MAX_TEXTURES];
-    GLint bump_scale_loc[NV2A_MAX_TEXTURES];
-    GLint bump_offset_loc[NV2A_MAX_TEXTURES];
-    GLint tex_scale_loc[NV2A_MAX_TEXTURES];
-
-    GLint surface_size_loc;
-    GLint clip_range_loc;
-    GLint depth_offset_loc;
-
-    GLint vsh_constant_loc[NV2A_VERTEXSHADER_CONSTANTS];
-    uint32_t vsh_constants[NV2A_VERTEXSHADER_CONSTANTS][4];
-
-    GLint inv_viewport_loc;
-    GLint ltctxa_loc[NV2A_LTCTXA_COUNT];
-    GLint ltctxb_loc[NV2A_LTCTXB_COUNT];
-    GLint ltc1_loc[NV2A_LTC1_COUNT];
-
-    GLint fog_color_loc;
-    GLint fog_param_loc;
-    GLint light_infinite_half_vector_loc[NV2A_MAX_LIGHTS];
-    GLint light_infinite_direction_loc[NV2A_MAX_LIGHTS];
-    GLint light_local_position_loc[NV2A_MAX_LIGHTS];
-    GLint light_local_attenuation_loc[NV2A_MAX_LIGHTS];
-
-    GLint clip_region_loc[8];
-
-    GLint material_alpha_loc;
+    struct {
+        PshUniformLocs psh;
+        VshUniformLocs vsh;
+    } uniform_locs;
 } ShaderBinding;
 
 typedef struct VertexKey {
@@ -201,6 +199,9 @@ typedef struct PGRAPHGLState {
     QemuMutex shader_cache_lock;
     QemuThread shader_disk_thread;
 
+    Lru shader_module_cache;
+    ShaderModuleCacheEntry *shader_module_cache_entries;
+
     unsigned int zpass_pixel_count_result;
     unsigned int gl_zpass_pixel_count_query_count;
     GLuint *gl_zpass_pixel_count_queries;

hw/xbox/nv2a/pgraph/gl/shaders.c

@@ -22,14 +22,9 @@
 #include "qemu/osdep.h"
 #include "qemu/fast-hash.h"
 #include "qemu/mstring.h"
-#include <locale.h>
 
 #include "xemu-version.h"
 #include "ui/xemu-settings.h"
-#include "hw/xbox/nv2a/pgraph/glsl/geom.h"
-#include "hw/xbox/nv2a/pgraph/glsl/vsh.h"
-#include "hw/xbox/nv2a/pgraph/glsl/psh.h"
-#include "hw/xbox/nv2a/pgraph/shaders.h"
 #include "hw/xbox/nv2a/pgraph/util.h"
 #include "debug.h"
 #include "renderer.h"
@@ -100,154 +95,131 @@ static GLuint create_gl_shader(GLenum gl_shader_type,
     return shader;
 }
 
-static void update_shader_constant_locations(ShaderBinding *binding)
+static void set_texture_sampler_uniforms(ShaderBinding *binding)
 {
-    char tmp[64];
-
-    /* set texture samplers */
     for (int i = 0; i < NV2A_MAX_TEXTURES; i++) {
         char samplerName[16];
         snprintf(samplerName, sizeof(samplerName), "texSamp%d", i);
-        GLint texSampLoc = glGetUniformLocation(binding->gl_program, samplerName);
+        GLint texSampLoc =
+            glGetUniformLocation(binding->gl_program, samplerName);
         if (texSampLoc >= 0) {
             glUniform1i(texSampLoc, i);
         }
     }
+}
 
-    /* validate the program */
-    glValidateProgram(binding->gl_program);
-    GLint valid = 0;
-    glGetProgramiv(binding->gl_program, GL_VALIDATE_STATUS, &valid);
-    if (!valid) {
-        GLchar log[1024];
-        glGetProgramInfoLog(binding->gl_program, 1024, NULL, log);
-        fprintf(stderr, "nv2a: shader validation failed: %s\n", log);
-        abort();
-    }
+static void update_shader_uniform_locs(ShaderBinding *binding)
+{
+    char tmp[64];
 
-    /* lookup fragment shader uniforms */
-    for (int i = 0; i < 9; i++) {
-        for (int j = 0; j < 2; j++) {
-            snprintf(tmp, sizeof(tmp), "c%d_%d", j, i);
-            binding->psh_constant_loc[i][j] = glGetUniformLocation(binding->gl_program, tmp);
+    for (int i = 0; i < ARRAY_SIZE(binding->uniform_locs.vsh); i++) {
+        const char *name = VshUniformInfo[i].name;
+        if (VshUniformInfo[i].count > 1) {
+            snprintf(tmp, sizeof(tmp), "%s[0]", name);
+            name = tmp;
         }
-    }
-    binding->alpha_ref_loc = glGetUniformLocation(binding->gl_program, "alphaRef");
-    for (int i = 1; i < NV2A_MAX_TEXTURES; i++) {
-        snprintf(tmp, sizeof(tmp), "bumpMat%d", i);
-        binding->bump_mat_loc[i] = glGetUniformLocation(binding->gl_program, tmp);
-        snprintf(tmp, sizeof(tmp), "bumpScale%d", i);
-        binding->bump_scale_loc[i] = glGetUniformLocation(binding->gl_program, tmp);
-        snprintf(tmp, sizeof(tmp), "bumpOffset%d", i);
-        binding->bump_offset_loc[i] = glGetUniformLocation(binding->gl_program, tmp);
+        binding->uniform_locs.vsh[i] = glGetUniformLocation(binding->gl_program, name);
     }
 
-    for (int i = 0; i < NV2A_MAX_TEXTURES; i++) {
-        snprintf(tmp, sizeof(tmp), "texScale%d", i);
-        binding->tex_scale_loc[i] = glGetUniformLocation(binding->gl_program, tmp);
-    }
-
-    /* lookup vertex shader uniforms */
-    for (int i = 0; i < NV2A_VERTEXSHADER_CONSTANTS; i++) {
-        snprintf(tmp, sizeof(tmp), "c[%d]", i);
-        binding->vsh_constant_loc[i] = glGetUniformLocation(binding->gl_program, tmp);
-    }
-    binding->surface_size_loc = glGetUniformLocation(binding->gl_program, "surfaceSize");
-    binding->clip_range_loc = glGetUniformLocation(binding->gl_program, "clipRange");
-    binding->depth_offset_loc = glGetUniformLocation(binding->gl_program, "depthOffset");
-    binding->fog_color_loc = glGetUniformLocation(binding->gl_program, "fogColor");
-    binding->fog_param_loc = glGetUniformLocation(binding->gl_program, "fogParam");
-
-    binding->inv_viewport_loc = glGetUniformLocation(binding->gl_program, "invViewport");
-    for (int i = 0; i < NV2A_LTCTXA_COUNT; i++) {
-        snprintf(tmp, sizeof(tmp), "ltctxa[%d]", i);
-        binding->ltctxa_loc[i] = glGetUniformLocation(binding->gl_program, tmp);
-    }
-    for (int i = 0; i < NV2A_LTCTXB_COUNT; i++) {
-        snprintf(tmp, sizeof(tmp), "ltctxb[%d]", i);
-        binding->ltctxb_loc[i] = glGetUniformLocation(binding->gl_program, tmp);
-    }
-    for (int i = 0; i < NV2A_LTC1_COUNT; i++) {
-        snprintf(tmp, sizeof(tmp), "ltc1[%d]", i);
-        binding->ltc1_loc[i] = glGetUniformLocation(binding->gl_program, tmp);
-    }
-    for (int i = 0; i < NV2A_MAX_LIGHTS; i++) {
-        snprintf(tmp, sizeof(tmp), "lightInfiniteHalfVector%d", i);
-        binding->light_infinite_half_vector_loc[i] =
-            glGetUniformLocation(binding->gl_program, tmp);
-        snprintf(tmp, sizeof(tmp), "lightInfiniteDirection%d", i);
-        binding->light_infinite_direction_loc[i] =
-            glGetUniformLocation(binding->gl_program, tmp);
-
-        snprintf(tmp, sizeof(tmp), "lightLocalPosition%d", i);
-        binding->light_local_position_loc[i] = glGetUniformLocation(binding->gl_program, tmp);
-        snprintf(tmp, sizeof(tmp), "lightLocalAttenuation%d", i);
-        binding->light_local_attenuation_loc[i] =
-            glGetUniformLocation(binding->gl_program, tmp);
-    }
-    for (int i = 0; i < 8; i++) {
-        snprintf(tmp, sizeof(tmp), "clipRegion[%d]", i);
-        binding->clip_region_loc[i] = glGetUniformLocation(binding->gl_program, tmp);
-    }
-
-    if (binding->state.fixed_function) {
-        binding->material_alpha_loc =
-            glGetUniformLocation(binding->gl_program, "material_alpha");
-    } else {
-        binding->material_alpha_loc = -1;
+    for (int i = 0; i < ARRAY_SIZE(binding->uniform_locs.psh); i++) {
+        const char *name = PshUniformInfo[i].name;
+        if (PshUniformInfo[i].count > 1) {
+            snprintf(tmp, sizeof(tmp), "%s[0]", name);
+            name = tmp;
+        }
+        binding->uniform_locs.psh[i] = glGetUniformLocation(binding->gl_program, name);
     }
 }
 
-static void generate_shaders(ShaderBinding *binding)
+static void shader_module_cache_entry_init(Lru *lru, LruNode *node,
+                                           const void *key)
 {
-    char *previous_numeric_locale = setlocale(LC_NUMERIC, NULL);
-    if (previous_numeric_locale) {
-        previous_numeric_locale = g_strdup(previous_numeric_locale);
+    ShaderModuleCacheEntry *module =
+        container_of(node, ShaderModuleCacheEntry, node);
+    memcpy(&module->key, key, sizeof(ShaderModuleCacheKey));
+
+    const char *kind_str;
+    MString *code;
+
+    switch (module->key.kind) {
+    case GL_VERTEX_SHADER:
+        kind_str = "vertex shader";
+        code = pgraph_glsl_gen_vsh(&module->key.vsh.state,
+                                   module->key.vsh.glsl_opts);
+        break;
+    case GL_GEOMETRY_SHADER:
+        kind_str = "geometry shader";
+        code = pgraph_glsl_gen_geom(&module->key.geom.state,
+                                    module->key.geom.glsl_opts);
+        break;
+    case GL_FRAGMENT_SHADER:
+        kind_str = "fragment shader";
+        code = pgraph_glsl_gen_psh(&module->key.psh.state,
+                                   module->key.psh.glsl_opts);
+        break;
+    default:
+        assert(!"Invalid shader module kind");
+        kind_str = "unknown";
+        code = NULL;
     }
 
-    /* Ensure numeric values are printed with '.' radix, no grouping */
-    setlocale(LC_NUMERIC, "C");
+    module->gl_shader =
+        create_gl_shader(module->key.kind, mstring_get_str(code), kind_str);
+    mstring_unref(code);
+}
+
+static void shader_module_cache_entry_post_evict(Lru *lru, LruNode *node)
+{
+    ShaderModuleCacheEntry *module =
+        container_of(node, ShaderModuleCacheEntry, node);
+    glDeleteShader(module->gl_shader);
+}
+
+static bool shader_module_cache_entry_compare(Lru *lru, LruNode *node,
+                                              const void *key)
+{
+    ShaderModuleCacheEntry *module =
+        container_of(node, ShaderModuleCacheEntry, node);
+    return memcmp(&module->key, key, sizeof(ShaderModuleCacheKey));
+}
+
+static GLuint get_shader_module_for_key(PGRAPHGLState *r,
+                                        const ShaderModuleCacheKey *key)
+{
+    uint64_t hash = fast_hash((void *)key, sizeof(ShaderModuleCacheKey));
+    LruNode *node = lru_lookup(&r->shader_module_cache, hash, key);
+    ShaderModuleCacheEntry *module =
+        container_of(node, ShaderModuleCacheEntry, node);
+    return module->gl_shader;
+}
+
+static void generate_shaders(PGRAPHGLState *r, ShaderBinding *binding)
+{
     GLuint program = glCreateProgram();
 
     ShaderState *state = &binding->state;
+    ShaderModuleCacheKey key;
 
-    /* Create an optional geometry shader and find primitive type */
-    GLenum gl_primitive_mode =
-        get_gl_primitive_mode(state->polygon_front_mode, state->primitive_mode);
-    MString* geometry_shader_code =
-        pgraph_gen_geom_glsl(state->polygon_front_mode,
-                                 state->polygon_back_mode,
-                                 state->primitive_mode,
-                                 state->smooth_shading,
-                                 false);
-    if (geometry_shader_code) {
-        const char* geometry_shader_code_str =
-             mstring_get_str(geometry_shader_code);
-        GLuint geometry_shader = create_gl_shader(GL_GEOMETRY_SHADER,
-                                                  geometry_shader_code_str,
-                                                  "geometry shader");
-        glAttachShader(program, geometry_shader);
-        mstring_unref(geometry_shader_code);
+    bool need_geometry_shader = pgraph_glsl_need_geom(&state->geom);
+    if (need_geometry_shader) {
+        memset(&key, 0, sizeof(key));
+        key.kind = GL_GEOMETRY_SHADER;
+        key.geom.state = state->geom;
+        glAttachShader(program, get_shader_module_for_key(r, &key));
     }
 
     /* create the vertex shader */
-    MString *vertex_shader_code =
-        pgraph_gen_vsh_glsl(state, geometry_shader_code != NULL);
-    GLuint vertex_shader = create_gl_shader(GL_VERTEX_SHADER,
-                                            mstring_get_str(vertex_shader_code),
-                                            "vertex shader");
-    glAttachShader(program, vertex_shader);
-    mstring_unref(vertex_shader_code);
+    memset(&key, 0, sizeof(key));
+    key.kind = GL_VERTEX_SHADER;
+    key.vsh.state = state->vsh;
+    key.vsh.glsl_opts.prefix_outputs = need_geometry_shader;
+    glAttachShader(program, get_shader_module_for_key(r, &key));
 
     /* generate a fragment shader from register combiners */
-    MString *fragment_shader_code = pgraph_gen_psh_glsl(state->psh);
-    const char *fragment_shader_code_str =
-        mstring_get_str(fragment_shader_code);
-    GLuint fragment_shader = create_gl_shader(GL_FRAGMENT_SHADER,
-                                              fragment_shader_code_str,
-                                              "fragment shader");
-    glAttachShader(program, fragment_shader);
-    mstring_unref(fragment_shader_code);
+    memset(&key, 0, sizeof(key));
+    key.kind = GL_FRAGMENT_SHADER;
+    key.psh.state = state->psh;
+    glAttachShader(program, get_shader_module_for_key(r, &key));
 
     /* link the program */
     glLinkProgram(program);
@@ -262,15 +234,25 @@ static void generate_shaders(ShaderBinding *binding)
 
     glUseProgram(program);
 
-    binding->initialized = true;
     binding->gl_program = program;
-    binding->gl_primitive_mode = gl_primitive_mode;
-    update_shader_constant_locations(binding);
+    binding->gl_primitive_mode = get_gl_primitive_mode(
+        state->geom.polygon_front_mode, state->geom.primitive_mode);
+    binding->initialized = true;
 
-    if (previous_numeric_locale) {
-        setlocale(LC_NUMERIC, previous_numeric_locale);
-        g_free(previous_numeric_locale);
+    set_texture_sampler_uniforms(binding);
+
+    /* validate the program */
+    GLint valid = 0;
+    glValidateProgram(program);
+    glGetProgramiv(program, GL_VALIDATE_STATUS, &valid);
+    if (!valid) {
+        GLchar log[1024];
+        glGetProgramInfoLog(program, 1024, NULL, log);
+        fprintf(stderr, "nv2a: shader validation failed: %s\n", log);
+        abort();
     }
+
+    update_shader_uniform_locs(binding);
 }
 
 static const char *shader_gl_vendor = NULL;
@@ -346,6 +328,19 @@ bool pgraph_gl_shader_load_from_memory(ShaderBinding *binding)
         return false;
     }
 
+    glUseProgram(gl_program);
+
+    g_free(binding->program);
+
+    binding->program = NULL;
+    binding->gl_program = gl_program;
+    binding->gl_primitive_mode =
+        get_gl_primitive_mode(binding->state.geom.polygon_front_mode,
+                              binding->state.geom.primitive_mode);
+    binding->initialized = true;
+
+    set_texture_sampler_uniforms(binding);
+
     glValidateProgram(gl_program);
     GLint valid = 0;
     glGetProgramiv(gl_program, GL_VALIDATE_STATUS, &valid);
@@ -357,17 +352,7 @@ bool pgraph_gl_shader_load_from_memory(ShaderBinding *binding)
         return false;
     }
 
-    glUseProgram(gl_program);
-
-    binding->gl_program = gl_program;
-    binding->gl_primitive_mode = get_gl_primitive_mode(
-        binding->state.polygon_front_mode, binding->state.primitive_mode);
-    binding->initialized = true;
-
-    g_free(binding->program);
-    binding->program = NULL;
-
-    update_shader_constant_locations(binding);
+    update_shader_uniform_locs(binding);
 
     return true;
 }
@@ -507,7 +492,7 @@ static void *shader_reload_lru_from_disk(void *arg)
     return NULL;
 }
 
-static void shader_cache_entry_init(Lru *lru, LruNode *node, void *state)
+static void shader_cache_entry_init(Lru *lru, LruNode *node, const void *state)
 {
     ShaderBinding *binding = container_of(node, ShaderBinding, node);
     memcpy(&binding->state, state, sizeof(ShaderState));
@@ -537,7 +522,7 @@ static void shader_cache_entry_post_evict(Lru *lru, LruNode *node)
     memset(&binding->state, 0, sizeof(ShaderState));
 }
 
-static bool shader_cache_entry_compare(Lru *lru, LruNode *node, void *key)
+static bool shader_cache_entry_compare(Lru *lru, LruNode *node, const void *key)
 {
     ShaderBinding *binding = container_of(node, ShaderBinding, node);
     return memcmp(&binding->state, key, sizeof(ShaderState));
@@ -571,6 +556,20 @@ void pgraph_gl_init_shaders(PGRAPHState *pg)
 
     qemu_thread_create(&r->shader_disk_thread, "pgraph.renderer_state->shader_cache",
                        shader_reload_lru_from_disk, pg, QEMU_THREAD_JOINABLE);
+
+    /* FIXME: Make this configurable */
+    const size_t shader_module_cache_size = 50*1024;
+    lru_init(&r->shader_module_cache);
+    r->shader_module_cache_entries =
+        g_malloc_n(shader_module_cache_size, sizeof(ShaderModuleCacheEntry));
+    assert(r->shader_module_cache_entries != NULL);
+    for (int i = 0; i < shader_module_cache_size; i++) {
+        lru_add_free(&r->shader_module_cache, &r->shader_module_cache_entries[i].node);
+    }
+
+    r->shader_module_cache.init_node = shader_module_cache_entry_init;
+    r->shader_module_cache.compare_nodes = shader_module_cache_entry_compare;
+    r->shader_module_cache.post_node_evict = shader_module_cache_entry_post_evict;
 }
 
 void pgraph_gl_finalize_shaders(PGRAPHState *pg)
@@ -582,6 +581,10 @@ void pgraph_gl_finalize_shaders(PGRAPHState *pg)
     free(r->shader_cache_entries);
     r->shader_cache_entries = NULL;
 
+    lru_flush(&r->shader_module_cache);
+    g_free(r->shader_module_cache_entries);
+    r->shader_module_cache_entries = NULL;
+
     qemu_mutex_destroy(&r->shader_cache_lock);
 }
 
@@ -685,341 +688,72 @@ void pgraph_gl_shader_cache_to_disk(ShaderBinding *binding)
     qemu_thread_create(binding->save_thread, name, shader_write_to_disk, binding, QEMU_THREAD_JOINABLE);
 }
 
-static void shader_update_constants(PGRAPHState *pg, ShaderBinding *binding,
-                                    bool binding_changed)
+static void apply_uniform_updates(const UniformInfo *info, int *locs,
+                                  void *values, size_t count)
 {
-    PGRAPHGLState *r = pg->gl_renderer_state;
-    int i, j;
-
-    /* update combiner constants */
-    for (i = 0; i < 9; i++) {
-        uint32_t constant[2];
-        if (i == 8) {
-            /* final combiner */
-            constant[0] = pgraph_reg_r(pg, NV_PGRAPH_SPECFOGFACTOR0);
-            constant[1] = pgraph_reg_r(pg, NV_PGRAPH_SPECFOGFACTOR1);
-        } else {
-            constant[0] = pgraph_reg_r(pg, NV_PGRAPH_COMBINEFACTOR0 + i * 4);
-            constant[1] = pgraph_reg_r(pg, NV_PGRAPH_COMBINEFACTOR1 + i * 4);
+    for (int i = 0; i < count; i++) {
+        if (locs[i] == -1) {
+            continue;
         }
 
-        for (j = 0; j < 2; j++) {
-            GLint loc = binding->psh_constant_loc[i][j];
-            if (loc != -1) {
-                float value[4];
-                pgraph_argb_pack32_to_rgba_float(constant[j], value);
-                glUniform4fv(loc, 1, value);
-            }
-        }
-    }
-    if (binding->alpha_ref_loc != -1) {
-        int alpha_ref = GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CONTROL_0),
-                                   NV_PGRAPH_CONTROL_0_ALPHAREF);
-        glUniform1i(binding->alpha_ref_loc, alpha_ref);
-    }
+        void *value = (char*)values + info[i].val_offs;
 
-
-    /* For each texture stage */
-    for (i = 0; i < NV2A_MAX_TEXTURES; i++) {
-        GLint loc;
-
-        /* Bump luminance only during stages 1 - 3 */
-        if (i > 0) {
-            loc = binding->bump_mat_loc[i];
-            if (loc != -1) {
-                uint32_t m_u32[4];
-                m_u32[0] = pgraph_reg_r(pg, NV_PGRAPH_BUMPMAT00 + 4 * (i - 1));
-                m_u32[1] = pgraph_reg_r(pg, NV_PGRAPH_BUMPMAT01 + 4 * (i - 1));
-                m_u32[2] = pgraph_reg_r(pg, NV_PGRAPH_BUMPMAT10 + 4 * (i - 1));
-                m_u32[3] = pgraph_reg_r(pg, NV_PGRAPH_BUMPMAT11 + 4 * (i - 1));
-                float m[4];
-                m[0] = *(float*)&m_u32[0];
-                m[1] = *(float*)&m_u32[1];
-                m[2] = *(float*)&m_u32[2];
-                m[3] = *(float*)&m_u32[3];
-                glUniformMatrix2fv(loc, 1, GL_FALSE, m);
-            }
-            loc = binding->bump_scale_loc[i];
-            if (loc != -1) {
-                uint32_t v =
-                    pgraph_reg_r(pg, NV_PGRAPH_BUMPSCALE1 + (i - 1) * 4);
-                glUniform1f(loc, *(float*)&v);
-            }
-            loc = binding->bump_offset_loc[i];
-            if (loc != -1) {
-                uint32_t v =
-                    pgraph_reg_r(pg, NV_PGRAPH_BUMPOFFSET1 + (i - 1) * 4);
-                glUniform1f(loc, *(float*)&v);
-            }
-        }
-
-        loc = r->shader_binding->tex_scale_loc[i];
-        if (loc != -1) {
-            assert(r->texture_binding[i] != NULL);
-            glUniform1f(loc, (float)r->texture_binding[i]->scale);
+        switch (info[i].type) {
+        case UniformElementType_uint:
+            glUniform1uiv(locs[i], info[i].count, value);
+            break;
+        case UniformElementType_int:
+            glUniform1iv(locs[i], info[i].count, value);
+            break;
+        case UniformElementType_ivec4:
+            glUniform4iv(locs[i], info[i].count, value);
+            break;
+        case UniformElementType_float:
+            glUniform1fv(locs[i], info[i].count, value);
+            break;
+        case UniformElementType_vec2:
+            glUniform2fv(locs[i], info[i].count, value);
+            break;
+        case UniformElementType_vec3:
+            glUniform3fv(locs[i], info[i].count, value);
+            break;
+        case UniformElementType_vec4:
+            glUniform4fv(locs[i], info[i].count, value);
+            break;
+        case UniformElementType_mat2:
+            glUniformMatrix2fv(locs[i], info[i].count, GL_FALSE, value);
+            break;
+        default:
+            g_assert_not_reached();
         }
     }
 
-    if (binding->fog_color_loc != -1) {
-        uint32_t fog_color = pgraph_reg_r(pg, NV_PGRAPH_FOGCOLOR);
-        glUniform4f(binding->fog_color_loc,
-                    GET_MASK(fog_color, NV_PGRAPH_FOGCOLOR_RED) / 255.0,
-                    GET_MASK(fog_color, NV_PGRAPH_FOGCOLOR_GREEN) / 255.0,
-                    GET_MASK(fog_color, NV_PGRAPH_FOGCOLOR_BLUE) / 255.0,
-                    GET_MASK(fog_color, NV_PGRAPH_FOGCOLOR_ALPHA) / 255.0);
-    }
-    if (binding->fog_param_loc != -1) {
-        uint32_t v[2];
-        v[0] = pgraph_reg_r(pg, NV_PGRAPH_FOGPARAM0);
-        v[1] = pgraph_reg_r(pg, NV_PGRAPH_FOGPARAM1);
-        glUniform2f(binding->fog_param_loc, *(float *)&v[0], *(float *)&v[1]);
-    }
-
-    float zmax;
-    switch (pg->surface_shape.zeta_format) {
-    case NV097_SET_SURFACE_FORMAT_ZETA_Z16:
-        zmax = pg->surface_shape.z_format ? f16_max : (float)0xFFFF;
-        break;
-    case NV097_SET_SURFACE_FORMAT_ZETA_Z24S8:
-        zmax = pg->surface_shape.z_format ? f24_max : (float)0xFFFFFF;
-        break;
-    default:
-        assert(0);
-    }
-
-    if (binding->state.fixed_function) {
-        /* update lighting constants */
-        struct {
-            uint32_t* v;
-            bool* dirty;
-            GLint* locs;
-            size_t len;
-        } lighting_arrays[] = {
-            {&pg->ltctxa[0][0], &pg->ltctxa_dirty[0], binding->ltctxa_loc, NV2A_LTCTXA_COUNT},
-            {&pg->ltctxb[0][0], &pg->ltctxb_dirty[0], binding->ltctxb_loc, NV2A_LTCTXB_COUNT},
-            {&pg->ltc1[0][0], &pg->ltc1_dirty[0], binding->ltc1_loc, NV2A_LTC1_COUNT},
-        };
-
-        for (i=0; i<ARRAY_SIZE(lighting_arrays); i++) {
-            uint32_t *lighting_v = lighting_arrays[i].v;
-            bool *lighting_dirty = lighting_arrays[i].dirty;
-            GLint *lighting_locs = lighting_arrays[i].locs;
-            size_t lighting_len = lighting_arrays[i].len;
-            for (j=0; j<lighting_len; j++) {
-                if (!lighting_dirty[j] && !binding_changed) continue;
-                GLint loc = lighting_locs[j];
-                if (loc != -1) {
-                    glUniform4fv(loc, 1, (const GLfloat*)&lighting_v[j*4]);
-                }
-                lighting_dirty[j] = false;
-            }
-        }
-
-        for (i = 0; i < NV2A_MAX_LIGHTS; i++) {
-            GLint loc;
-            loc = binding->light_infinite_half_vector_loc[i];
-            if (loc != -1) {
-                glUniform3fv(loc, 1, pg->light_infinite_half_vector[i]);
-            }
-            loc = binding->light_infinite_direction_loc[i];
-            if (loc != -1) {
-                glUniform3fv(loc, 1, pg->light_infinite_direction[i]);
-            }
-
-            loc = binding->light_local_position_loc[i];
-            if (loc != -1) {
-                glUniform3fv(loc, 1, pg->light_local_position[i]);
-            }
-            loc = binding->light_local_attenuation_loc[i];
-            if (loc != -1) {
-                glUniform3fv(loc, 1, pg->light_local_attenuation[i]);
-            }
-        }
-
-        /* estimate the viewport by assuming it matches the surface ... */
-        unsigned int aa_width = 1, aa_height = 1;
-        pgraph_apply_anti_aliasing_factor(pg, &aa_width, &aa_height);
-
-        float m11 = 0.5 * (pg->surface_binding_dim.width/aa_width);
-        float m22 = -0.5 * (pg->surface_binding_dim.height/aa_height);
-        float m33 = zmax;
-        float m41 = *(float*)&pg->vsh_constants[NV_IGRAPH_XF_XFCTX_VPOFF][0];
-        float m42 = *(float*)&pg->vsh_constants[NV_IGRAPH_XF_XFCTX_VPOFF][1];
-
-        float invViewport[16] = {
-            1.0/m11, 0, 0, 0,
-            0, 1.0/m22, 0, 0,
-            0, 0, 1.0/m33, 0,
-            -1.0+m41/m11, 1.0+m42/m22, 0, 1.0
-        };
-
-        if (binding->inv_viewport_loc != -1) {
-            glUniformMatrix4fv(binding->inv_viewport_loc,
-                               1, GL_FALSE, &invViewport[0]);
-        }
-    }
-
-    /* update vertex program constants */
-    for (i=0; i<NV2A_VERTEXSHADER_CONSTANTS; i++) {
-        if (!pg->vsh_constants_dirty[i] && !binding_changed) continue;
-
-        GLint loc = binding->vsh_constant_loc[i];
-        if ((loc != -1) &&
-            memcmp(binding->vsh_constants[i], pg->vsh_constants[i],
-                   sizeof(pg->vsh_constants[1]))) {
-            glUniform4fv(loc, 1, (const GLfloat *)pg->vsh_constants[i]);
-            memcpy(binding->vsh_constants[i], pg->vsh_constants[i],
-                   sizeof(pg->vsh_constants[i]));
-        }
-
-        pg->vsh_constants_dirty[i] = false;
-    }
-
-    if (binding->surface_size_loc != -1) {
-        unsigned int aa_width = 1, aa_height = 1;
-        pgraph_apply_anti_aliasing_factor(pg, &aa_width, &aa_height);
-        glUniform2f(binding->surface_size_loc,
-                    pg->surface_binding_dim.width / aa_width,
-                    pg->surface_binding_dim.height / aa_height);
-    }
-
-    if (binding->clip_range_loc != -1) {
-        uint32_t v[2];
-        v[0] = pgraph_reg_r(pg, NV_PGRAPH_ZCLIPMIN);
-        v[1] = pgraph_reg_r(pg, NV_PGRAPH_ZCLIPMAX);
-        float zclip_min = *(float *)&v[0];
-        float zclip_max = *(float *)&v[1];
-        glUniform4f(binding->clip_range_loc, 0, zmax, zclip_min, zclip_max);
-    }
-
-    if (binding->depth_offset_loc != -1) {
-        float zbias = 0.0f;
-
-        if (pgraph_reg_r(pg, NV_PGRAPH_SETUPRASTER) &
-            (NV_PGRAPH_SETUPRASTER_POFFSETFILLENABLE |
-             NV_PGRAPH_SETUPRASTER_POFFSETLINEENABLE |
-             NV_PGRAPH_SETUPRASTER_POFFSETPOINTENABLE)) {
-            uint32_t zbias_u32 = pgraph_reg_r(pg, NV_PGRAPH_ZOFFSETBIAS);
-            zbias = *(float *)&zbias_u32;
-
-            if (pgraph_reg_r(pg, NV_PGRAPH_ZOFFSETFACTOR) != 0 &&
-                (pgraph_reg_r(pg, NV_PGRAPH_CONTROL_0) &
-                 NV_PGRAPH_CONTROL_0_Z_PERSPECTIVE_ENABLE)) {
-                /* TODO: emulate zfactor when z_perspective true, i.e.
-                 * w-buffering. Perhaps calculate an additional offset based on
-                 * triangle orientation in geometry shader and pass the result
-                 * to fragment shader and add it to gl_FragDepth as well.
-                 */
-                NV2A_UNIMPLEMENTED("NV_PGRAPH_ZOFFSETFACTOR for w-buffering");
-            }
-        }
-
-        glUniform1f(binding->depth_offset_loc, zbias);
-    }
-
-    /* Clipping regions */
-    unsigned int max_gl_width = pg->surface_binding_dim.width;
-    unsigned int max_gl_height = pg->surface_binding_dim.height;
-    pgraph_apply_scaling_factor(pg, &max_gl_width, &max_gl_height);
-
-    for (i = 0; i < 8; i++) {
-        uint32_t x = pgraph_reg_r(pg, NV_PGRAPH_WINDOWCLIPX0 + i * 4);
-        unsigned int x_min = GET_MASK(x, NV_PGRAPH_WINDOWCLIPX0_XMIN);
-        unsigned int x_max = GET_MASK(x, NV_PGRAPH_WINDOWCLIPX0_XMAX) + 1;
-        uint32_t y = pgraph_reg_r(pg, NV_PGRAPH_WINDOWCLIPY0 + i * 4);
-        unsigned int y_min = GET_MASK(y, NV_PGRAPH_WINDOWCLIPY0_YMIN);
-        unsigned int y_max = GET_MASK(y, NV_PGRAPH_WINDOWCLIPY0_YMAX) + 1;
-        pgraph_apply_anti_aliasing_factor(pg, &x_min, &y_min);
-        pgraph_apply_anti_aliasing_factor(pg, &x_max, &y_max);
-
-        pgraph_apply_scaling_factor(pg, &x_min, &y_min);
-        pgraph_apply_scaling_factor(pg, &x_max, &y_max);
-
-        /* Translate for the GL viewport origin */
-        int y_min_xlat = MAX((int)max_gl_height - (int)y_max, 0);
-        int y_max_xlat = MIN((int)max_gl_height - (int)y_min, max_gl_height);
-
-        glUniform4i(r->shader_binding->clip_region_loc[i],
-                    x_min, y_min_xlat, x_max, y_max_xlat);
-    }
-
-    if (binding->material_alpha_loc != -1) {
-        glUniform1f(binding->material_alpha_loc, pg->material_alpha);
-    }
+    assert(glGetError() == GL_NO_ERROR);
 }
 
-static bool test_shaders_dirty(PGRAPHState *pg)
+// FIXME: Dirty tracking
+// FIXME: Consider UBO to align with VK renderer
+static void update_shader_uniforms(PGRAPHState *pg, ShaderBinding *binding)
 {
-    #define CR_1(reg) CR_x(reg, 1)
-    #define CR_4(reg) CR_x(reg, 4)
-    #define CR_8(reg) CR_x(reg, 8)
-    #define CF(src, name)  CF_x(typeof(src), (&src), name, 1)
-    #define CFA(src, name) CF_x(typeof(src[0]), src, name, ARRAY_SIZE(src))
-    #define CNAME(name) reg_check__ ## name
-    #define CX_x__define(type, name, x) static type CNAME(name)[x];
-    #define CR_x__define(reg, x) CX_x__define(uint32_t, reg, x)
-    #define CF_x__define(type, src, name, x) CX_x__define(type, name, x)
-    #define CR_x__check(reg, x) \
-        for (int i = 0; i < x; i++) { if (pgraph_reg_r(pg, reg+i*4) != CNAME(reg)[i]) goto dirty; }
-    #define CF_x__check(type, src, name, x) \
-        for (int i = 0; i < x; i++) { if (src[i] != CNAME(name)[i]) goto dirty; }
-    #define CR_x__update(reg, x) \
-        for (int i = 0; i < x; i++) { CNAME(reg)[i] = pgraph_reg_r(pg, reg+i*4); }
-    #define CF_x__update(type, src, name, x) \
-        for (int i = 0; i < x; i++) { CNAME(name)[i] = src[i]; }
+    PGRAPHGLState *r = pg->gl_renderer_state;
 
-    #define DIRTY_REGS \
-        CR_1(NV_PGRAPH_COMBINECTL) \
-        CR_1(NV_PGRAPH_SHADERCTL) \
-        CR_1(NV_PGRAPH_SHADOWCTL) \
-        CR_1(NV_PGRAPH_COMBINESPECFOG0) \
-        CR_1(NV_PGRAPH_COMBINESPECFOG1) \
-        CR_1(NV_PGRAPH_CONTROL_0) \
-        CR_1(NV_PGRAPH_CONTROL_3) \
-        CR_1(NV_PGRAPH_CSV0_C) \
-        CR_1(NV_PGRAPH_CSV0_D) \
-        CR_1(NV_PGRAPH_CSV1_A) \
-        CR_1(NV_PGRAPH_CSV1_B) \
-        CR_1(NV_PGRAPH_SETUPRASTER) \
-        CR_1(NV_PGRAPH_SHADERPROG) \
-        CR_1(NV_PGRAPH_ZCOMPRESSOCCLUDE) \
-        CR_8(NV_PGRAPH_COMBINECOLORI0) \
-        CR_8(NV_PGRAPH_COMBINECOLORO0) \
-        CR_8(NV_PGRAPH_COMBINEALPHAI0) \
-        CR_8(NV_PGRAPH_COMBINEALPHAO0) \
-        CR_8(NV_PGRAPH_COMBINEFACTOR0) \
-        CR_8(NV_PGRAPH_COMBINEFACTOR1) \
-        CR_1(NV_PGRAPH_SHADERCLIPMODE) \
-        CR_4(NV_PGRAPH_TEXCTL0_0) \
-        CR_4(NV_PGRAPH_TEXFMT0) \
-        CR_4(NV_PGRAPH_TEXFILTER0) \
-        CR_8(NV_PGRAPH_WINDOWCLIPX0) \
-        CR_8(NV_PGRAPH_WINDOWCLIPY0) \
-        CF(pg->primitive_mode, primitive_mode) \
-        CF(pg->surface_scale_factor, surface_scale_factor) \
-        CF(pg->compressed_attrs, compressed_attrs) \
-        CFA(pg->texture_matrix_enable, texture_matrix_enable)
+    VshUniformValues vsh_values;
+    pgraph_glsl_set_vsh_uniform_values(pg, &binding->state.vsh,
+                                  binding->uniform_locs.vsh, &vsh_values);
+    apply_uniform_updates(VshUniformInfo, binding->uniform_locs.vsh,
+                          &vsh_values, VshUniform__COUNT);
 
-    #define CR_x(reg, x) CR_x__define(reg, x)
-    #define CF_x(type, src, name, x) CF_x__define(type, src, name, x)
-    DIRTY_REGS
-    #undef CR_x
-    #undef CF_x
+    PshUniformValues psh_values;
+    pgraph_glsl_set_psh_uniform_values(pg, binding->uniform_locs.psh, &psh_values);
 
-    #define CR_x(reg, x) CR_x__check(reg, x)
-    #define CF_x(type, src, name, x) CF_x__check(type, src, name, x)
-    DIRTY_REGS
-    #undef CR_x
-    #undef CF_x
-    return false;
-
-dirty:
-    #define CR_x(reg, x) CR_x__update(reg, x)
-    #define CF_x(type, src, name, x) CF_x__update(type, src, name, x)
-    DIRTY_REGS
-    #undef CR_x
-    #undef CF_x
-    return true;
+    for (int i = 0; i < 4; i++) {
+        if (r->texture_binding[i] != NULL) {
+            float scale = r->texture_binding[i]->scale;
+            psh_values.texScale[i] = scale;
+        }
+    }
+    apply_uniform_updates(PshUniformInfo, binding->uniform_locs.psh,
+                          &psh_values, PshUniform__COUNT);
 }
 
 void pgraph_gl_bind_shaders(PGRAPHState *pg)
@@ -1027,18 +761,17 @@ void pgraph_gl_bind_shaders(PGRAPHState *pg)
     PGRAPHGLState *r = pg->gl_renderer_state;
 
     bool binding_changed = false;
-    if (r->shader_binding && !test_shaders_dirty(pg) && !pg->program_data_dirty) {
+    if (r->shader_binding &&
+        !pgraph_glsl_check_shader_state_dirty(pg, &r->shader_binding->state)) {
         nv2a_profile_inc_counter(NV2A_PROF_SHADER_BIND_NOTDIRTY);
-        goto update_constants;
+        goto update_uniforms;
     }
 
     ShaderBinding *old_binding = r->shader_binding;
-    ShaderState state = pgraph_get_shader_state(pg);
-    assert(!state.vulkan);
+    ShaderState state = pgraph_glsl_get_shader_state(pg);
 
-    NV2A_GL_DGROUP_BEGIN("%s (VP: %s FFP: %s)", __func__,
-                         state.vertex_program ? "yes" : "no",
-                         state.fixed_function ? "yes" : "no");
+    NV2A_GL_DGROUP_BEGIN("%s (%s)", __func__,
+                         state.vsh.is_fixed_function ? "FF" : "PROG");
 
     qemu_mutex_lock(&r->shader_cache_lock);
 
@@ -1050,7 +783,7 @@ void pgraph_gl_bind_shaders(PGRAPHState *pg)
 
     if (!binding->initialized && !pgraph_gl_shader_load_from_memory(binding)) {
         nv2a_profile_inc_counter(NV2A_PROF_SHADER_GEN);
-        generate_shaders(binding);
+        generate_shaders(r, binding);
         if (g_config.perf.cache_shaders) {
             pgraph_gl_shader_cache_to_disk(binding);
         }
@@ -1069,10 +802,10 @@ void pgraph_gl_bind_shaders(PGRAPHState *pg)
 
     NV2A_GL_DGROUP_END();
 
-update_constants:
+update_uniforms:
     assert(r->shader_binding);
     assert(r->shader_binding->initialized);
-    shader_update_constants(pg, r->shader_binding, binding_changed);
+    update_shader_uniforms(pg, r->shader_binding);
 }
 
 GLuint pgraph_gl_compile_shader(const char *vs_src, const char *fs_src)

hw/xbox/nv2a/pgraph/gl/surface.c

@@ -137,11 +137,7 @@ static void init_render_to_texture(PGRAPHState *pg)
         "layout(location = 0) out vec4 out_Color;\n"
         "void main()\n"
         "{\n"
-        "    vec2 texCoord;\n"
-        "    texCoord.x = gl_FragCoord.x;\n"
-        "    texCoord.y = (surface_size.y - gl_FragCoord.y)\n"
-        "                 + (textureSize(tex,0).y - surface_size.y);\n"
-        "    texCoord /= textureSize(tex,0).xy;\n"
+        "    vec2 texCoord = gl_FragCoord.xy / textureSize(tex, 0).xy;\n"
         "    out_Color.rgba = texture(tex, texCoord);\n"
         "}\n";
 
@@ -298,7 +294,7 @@ static void render_surface_to_texture_slow(NV2AState *d,
     size_t bufsize = width * height * surface->fmt.bytes_per_pixel;
 
     uint8_t *buf = g_malloc(bufsize);
-    surface_download_to_buffer(d, surface, false, true, false, buf);
+    surface_download_to_buffer(d, surface, false, false, false, buf);
 
     width = texture_shape->width;
     height = texture_shape->height;
@@ -417,16 +413,52 @@ bool pgraph_gl_check_surface_to_texture_compatibility(
     return false;
 }
 
-static void wait_for_surface_download(SurfaceBinding *e)
+static bool check_surface_overlaps_range(const SurfaceBinding *surface,
+                                         hwaddr range_start, hwaddr range_len)
 {
-    NV2AState *d = g_nv2a;
-    PGRAPHState *pg = &d->pgraph;
-    PGRAPHGLState *r = pg->gl_renderer_state;
+    hwaddr surface_end = surface->vram_addr + surface->size;
+    hwaddr range_end = range_start + range_len;
+    return !(surface->vram_addr >= range_end || range_start >= surface_end);
+}
 
-    if (qatomic_read(&e->draw_dirty)) {
+static void surface_access_callback(void *opaque, MemoryRegion *mr, hwaddr addr,
+                                    hwaddr len, bool write)
+{
+    NV2AState *d = (NV2AState *)opaque;
+    qemu_mutex_lock(&d->pgraph.lock);
+
+    PGRAPHGLState *r = d->pgraph.gl_renderer_state;
+    bool wait_for_downloads = false;
+
+    SurfaceBinding *surface;
+    QTAILQ_FOREACH(surface, &r->surfaces, entry) {
+        if (!check_surface_overlaps_range(surface, addr, len)) {
+            continue;
+        }
+
+        hwaddr offset = addr - surface->vram_addr;
+
+        if (write) {
+            trace_nv2a_pgraph_surface_cpu_write(surface->vram_addr, offset);
+        } else {
+            trace_nv2a_pgraph_surface_cpu_read(surface->vram_addr, offset);
+        }
+
+        if (surface->draw_dirty) {
+            surface->download_pending = true;
+            wait_for_downloads = true;
+        }
+
+        if (write) {
+            surface->upload_pending = true;
+        }
+    }
+
+    qemu_mutex_unlock(&d->pgraph.lock);
+
+    if (wait_for_downloads) {
         qemu_mutex_lock(&d->pfifo.lock);
         qemu_event_reset(&r->downloads_complete);
-        qatomic_set(&e->download_pending, true);
         qatomic_set(&r->downloads_pending, true);
         pfifo_kick(d);
         qemu_mutex_unlock(&d->pfifo.lock);
@@ -434,22 +466,44 @@ static void wait_for_surface_download(SurfaceBinding *e)
     }
 }
 
-static void surface_access_callback(void *opaque, MemoryRegion *mr, hwaddr addr,
-                                    hwaddr len, bool write)
+static void register_cpu_access_callback(NV2AState *d, SurfaceBinding *surface)
 {
-    SurfaceBinding *e = opaque;
-    assert(addr >= e->vram_addr);
-    hwaddr offset = addr - e->vram_addr;
-    assert(offset < e->size);
-
-    if (qatomic_read(&e->draw_dirty)) {
-        trace_nv2a_pgraph_surface_cpu_access(e->vram_addr, offset);
-        wait_for_surface_download(e);
+    if (tcg_enabled()) {
+        surface->access_cb = mem_access_callback_insert(
+            qemu_get_cpu(0), d->vram, surface->vram_addr, surface->size,
+            &surface_access_callback, d);
     }
+}
 
-    if (write && !qatomic_read(&e->upload_pending)) {
-        trace_nv2a_pgraph_surface_cpu_access(e->vram_addr, offset);
-        qatomic_set(&e->upload_pending, true);
+static void unregister_cpu_access_callback(NV2AState *d,
+                                           SurfaceBinding const *surface)
+{
+    if (tcg_enabled()) {
+        mem_access_callback_remove_by_ref(qemu_get_cpu(0), surface->access_cb);
+    }
+}
+
+static bool check_surfaces_overlap(const SurfaceBinding *surface,
+                                   const SurfaceBinding *other_surface)
+{
+    return check_surface_overlaps_range(surface, other_surface->vram_addr,
+                                        other_surface->size);
+}
+
+static void invalidate_overlapping_surfaces(NV2AState *d, SurfaceBinding *surface)
+{
+    PGRAPHState *pg = &d->pgraph;
+    PGRAPHGLState *r = pg->gl_renderer_state;
+
+    SurfaceBinding *other_surface, *next_surface;
+    QTAILQ_FOREACH_SAFE(other_surface, &r->surfaces, entry, next_surface) {
+        if (check_surfaces_overlap(surface, other_surface)) {
+            trace_nv2a_pgraph_surface_evict_overlapping(
+                other_surface->vram_addr, other_surface->width, other_surface->height,
+                other_surface->pitch);
+            pgraph_gl_surface_download_if_dirty(d, other_surface);
+            pgraph_gl_surface_invalidate(d, other_surface);
+        }
     }
 }
 
@@ -461,35 +515,13 @@ static SurfaceBinding *surface_put(NV2AState *d, hwaddr addr,
 
     assert(pgraph_gl_surface_get(d, addr) == NULL);
 
-    SurfaceBinding *surface, *next;
-    uintptr_t e_end = surface_in->vram_addr + surface_in->size - 1;
-    QTAILQ_FOREACH_SAFE(surface, &r->surfaces, entry, next) {
-        uintptr_t s_end = surface->vram_addr + surface->size - 1;
-        bool overlapping = !(surface->vram_addr > e_end
-                             || surface_in->vram_addr > s_end);
-        if (overlapping) {
-            trace_nv2a_pgraph_surface_evict_overlapping(
-                surface->vram_addr, surface->width, surface->height,
-                surface->pitch);
-            pgraph_gl_surface_download_if_dirty(d, surface);
-            pgraph_gl_surface_invalidate(d, surface);
-        }
-    }
+    invalidate_overlapping_surfaces(d, surface_in);
 
     SurfaceBinding *surface_out = g_malloc(sizeof(SurfaceBinding));
     assert(surface_out != NULL);
     *surface_out = *surface_in;
 
-    if (tcg_enabled()) {
-        qemu_mutex_unlock(&d->pgraph.lock);
-        bql_lock();
-        mem_access_callback_insert(qemu_get_cpu(0),
-            d->vram, surface_out->vram_addr, surface_out->size,
-            &surface_out->access_cb, &surface_access_callback,
-            surface_out);
-        bql_unlock();
-        qemu_mutex_lock(&d->pgraph.lock);
-    }
+    register_cpu_access_callback(d, surface_out);
 
     QTAILQ_INSERT_TAIL(&r->surfaces, surface_out, entry);
 
@@ -543,13 +575,7 @@ void pgraph_gl_surface_invalidate(NV2AState *d, SurfaceBinding *surface)
         pgraph_gl_unbind_surface(d, false);
     }
 
-    if (tcg_enabled()) {
-        qemu_mutex_unlock(&d->pgraph.lock);
-        bql_lock();
-        mem_access_callback_remove_by_ref(qemu_get_cpu(0), surface->access_cb);
-        bql_unlock();
-        qemu_mutex_lock(&d->pgraph.lock);
-    }
+    unregister_cpu_access_callback(d, surface);
 
     glDeleteTextures(1, &surface->gl_buffer);
 
@@ -738,7 +764,7 @@ static void surface_download(NV2AState *d, SurfaceBinding *surface, bool force)
 
     nv2a_profile_inc_counter(NV2A_PROF_SURF_DOWNLOAD);
 
-    surface_download_to_buffer(d, surface, true, true, true,
+    surface_download_to_buffer(d, surface, true, false, true,
                                d->vram_ptr + surface->vram_addr);
 
     memory_region_set_client_dirty(d->vram, surface->vram_addr,
@@ -875,20 +901,26 @@ void pgraph_gl_upload_surface_data(NV2AState *d, SurfaceBinding *surface,
                        surface->fmt.bytes_per_pixel);
     }
 
-    /* FIXME: Replace this flip/scaling */
+    /* FIXME: Replace this scaling */
 
     // This is VRAM so we can't do this inplace!
-    uint8_t *flipped_buf = (uint8_t *)g_malloc(
-        surface->height * surface->width * surface->fmt.bytes_per_pixel);
-    unsigned int irow;
-    for (irow = 0; irow < surface->height; irow++) {
-        memcpy(&flipped_buf[surface->width * (surface->height - irow - 1)
-                                 * surface->fmt.bytes_per_pixel],
-               &buf[surface->pitch * irow],
-               surface->width * surface->fmt.bytes_per_pixel);
+    uint8_t *optimal_buf = buf;
+    unsigned int optimal_pitch = surface->width * surface->fmt.bytes_per_pixel;
+
+    if (surface->pitch != optimal_pitch) {
+        optimal_buf = (uint8_t *)g_malloc(surface->height * optimal_pitch);
+
+        uint8_t *src = buf;
+        uint8_t *dst = optimal_buf;
+        unsigned int irow;
+        for (irow = 0; irow < surface->height; irow++) {
+            memcpy(dst, src, optimal_pitch);
+            src += surface->pitch;
+            dst += optimal_pitch;
+        }
     }
 
-    uint8_t *gl_read_buf = flipped_buf;
+    uint8_t *gl_read_buf = optimal_buf;
     unsigned int width = surface->width, height = surface->height;
 
     if (pg->surface_scale_factor > 1) {
@@ -896,7 +928,7 @@ void pgraph_gl_upload_surface_data(NV2AState *d, SurfaceBinding *surface,
         pg->scale_buf = (uint8_t *)g_realloc(
             pg->scale_buf, width * height * surface->fmt.bytes_per_pixel);
         gl_read_buf = pg->scale_buf;
-        uint8_t *out = gl_read_buf, *in = flipped_buf;
+        uint8_t *out = gl_read_buf, *in = optimal_buf;
         surface_copy_expand(out, in, surface->width, surface->height,
                             surface->fmt.bytes_per_pixel,
                             d->pgraph.surface_scale_factor);
@@ -915,7 +947,9 @@ void pgraph_gl_upload_surface_data(NV2AState *d, SurfaceBinding *surface,
                  height, 0, surface->fmt.gl_format, surface->fmt.gl_type,
                  gl_read_buf);
     glPixelStorei(GL_UNPACK_ALIGNMENT, prev_unpack_alignment);
-    g_free(flipped_buf);
+    if (optimal_buf != buf) {
+        g_free(optimal_buf);
+    }
     if (surface->swizzle) {
         g_free(buf);
     }

hw/xbox/nv2a/pgraph/gl/texture.c

@@ -746,7 +746,7 @@ static void texture_binding_destroy(gpointer data)
 }
 
 /* functions for texture LRU cache */
-static void texture_cache_entry_init(Lru *lru, LruNode *node, void *key)
+static void texture_cache_entry_init(Lru *lru, LruNode *node, const void *key)
 {
     TextureLruNode *tnode = container_of(node, TextureLruNode, node);
     memcpy(&tnode->key, key, sizeof(TextureKey));
@@ -765,7 +765,8 @@ static void texture_cache_entry_post_evict(Lru *lru, LruNode *node)
     }
 }
 
-static bool texture_cache_entry_compare(Lru *lru, LruNode *node, void *key)
+static bool texture_cache_entry_compare(Lru *lru, LruNode *node,
+                                        const void *key)
 {
     TextureLruNode *tnode = container_of(node, TextureLruNode, node);
     return memcmp(&tnode->key, key, sizeof(TextureKey));

hw/xbox/nv2a/pgraph/gl/vertex.c

@@ -232,14 +232,14 @@ unsigned int pgraph_gl_bind_inline_array(NV2AState *d)
     return index_count;
 }
 
-static void vertex_cache_entry_init(Lru *lru, LruNode *node, void *key)
+static void vertex_cache_entry_init(Lru *lru, LruNode *node, const void *key)
 {
     VertexLruNode *vnode = container_of(node, VertexLruNode, node);
     memcpy(&vnode->key, key, sizeof(struct VertexKey));
     vnode->initialized = false;
 }
 
-static bool vertex_cache_entry_compare(Lru *lru, LruNode *node, void *key)
+static bool vertex_cache_entry_compare(Lru *lru, LruNode *node, const void *key)
 {
     VertexLruNode *vnode = container_of(node, VertexLruNode, node);
     return memcmp(&vnode->key, key, sizeof(VertexKey));

hw/xbox/nv2a/pgraph/glsl/common.c

@@ -1,7 +1,7 @@
 /*
  * Geforce NV2A PGRAPH GLSL Shader Generator
  *
- * Copyright (c) 2024 Matt Borgerson
+ * Copyright (c) 2024-2025 Matt Borgerson
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
@@ -18,8 +18,15 @@
  */
 
 #include "common.h"
+#include "hw/xbox/nv2a/pgraph/pgraph.h"
 
-MString *pgraph_get_glsl_vtx_header(MString *out, bool location, bool smooth, bool in, bool prefix, bool array)
+#define DECL_UNIFORM_ELEMENT_NAME(type) #type,
+const char *uniform_element_type_to_str[] = {
+    UNIFORM_ELEMENT_TYPE_X(DECL_UNIFORM_ELEMENT_NAME)
+};
+
+MString *pgraph_glsl_get_vtx_header(MString *out, bool location, bool smooth,
+                                    bool in, bool prefix, bool array)
 {
     const char *smooth_s = "";
     const char *flat_s = "flat ";
@@ -54,3 +61,26 @@ MString *pgraph_get_glsl_vtx_header(MString *out, bool location, bool smooth, bo
 
     return out;
 }
+
+void pgraph_glsl_set_clip_range_uniform_value(PGRAPHState *pg, float clipRange[4])
+{
+    float zmax;
+    switch (pg->surface_shape.zeta_format) {
+    case NV097_SET_SURFACE_FORMAT_ZETA_Z16:
+        zmax = pg->surface_shape.z_format ? f16_max : (float)0xFFFF;
+        break;
+    case NV097_SET_SURFACE_FORMAT_ZETA_Z24S8:
+        zmax = pg->surface_shape.z_format ? f24_max : (float)0xFFFFFF;
+        break;
+    default:
+        assert(0);
+    }
+
+    uint32_t zclip_min = pgraph_reg_r(pg, NV_PGRAPH_ZCLIPMIN);
+    uint32_t zclip_max = pgraph_reg_r(pg, NV_PGRAPH_ZCLIPMAX);
+
+    clipRange[0] = 0;
+    clipRange[1] = zmax;
+    clipRange[2] = *(float *)&zclip_min;
+    clipRange[3] = *(float *)&zclip_max;
+}

hw/xbox/nv2a/pgraph/glsl/common.h

@@ -3,6 +3,7 @@
  *
  * Copyright (c) 2015 espes
  * Copyright (c) 2015 Jannik Vogel
+ * Copyright (c) 2025 Matt Borgerson
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
@@ -21,8 +22,69 @@
 #ifndef HW_NV2A_SHADERS_COMMON_H
 #define HW_NV2A_SHADERS_COMMON_H
 
+#include "qemu/osdep.h"
 #include "qemu/mstring.h"
-#include <stdbool.h>
+
+typedef int ivec4[4];
+typedef float mat2[2 * 2];
+typedef unsigned int uint;
+typedef float vec2[2];
+typedef float vec3[3];
+typedef float vec4[4];
+
+#define UNIFORM_ELEMENT_TYPE_X(DECL) \
+    DECL(float)                      \
+    DECL(int)                        \
+    DECL(ivec4)                      \
+    DECL(mat2)                       \
+    DECL(uint)                       \
+    DECL(vec2)                       \
+    DECL(vec3)                       \
+    DECL(vec4)
+
+enum UniformElementType {
+#define DECL_UNIFORM_ELEMENT_TYPE(type) UniformElementType_##type,
+    UNIFORM_ELEMENT_TYPE_X(DECL_UNIFORM_ELEMENT_TYPE)
+};
+
+extern const char *uniform_element_type_to_str[];
+
+#define DECL_UNIFORM_ENUM_VALUE(s, name, type, count) s##_##name,
+#define DECL_UNIFORM_ENUM_TYPE(name, decls)                 \
+    enum name##Indices{                                     \
+        decls(name, DECL_UNIFORM_ENUM_VALUE) name##__COUNT, \
+    };
+
+#define DECL_UNIFORM_LOC_STRUCT_TYPE(name, decls) \
+    typedef int name##Locs[name##__COUNT];
+
+#define DECL_UNIFORM_VAL_STRUCT_FIELD(s, name, type, count) type name[count];
+#define DECL_UNIFORM_VAL_STRUCT_TYPE(name, decls)  \
+    typedef struct name##Values {                  \
+        decls(name, DECL_UNIFORM_VAL_STRUCT_FIELD) \
+    } name##Values;
+
+typedef struct UniformInfo {
+    const char *name;
+    enum UniformElementType type;
+    size_t size;
+    size_t count;
+    size_t val_offs;
+} UniformInfo;
+
+#define DECL_UNIFORM_INFO_ITEM(s, name, type, count)         \
+    { #name, UniformElementType_##type, sizeof(type), count, \
+      offsetof(s##Values, name) },
+#define DECL_UNIFORM_INFO_ARR(name, decls) \
+    extern const UniformInfo name##Info[];
+#define DEF_UNIFORM_INFO_ARR(name, decls) \
+    const UniformInfo name##Info[] = { decls(name, DECL_UNIFORM_INFO_ITEM) };
+
+#define DECL_UNIFORM_TYPES(name, decls)       \
+    DECL_UNIFORM_ENUM_TYPE(name, decls)       \
+    DECL_UNIFORM_LOC_STRUCT_TYPE(name, decls) \
+    DECL_UNIFORM_VAL_STRUCT_TYPE(name, decls) \
+    DECL_UNIFORM_INFO_ARR(name, decls)
 
 #define GLSL_C(idx) "c[" stringify(idx) "]"
 #define GLSL_LTCTXA(idx) "ltctxa[" stringify(idx) "]"
@@ -33,6 +95,12 @@
 
 #define GLSL_DEFINE(a, b) "#define " stringify(a) " " b "\n"
 
-MString *pgraph_get_glsl_vtx_header(MString *out, bool location, bool smooth, bool in, bool prefix, bool array);
+MString *pgraph_glsl_get_vtx_header(MString *out, bool location, bool smooth,
+                                    bool in, bool prefix, bool array);
+
+typedef struct PGRAPHState PGRAPHState;
+
+void pgraph_glsl_set_clip_range_uniform_value(PGRAPHState *pg,
+                                              float clipRange[4]);
 
 #endif

hw/xbox/nv2a/pgraph/glsl/geom.c

@@ -3,7 +3,7 @@
  *
  * Copyright (c) 2015 espes
  * Copyright (c) 2015 Jannik Vogel
- * Copyright (c) 2020-2024 Matt Borgerson
+ * Copyright (c) 2020-2025 Matt Borgerson
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
@@ -19,19 +19,99 @@
  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  */
 
-#include "hw/xbox/nv2a/pgraph/shaders.h"
-#include "common.h"
+#include "qemu/osdep.h"
+#include "hw/xbox/nv2a/pgraph/pgraph.h"
 #include "geom.h"
 
-MString *pgraph_gen_geom_glsl(enum ShaderPolygonMode polygon_front_mode,
-                              enum ShaderPolygonMode polygon_back_mode,
-                              enum ShaderPrimitiveMode primitive_mode,
-                              bool smooth_shading,
-                              bool vulkan)
+void pgraph_glsl_set_geom_state(PGRAPHState *pg, GeomState *state)
+{
+    state->primitive_mode = (enum ShaderPrimitiveMode)pg->primitive_mode;
+
+    state->polygon_front_mode = (enum ShaderPolygonMode)GET_MASK(
+        pgraph_reg_r(pg, NV_PGRAPH_SETUPRASTER),
+        NV_PGRAPH_SETUPRASTER_FRONTFACEMODE);
+    state->polygon_back_mode = (enum ShaderPolygonMode)GET_MASK(
+        pgraph_reg_r(pg, NV_PGRAPH_SETUPRASTER),
+        NV_PGRAPH_SETUPRASTER_BACKFACEMODE);
+
+    state->smooth_shading = GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CONTROL_3),
+                                     NV_PGRAPH_CONTROL_3_SHADEMODE) ==
+                            NV_PGRAPH_CONTROL_3_SHADEMODE_SMOOTH;
+}
+
+bool pgraph_glsl_need_geom(const GeomState *state)
 {
     /* FIXME: Missing support for 2-sided-poly mode */
-    assert(polygon_front_mode == polygon_back_mode);
-    enum ShaderPolygonMode polygon_mode = polygon_front_mode;
+    assert(state->polygon_front_mode == state->polygon_back_mode);
+    enum ShaderPolygonMode polygon_mode = state->polygon_front_mode;
+
+    /* POINT mode shouldn't require any special work */
+    if (polygon_mode == POLY_MODE_POINT) {
+        return false;
+    }
+
+    switch (state->primitive_mode) {
+    case PRIM_TYPE_TRIANGLES:
+        if (polygon_mode == POLY_MODE_FILL) {
+            return false;
+        }
+        return true;
+    case PRIM_TYPE_TRIANGLE_STRIP:
+        if (polygon_mode == POLY_MODE_FILL) {
+            return false;
+        }
+        assert(polygon_mode == POLY_MODE_LINE);
+        return true;
+    case PRIM_TYPE_TRIANGLE_FAN:
+        if (polygon_mode == POLY_MODE_FILL) {
+            return false;
+        }
+        assert(polygon_mode == POLY_MODE_LINE);
+        return true;
+    case PRIM_TYPE_QUADS:
+        if (polygon_mode == POLY_MODE_LINE) {
+            return true;
+        } else if (polygon_mode == POLY_MODE_FILL) {
+            return true;
+        } else {
+            assert(false);
+            return false;
+        }
+        break;
+    case PRIM_TYPE_QUAD_STRIP:
+        if (polygon_mode == POLY_MODE_LINE) {
+            return true;
+        } else if (polygon_mode == POLY_MODE_FILL) {
+            return true;
+        } else {
+            assert(false);
+            return false;
+        }
+        break;
+    case PRIM_TYPE_POLYGON:
+        if (polygon_mode == POLY_MODE_LINE) {
+            return false;
+        }
+        if (polygon_mode == POLY_MODE_FILL) {
+            if (state->smooth_shading) {
+                return false;
+            }
+            return true;
+        } else {
+            assert(false);
+            return false;
+        }
+        break;
+    default:
+        return false;
+    }
+}
+
+MString *pgraph_glsl_gen_geom(const GeomState *state, GenGeomGlslOptions opts)
+{
+    /* FIXME: Missing support for 2-sided-poly mode */
+    assert(state->polygon_front_mode == state->polygon_back_mode);
+    enum ShaderPolygonMode polygon_mode = state->polygon_front_mode;
 
     /* POINT mode shouldn't require any special work */
     if (polygon_mode == POLY_MODE_POINT) {
@@ -42,7 +122,7 @@ MString *pgraph_gen_geom_glsl(enum ShaderPolygonMode polygon_front_mode,
     const char *layout_in = NULL;
     const char *layout_out = NULL;
     const char *body = NULL;
-    switch (primitive_mode) {
+    switch (state->primitive_mode) {
     case PRIM_TYPE_POINTS: return NULL;
     case PRIM_TYPE_LINES: return NULL;
     case PRIM_TYPE_LINE_LOOP: return NULL;
@@ -145,7 +225,7 @@ MString *pgraph_gen_geom_glsl(enum ShaderPolygonMode polygon_front_mode,
             return NULL;
         }
         if (polygon_mode == POLY_MODE_FILL) {
-            if (smooth_shading) {
+            if (state->smooth_shading) {
                 return NULL;
             }
             layout_in = "layout(triangles) in;\n";
@@ -169,16 +249,19 @@ MString *pgraph_gen_geom_glsl(enum ShaderPolygonMode polygon_front_mode,
     assert(layout_in);
     assert(layout_out);
     assert(body);
-    MString *s = mstring_new();
-    mstring_append_fmt(s, "#version %d\n\n", vulkan ? 450 : 400);
-    mstring_append(s, layout_in);
-    mstring_append(s, layout_out);
-    mstring_append(s, "\n");
-    pgraph_get_glsl_vtx_header(s, vulkan, smooth_shading, true, true, true);
-    pgraph_get_glsl_vtx_header(s, vulkan, smooth_shading, false, false, false);
+    MString *output =
+        mstring_from_fmt("#version %d\n\n"
+                         "%s"
+                         "%s"
+                         "\n",
+                         opts.vulkan ? 450 : 400, layout_in, layout_out);
+    pgraph_glsl_get_vtx_header(output, opts.vulkan, state->smooth_shading, true,
+                               true, true);
+    pgraph_glsl_get_vtx_header(output, opts.vulkan, state->smooth_shading,
+                               false, false, false);
 
-    if (smooth_shading) {
-        mstring_append(s,
+    if (state->smooth_shading) {
+        mstring_append(output,
                        "void emit_vertex(int index, int _unused) {\n"
                        "  gl_Position = gl_in[index].gl_Position;\n"
                        "  gl_PointSize = gl_in[index].gl_PointSize;\n"
@@ -194,7 +277,7 @@ MString *pgraph_gen_geom_glsl(enum ShaderPolygonMode polygon_front_mode,
                        "  EmitVertex();\n"
                        "}\n");
     } else {
-        mstring_append(s,
+        mstring_append(output,
                        "void emit_vertex(int index, int provoking_index) {\n"
                        "  gl_Position = gl_in[index].gl_Position;\n"
                        "  gl_PointSize = gl_in[index].gl_PointSize;\n"
@@ -211,10 +294,12 @@ MString *pgraph_gen_geom_glsl(enum ShaderPolygonMode polygon_front_mode,
                        "}\n");
     }
 
-    mstring_append(s, "\n"
-                      "void main() {\n");
-    mstring_append(s, body);
-    mstring_append(s, "}\n");
+    mstring_append_fmt(output,
+                       "\n"
+                       "void main() {\n"
+                       "%s"
+                       "}\n",
+                       body);
 
-    return s;
+    return output;
 }

hw/xbox/nv2a/pgraph/glsl/geom.h

@@ -3,7 +3,7 @@
  *
  * Copyright (c) 2015 espes
  * Copyright (c) 2015 Jannik Vogel
- * Copyright (c) 2020-2024 Matt Borgerson
+ * Copyright (c) 2020-2025 Matt Borgerson
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
@@ -22,13 +22,24 @@
 #ifndef HW_XBOX_NV2A_PGRAPH_GLSL_GEOM_H
 #define HW_XBOX_NV2A_PGRAPH_GLSL_GEOM_H
 
-#include "qemu/mstring.h"
-#include "hw/xbox/nv2a/pgraph/shaders.h"
+#include "common.h"
+#include "hw/xbox/nv2a/pgraph/vsh_regs.h"
 
-MString *pgraph_gen_geom_glsl(enum ShaderPolygonMode polygon_front_mode,
-                              enum ShaderPolygonMode polygon_back_mode,
-                              enum ShaderPrimitiveMode primitive_mode,
-                              bool smooth_shading,
-                              bool vulkan);
+typedef struct {
+    enum ShaderPrimitiveMode primitive_mode;
+    enum ShaderPolygonMode polygon_front_mode;
+    enum ShaderPolygonMode polygon_back_mode;
+    bool smooth_shading;
+} GeomState;
+
+typedef struct GenGeomGlslOptions {
+    bool vulkan;
+} GenGeomGlslOptions;
+
+void pgraph_glsl_set_geom_state(PGRAPHState *pg, GeomState *geom);
+
+bool pgraph_glsl_need_geom(const GeomState *state);
+
+MString *pgraph_glsl_gen_geom(const GeomState *state, GenGeomGlslOptions opts);
 
 #endif

hw/xbox/nv2a/pgraph/glsl/meson.build

@@ -2,6 +2,7 @@ specific_ss.add([files(
 	'common.c',
 	'geom.c',
 	'psh.c',
+	'shaders.c',
 	'vsh.c',
 	'vsh-ff.c',
 	'vsh-prog.c',

hw/xbox/nv2a/pgraph/glsl/psh.h

@@ -3,39 +3,94 @@
  *
  * Copyright (c) 2013 espes
  * Copyright (c) 2015 Jannik Vogel
- * Copyright (c) 2020-2024 Matt Borgerson
+ * Copyright (c) 2020-2025 Matt Borgerson
  *
- * Based on:
- * Cxbx, PixelShader.cpp
- * Copyright (c) 2004 Aaron Robinson <caustik@caustik.com>
- *                    Kingofc <kingofc@freenet.de>
- * Xeon, XBD3DPixelShader.cpp
- * Copyright (c) 2003 _SF_
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
  *
- * This program is free software; you can redistribute it and/or
- * modify it under the terms of the GNU General Public License as
- * published by the Free Software Foundation; either version 2 or
- * (at your option) version 3 of the License.
- *
- * This program is distributed in the hope that it will be useful,
+ * This library is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * GNU General Public License for more details.
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
  *
- * You should have received a copy of the GNU General Public License
- * along with this program; if not, see <http://www.gnu.org/licenses/>.
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  */
 
 #ifndef HW_XBOX_NV2A_PGRAPH_GLSL_PSH_H
 #define HW_XBOX_NV2A_PGRAPH_GLSL_PSH_H
 
-#include "qemu/mstring.h"
-#include "hw/xbox/nv2a/pgraph/shaders.h"
+#include "common.h"
+#include "hw/xbox/nv2a/pgraph/psh_regs.h"
 
-// FIXME: Move to struct
-#define PSH_UBO_BINDING 1
-#define PSH_TEX_BINDING 2
+typedef struct PGRAPHState PGRAPHState;
 
-MString *pgraph_gen_psh_glsl(const PshState state);
+typedef struct PshState {
+    uint32_t combiner_control;
+    uint32_t shader_stage_program;
+    uint32_t other_stage_input;
+    uint32_t final_inputs_0;
+    uint32_t final_inputs_1;
+
+    uint32_t rgb_inputs[8], rgb_outputs[8];
+    uint32_t alpha_inputs[8], alpha_outputs[8];
+
+    bool point_sprite;
+    bool rect_tex[4];
+    bool snorm_tex[4];
+    bool compare_mode[4][4];
+    bool alphakill[4];
+    int colorkey_mode[4];
+    enum ConvolutionFilter conv_tex[4];
+    bool tex_x8y24[4];
+    int dim_tex[4];
+
+    float border_logical_size[4][3];
+    float border_inv_real_size[4][3];
+
+    bool shadow_map[4];
+    enum PshShadowDepthFunc shadow_depth_func;
+
+    bool alpha_test;
+    enum PshAlphaFunc alpha_func;
+
+    bool window_clip_exclusive;
+
+    bool smooth_shading;
+    bool depth_clipping;
+    bool z_perspective;
+} PshState;
+
+void pgraph_glsl_set_psh_state(PGRAPHState *pg, PshState *state);
+
+#define PSH_UNIFORM_DECL_X(S, DECL) \
+    DECL(S, alphaRef, int, 1)       \
+    DECL(S, bumpMat, mat2, 4)       \
+    DECL(S, bumpOffset, float, 4)   \
+    DECL(S, bumpScale, float, 4)    \
+    DECL(S, clipRange, vec4, 1)     \
+    DECL(S, clipRegion, ivec4, 8)   \
+    DECL(S, colorKey, uint, 4)      \
+    DECL(S, colorKeyMask, uint, 4)  \
+    DECL(S, consts, vec4, 18)       \
+    DECL(S, depthOffset, float, 1)  \
+    DECL(S, fogColor, vec4, 1)      \
+    DECL(S, texScale, float, 4)
+
+DECL_UNIFORM_TYPES(PshUniform, PSH_UNIFORM_DECL_X)
+
+typedef struct GenPshGlslOptions {
+    bool vulkan;
+    int ubo_binding;
+    int tex_binding;
+} GenPshGlslOptions;
+
+MString *pgraph_glsl_gen_psh(const PshState *state, GenPshGlslOptions opts);
+
+void pgraph_glsl_set_psh_uniform_values(PGRAPHState *pg,
+                                        const PshUniformLocs locs,
+                                        PshUniformValues *values);
 
 #endif

hw/xbox/nv2a/pgraph/glsl/shaders.c

@@ -0,0 +1,94 @@
+/*
+ * Geforce NV2A PGRAPH GLSL Shader Generator
+ *
+ * Copyright (c) 2025 Matt Borgerson
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "hw/xbox/nv2a/pgraph/pgraph.h"
+#include "shaders.h"
+
+ShaderState pgraph_glsl_get_shader_state(PGRAPHState *pg)
+{
+    pg->program_data_dirty = false; /* fixme */
+
+    ShaderState state;
+
+    // We will hash it, so make sure any padding is zeroed
+    memset(&state, 0, sizeof(ShaderState));
+
+    pgraph_glsl_set_vsh_state(pg, &state.vsh);
+    pgraph_glsl_set_geom_state(pg, &state.geom);
+    pgraph_glsl_set_psh_state(pg, &state.psh);
+
+    return state;
+}
+
+bool pgraph_glsl_check_shader_state_dirty(PGRAPHState *pg,
+                                          const ShaderState *state)
+{
+    if (pg->program_data_dirty) {
+        return true;
+    }
+
+    unsigned int regs[] = {
+        NV_PGRAPH_COMBINECTL,      NV_PGRAPH_COMBINESPECFOG0,
+        NV_PGRAPH_COMBINESPECFOG1, NV_PGRAPH_CONTROL_0,
+        NV_PGRAPH_CONTROL_3,       NV_PGRAPH_CSV0_C,
+        NV_PGRAPH_CSV0_D,          NV_PGRAPH_CSV1_A,
+        NV_PGRAPH_CSV1_B,          NV_PGRAPH_POINTSIZE,
+        NV_PGRAPH_SETUPRASTER,     NV_PGRAPH_SHADERCLIPMODE,
+        NV_PGRAPH_SHADERCTL,       NV_PGRAPH_SHADERPROG,
+        NV_PGRAPH_SHADOWCTL,       NV_PGRAPH_ZCOMPRESSOCCLUDE,
+    };
+    for (int i = 0; i < ARRAY_SIZE(regs); i++) {
+        if (pgraph_is_reg_dirty(pg, regs[i])) {
+            return true;
+        }
+    }
+
+    int num_stages = pgraph_reg_r(pg, NV_PGRAPH_COMBINECTL) & 0xFF;
+    for (int i = 0; i < num_stages; i++) {
+        if (pgraph_is_reg_dirty(pg, NV_PGRAPH_COMBINEALPHAI0 + i * 4) ||
+            pgraph_is_reg_dirty(pg, NV_PGRAPH_COMBINEALPHAO0 + i * 4) ||
+            pgraph_is_reg_dirty(pg, NV_PGRAPH_COMBINECOLORI0 + i * 4) ||
+            pgraph_is_reg_dirty(pg, NV_PGRAPH_COMBINECOLORO0 + i * 4)) {
+            return true;
+        }
+    }
+
+    if (pg->uniform_attrs != state->vsh.uniform_attrs ||
+        pg->swizzle_attrs != state->vsh.swizzle_attrs ||
+        pg->compressed_attrs != state->vsh.compressed_attrs ||
+        pg->primitive_mode != state->geom.primitive_mode ||
+        pg->surface_scale_factor != state->vsh.surface_scale_factor) {
+        return true;
+    }
+
+    for (int i = 0; i < 4; i++) {
+        if (pgraph_is_reg_dirty(pg, NV_PGRAPH_TEXCTL0_0 + i * 4) ||
+            pgraph_is_reg_dirty(pg, NV_PGRAPH_TEXFILTER0 + i * 4) ||
+            pgraph_is_reg_dirty(pg, NV_PGRAPH_TEXFMT0 + i * 4)) {
+            return true;
+        }
+
+        if (pg->texture_matrix_enable[i] !=
+            state->vsh.fixed_function.texture_matrix_enable[i]) {
+            return true;
+        }
+    }
+
+    return false;
+}

hw/xbox/nv2a/pgraph/glsl/shaders.h

@@ -0,0 +1,40 @@
+/*
+ * Geforce NV2A PGRAPH GLSL Shader Generator
+ *
+ * Copyright (c) 2025 Matt Borgerson
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef HW_XBOX_NV2A_PGRAPH_GLSL_SHADERS_H
+#define HW_XBOX_NV2A_PGRAPH_GLSL_SHADERS_H
+
+#include "vsh.h"
+#include "geom.h"
+#include "psh.h"
+
+typedef struct ShaderState {
+    VshState vsh;
+    GeomState geom;
+    PshState psh;
+} ShaderState;
+
+typedef struct PGRAPHState PGRAPHState;
+
+ShaderState pgraph_glsl_get_shader_state(PGRAPHState *pg);
+
+bool pgraph_glsl_check_shader_state_dirty(PGRAPHState *pg,
+                                          const ShaderState *state);
+
+#endif

hw/xbox/nv2a/pgraph/glsl/vsh-ff.c

@@ -3,7 +3,7 @@
  *
  * Copyright (c) 2015 espes
  * Copyright (c) 2015 Jannik Vogel
- * Copyright (c) 2020-2024 Matt Borgerson
+ * Copyright (c) 2020-2025 Matt Borgerson
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
@@ -20,22 +20,55 @@
  */
 
 #include "qemu/osdep.h"
-#include "hw/xbox/nv2a/pgraph/shaders.h"
 #include "common.h"
 #include "vsh-ff.h"
 
-static void append_skinning_code(MString* str, bool mix,
-                                 unsigned int count, const char* type,
-                                 const char* output, const char* input,
-                                 const char* matrix, const char* swizzle);
+static void append_skinning_code(MString *str, bool mix, unsigned int count,
+                                 const char *type, const char *output,
+                                 const char *input, const char *matrix,
+                                 const char *swizzle)
+{
+    if (count == 0) {
+        mstring_append_fmt(str, "%s %s = (%s * %s0).%s;\n",
+                           type, output, input, matrix, swizzle);
+    } else {
+        mstring_append_fmt(str, "%s %s = %s(0.0);\n", type, output, type);
+        if (mix) {
+            /* Generated final weight (like GL_WEIGHT_SUM_UNITY_ARB) */
+            mstring_append(str, "{\n"
+                                "  float weight_i;\n"
+                                "  float weight_n = 1.0;\n");
+            int i;
+            for (i = 0; i < count; i++) {
+                if (i < (count - 1)) {
+                    char c = "xyzw"[i];
+                    mstring_append_fmt(str, "  weight_i = weight.%c;\n"
+                                            "  weight_n -= weight_i;\n",
+                                       c);
+                } else {
+                    mstring_append(str, "  weight_i = weight_n;\n");
+                }
+                mstring_append_fmt(str, "  %s += (%s * %s%d).%s * weight_i;\n",
+                                   output, input, matrix, i, swizzle);
+            }
+            mstring_append(str, "}\n");
+        } else {
+            /* Individual weights */
+            int i;
+            for (i = 0; i < count; i++) {
+                char c = "xyzw"[i];
+                mstring_append_fmt(str, "%s += (%s * %s%d).%s * weight.%c;\n",
+                                   output, input, matrix, i, swizzle, c);
+            }
+        }
+    }
+}
 
-void pgraph_gen_vsh_ff_glsl(const ShaderState *state, MString *header,
-                             MString *body, MString *uniforms)
+void pgraph_glsl_gen_vsh_ff(const VshState *state, MString *header,
+                            MString *body)
 {
     int i, j;
-    const char *u = state->vulkan ? "" : "uniform "; // FIXME: Remove
 
-    /* generate vertex shader mimicking fixed function */
     mstring_append(header,
 "#define position      v0\n"
 "#define weight        v1\n"
@@ -54,11 +87,6 @@ void pgraph_gen_vsh_ff_glsl(const ShaderState *state, MString *header,
 "#define reserved2     v14\n"
 "#define reserved3     v15\n"
 "\n");
-    mstring_append_fmt(uniforms,
-"%svec4 ltctxa[" stringify(NV2A_LTCTXA_COUNT) "];\n"
-"%svec4 ltctxb[" stringify(NV2A_LTCTXB_COUNT) "];\n"
-"%svec4 ltc1[" stringify(NV2A_LTC1_COUNT) "];\n", u, u, u
-);
     mstring_append(header,
 "\n"
 GLSL_DEFINE(projectionMat, GLSL_C_MAT4(NV_IGRAPH_XF_XFCTX_PMAT0))
@@ -115,13 +143,10 @@ GLSL_DEFINE(sceneAmbientColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_FR_AMB) ".xyz")
 GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz")
 "\n"
 );
-    mstring_append_fmt(uniforms,
-"%smat4 invViewport;\n", u);
 
-    /* Skinning */
     unsigned int count;
     bool mix;
-    switch (state->skinning) {
+    switch (state->fixed_function.skinning) {
     case SKINNING_OFF:
         mix = false; count = 0; break;
     case SKINNING_1WEIGHTS:
@@ -141,7 +166,7 @@ GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz
         break;
     }
     mstring_append_fmt(body, "/* Skinning mode %d */\n",
-                       state->skinning);
+                       state->fixed_function.skinning);
 
     append_skinning_code(body, mix, count, "vec4",
                          "tPosition", "position",
@@ -150,12 +175,10 @@ GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz
                          "tNormal", "vec4(normal, 0.0)",
                          "invModelViewMat", "xyz");
 
-    /* Normalization */
-    if (state->normalization) {
+    if (state->fixed_function.normalization) {
         mstring_append(body, "tNormal = normalize(tNormal);\n");
     }
 
-    /* Texgen */
     for (i = 0; i < NV2A_MAX_TEXTURES; i++) {
         mstring_append_fmt(body, "/* Texgen for stage %d */\n",
                            i);
@@ -165,7 +188,7 @@ GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz
             /* TODO: TexGen View Model missing! */
             char c = "xyzw"[j];
             char cSuffix = "STRQ"[j];
-            switch (state->texgen[i][j]) {
+            switch (state->fixed_function.texgen[i][j]) {
             case TEXGEN_DISABLE:
                 mstring_append_fmt(body, "oT%d.%c = texture%d.%c;\n",
                                    i, c, i, c);
@@ -220,104 +243,104 @@ GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz
         }
     }
 
-    /* Apply texture matrices */
     for (i = 0; i < NV2A_MAX_TEXTURES; i++) {
-        if (state->texture_matrix_enable[i]) {
+        if (state->fixed_function.texture_matrix_enable[i]) {
             mstring_append_fmt(body,
                                "oT%d = oT%d * texMat%d;\n",
                                i, i, i);
         }
     }
 
-    /* Lighting */
-    if (state->lighting) {
-
+    if (!state->fixed_function.lighting) {
+        mstring_append(body, "  oD0 = diffuse;\n");
+        mstring_append(body, "  oD1 = specular;\n");
+        mstring_append(body, "  oB0 = backDiffuse;\n");
+        mstring_append(body, "  oB1 = backSpecular;\n");
+    } else {
         //FIXME: Do 2 passes if we want 2 sided-lighting?
-
         static char alpha_source_diffuse[] = "diffuse.a";
         static char alpha_source_specular[] = "specular.a";
         static char alpha_source_material[] = "material_alpha";
         const char *alpha_source = alpha_source_diffuse;
-        if (state->diffuse_src == MATERIAL_COLOR_SRC_MATERIAL) {
-            mstring_append_fmt(uniforms, "%sfloat material_alpha;\n", u);
+        if (state->fixed_function.diffuse_src == MATERIAL_COLOR_SRC_MATERIAL) {
             alpha_source = alpha_source_material;
-        } else if (state->diffuse_src == MATERIAL_COLOR_SRC_SPECULAR) {
+        } else if (state->fixed_function.diffuse_src == MATERIAL_COLOR_SRC_SPECULAR) {
             alpha_source = alpha_source_specular;
         }
 
-        if (state->ambient_src == MATERIAL_COLOR_SRC_MATERIAL) {
+        if (state->fixed_function.ambient_src == MATERIAL_COLOR_SRC_MATERIAL) {
             mstring_append_fmt(body, "oD0 = vec4(sceneAmbientColor, %s);\n", alpha_source);
-        } else if (state->ambient_src == MATERIAL_COLOR_SRC_DIFFUSE) {
+        } else if (state->fixed_function.ambient_src == MATERIAL_COLOR_SRC_DIFFUSE) {
             mstring_append_fmt(body, "oD0 = vec4(diffuse.rgb, %s);\n", alpha_source);
-        } else if (state->ambient_src == MATERIAL_COLOR_SRC_SPECULAR) {
+        } else if (state->fixed_function.ambient_src == MATERIAL_COLOR_SRC_SPECULAR) {
             mstring_append_fmt(body, "oD0 = vec4(specular.rgb, %s);\n", alpha_source);
         }
 
         mstring_append(body, "oD0.rgb *= materialEmissionColor.rgb;\n");
-        if (state->emission_src == MATERIAL_COLOR_SRC_MATERIAL) {
+        if (state->fixed_function.emission_src == MATERIAL_COLOR_SRC_MATERIAL) {
             mstring_append(body, "oD0.rgb += sceneAmbientColor;\n");
-        } else if (state->emission_src == MATERIAL_COLOR_SRC_DIFFUSE) {
+        } else if (state->fixed_function.emission_src == MATERIAL_COLOR_SRC_DIFFUSE) {
             mstring_append(body, "oD0.rgb += diffuse.rgb;\n");
-        } else if (state->emission_src == MATERIAL_COLOR_SRC_SPECULAR) {
+        } else if (state->fixed_function.emission_src == MATERIAL_COLOR_SRC_SPECULAR) {
             mstring_append(body, "oD0.rgb += specular.rgb;\n");
         }
 
         mstring_append(body, "oD1 = vec4(0.0, 0.0, 0.0, specular.a);\n");
 
+        if (state->fixed_function.local_eye) {
+            mstring_append(body,
+                "vec3 VPeye = normalize(eyePosition.xyz / eyePosition.w - tPosition.xyz / tPosition.w);\n"
+            );
+        }
+
         for (i = 0; i < NV2A_MAX_LIGHTS; i++) {
-            if (state->light[i] == LIGHT_OFF) {
+            if (state->fixed_function.light[i] == LIGHT_OFF) {
                 continue;
             }
 
-            /* FIXME: It seems that we only have to handle the surface colors if
-             *        they are not part of the material [= vertex colors].
-             *        If they are material the cpu will premultiply light
-             *        colors
-             */
-
             mstring_append_fmt(body, "/* Light %d */ {\n", i);
 
-            if (state->light[i] == LIGHT_LOCAL
-                    || state->light[i] == LIGHT_SPOT) {
+            if (state->fixed_function.light[i] == LIGHT_LOCAL
+                    || state->fixed_function.light[i] == LIGHT_SPOT) {
 
-                mstring_append_fmt(uniforms,
-                    "%svec3 lightLocalPosition%d;\n"
-                    "%svec3 lightLocalAttenuation%d;\n",
-                    u, i, u, i);
                 mstring_append_fmt(body,
-                    "  vec3 VP = lightLocalPosition%d - tPosition.xyz/tPosition.w;\n"
+                    "  vec3 tPos = tPosition.xyz/tPosition.w;\n"
+                    "  vec3 VP = lightLocalPosition[%d] - tPos;\n"
                     "  float d = length(VP);\n"
-//FIXME: if (d > lightLocalRange) { .. don't process this light .. } /* inclusive?! */ - what about directional lights?
-                    "  VP = normalize(VP);\n"
-                    "  float attenuation = 1.0 / (lightLocalAttenuation%d.x\n"
-                    "                               + lightLocalAttenuation%d.y * d\n"
-                    "                               + lightLocalAttenuation%d.z * d * d);\n"
-                    "  vec3 halfVector = normalize(VP + eyePosition.xyz / eyePosition.w);\n" /* FIXME: Not sure if eyePosition is correct */
-                    "  float nDotVP = max(0.0, dot(tNormal, VP));\n"
-                    "  float nDotHV = max(0.0, dot(tNormal, halfVector));\n",
-                    i, i, i, i);
-
+                    "  if (d <= lightLocalRange(%d)) {\n"  /* FIXME: Double check that range is inclusive */
+                    "    VP = normalize(VP);\n"
+                    "    float attenuation = 1.0 / (lightLocalAttenuation[%d].x\n"
+                    "                                 + lightLocalAttenuation[%d].y * d\n"
+                    "                                 + lightLocalAttenuation[%d].z * d * d);\n"
+                    "    vec3 halfVector = normalize(VP + %s);\n"
+                    "    float nDotVP = max(0.0, dot(tNormal, VP));\n"
+                    "    float nDotHV = max(0.0, dot(tNormal, halfVector));\n",
+                    i, i, i, i, i,
+                    state->fixed_function.local_eye ? "VPeye" : "vec3(0.0, 0.0, 0.0)"
+                );
             }
 
-            switch(state->light[i]) {
+            switch(state->fixed_function.light[i]) {
             case LIGHT_INFINITE:
 
                 /* lightLocalRange will be 1e+30 here */
 
-                mstring_append_fmt(uniforms,
-                    "%svec3 lightInfiniteHalfVector%d;\n"
-                    "%svec3 lightInfiniteDirection%d;\n",
-                    u, i, u, i);
                 mstring_append_fmt(body,
-                    "  float attenuation = 1.0;\n"
-                    "  float nDotVP = max(0.0, dot(tNormal, normalize(vec3(lightInfiniteDirection%d))));\n"
-                    "  float nDotHV = max(0.0, dot(tNormal, vec3(lightInfiniteHalfVector%d)));\n",
-                    i, i);
-
-                /* FIXME: Do specular */
-
-                /* FIXME: tBackDiffuse */
-
+                    "  {\n"
+                    "    float attenuation = 1.0;\n"
+                    "    vec3 lightDirection = normalize(lightInfiniteDirection[%d]);\n"
+                    "    float nDotVP = max(0.0, dot(tNormal, lightDirection));\n",
+                    i);
+                if (state->fixed_function.local_eye) {
+                    mstring_append(body,
+                        "    float nDotHV = max(0.0, dot(tNormal, normalize(lightDirection + VPeye)));\n"
+                    );
+                } else {
+                    mstring_append_fmt(body,
+                        "    float nDotHV = max(0.0, dot(tNormal, lightInfiniteHalfVector[%d]));\n",
+                        i
+                    );
+                }
                 break;
             case LIGHT_LOCAL:
                 /* Everything done already */
@@ -325,18 +348,18 @@ GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz
             case LIGHT_SPOT:
                 /* https://docs.microsoft.com/en-us/windows/win32/direct3d9/attenuation-and-spotlight-factor#spotlight-factor */
                 mstring_append_fmt(body,
-                    "  vec4 spotDir = lightSpotDirection(%d);\n"
-                    "  float invScale = 1/length(spotDir.xyz);\n"
-                    "  float cosHalfPhi = -invScale*spotDir.w;\n"
-                    "  float cosHalfTheta = invScale + cosHalfPhi;\n"
-                    "  float spotDirDotVP = dot(spotDir.xyz, VP);\n"
-                    "  float rho = invScale*spotDirDotVP;\n"
-                    "  if (rho > cosHalfTheta) {\n"
-                    "  } else if (rho <= cosHalfPhi) {\n"
-                    "    attenuation = 0.0;\n"
-                    "  } else {\n"
-                    "    attenuation *= spotDirDotVP + spotDir.w;\n" /* FIXME: lightSpotFalloff */
-                    "  }\n",
+                    "    vec4 spotDir = lightSpotDirection(%d);\n"
+                    "    float invScale = 1/length(spotDir.xyz);\n"
+                    "    float cosHalfPhi = -invScale*spotDir.w;\n"
+                    "    float cosHalfTheta = invScale + cosHalfPhi;\n"
+                    "    float spotDirDotVP = dot(spotDir.xyz, VP);\n"
+                    "    float rho = invScale*spotDirDotVP;\n"
+                    "    if (rho > cosHalfTheta) {\n"
+                    "    } else if (rho <= cosHalfPhi) {\n"
+                    "      attenuation = 0.0;\n"
+                    "    } else {\n"
+                    "      attenuation *= spotDirDotVP + spotDir.w;\n" /* FIXME: lightSpotFalloff */
+                    "    }\n",
                     i);
                 break;
             default:
@@ -345,57 +368,86 @@ GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz
             }
 
             mstring_append_fmt(body,
-                "  float pf;\n"
-                "  if (nDotVP == 0.0) {\n"
-                "    pf = 0.0;\n"
-                "  } else {\n"
-                "    pf = pow(nDotHV, /* specular(l, m, n, l1, m1, n1) */ 0.001);\n"
-                "  }\n"
-                "  vec3 lightAmbient = lightAmbientColor(%d) * attenuation;\n"
-                "  vec3 lightDiffuse = lightDiffuseColor(%d) * attenuation * nDotVP;\n"
-                "  vec3 lightSpecular = lightSpecularColor(%d) * pf;\n",
+                "    float pf;\n"
+                "    if (nDotVP == 0.0 || nDotHV == 0.0) {\n"
+                "      pf = 0.0;\n"
+                "    } else {\n"
+                "      pf = pow(nDotHV, specularPower);\n"
+                "    }\n"
+                "    vec3 lightAmbient = lightAmbientColor(%d) * attenuation;\n"
+                "    vec3 lightDiffuse = lightDiffuseColor(%d) * attenuation * nDotVP;\n"
+                "    vec3 lightSpecular = lightSpecularColor(%d) * attenuation * pf;\n",
                 i, i, i);
 
             mstring_append(body,
-                "  oD0.xyz += lightAmbient;\n");
+                "    oD0.xyz += lightAmbient;\n");
 
-            switch (state->diffuse_src) {
+            switch (state->fixed_function.diffuse_src) {
             case MATERIAL_COLOR_SRC_MATERIAL:
                 mstring_append(body,
-                               "  oD0.xyz += lightDiffuse;\n");
+                               "    oD0.xyz += lightDiffuse;\n");
                 break;
             case MATERIAL_COLOR_SRC_DIFFUSE:
                 mstring_append(body,
-                               "  oD0.xyz += diffuse.xyz * lightDiffuse;\n");
+                               "    oD0.xyz += diffuse.xyz * lightDiffuse;\n");
                 break;
             case MATERIAL_COLOR_SRC_SPECULAR:
                 mstring_append(body,
-                               "  oD0.xyz += specular.xyz * lightDiffuse;\n");
+                               "    oD0.xyz += specular.xyz * lightDiffuse;\n");
                 break;
             }
 
-            mstring_append(body,
-                "  oD1.xyz += specular.xyz * lightSpecular;\n");
+            switch (state->fixed_function.specular_src) {
+            case MATERIAL_COLOR_SRC_MATERIAL:
+                mstring_append(body,
+                               "    oD1.xyz += lightSpecular;\n");
+                break;
+            case MATERIAL_COLOR_SRC_DIFFUSE:
+                mstring_append(body,
+                               "    oD1.xyz += diffuse.xyz * lightSpecular;\n");
+                break;
+            case MATERIAL_COLOR_SRC_SPECULAR:
+                mstring_append(body,
+                               "    oD1.xyz += specular.xyz * lightSpecular;\n");
+                break;
+            }
 
-            mstring_append(body, "}\n");
+            mstring_append(body, "  }\n"
+                                 "}\n");
         }
-    } else {
-        mstring_append(body, "  oD0 = diffuse;\n");
-        mstring_append(body, "  oD1 = specular;\n");
+
+        /* TODO: Implement two-sided lighting */
+        mstring_append(body, "  oB0 = backDiffuse;\n");
+        mstring_append(body, "  oB1 = backSpecular;\n");
     }
 
     if (!state->specular_enable) {
         mstring_append(body, "  oD1 = vec4(0.0, 0.0, 0.0, 1.0);\n");
+        mstring_append(body, "  oB1 = vec4(0.0, 0.0, 0.0, 1.0);\n");
+    } else {
+        if (!state->separate_specular) {
+            if (state->fixed_function.lighting) {
+				mstring_append(body,
+				               "  oD0.xyz += oD1.xyz;\n"
+				               "  oB0.xyz += oB1.xyz;\n"
+				);
+            }
+			mstring_append(body,
+				           "  oD1 = specular;\n"
+				           "  oB1 = backSpecular;\n"
+			);
+        }
+        if (state->ignore_specular_alpha) {
+            mstring_append(body,
+                           "  oD1.a = 1.0;\n"
+                           "  oB1.a = 1.0;\n"
+            );
+        }
     }
 
-    mstring_append(body, "  oB0 = backDiffuse;\n");
-    mstring_append(body, "  oB1 = backSpecular;\n");
-
-    /* Fog */
     if (state->fog_enable) {
-
         /* From: https://www.opengl.org/registry/specs/NV/fog_distance.txt */
-        switch(state->foggen) {
+        switch(state->fixed_function.foggen) {
         case FOGGEN_SPEC_ALPHA:
             /* FIXME: Do we have to clamp here? */
             mstring_append(body, "  float fogDistance = clamp(specular.a, 0.0, 1.0);\n");
@@ -406,7 +458,7 @@ GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz
         case FOGGEN_PLANAR:
         case FOGGEN_ABS_PLANAR:
             mstring_append(body, "  float fogDistance = dot(fogPlane.xyz, tPosition.xyz) + fogPlane.w;\n");
-            if (state->foggen == FOGGEN_ABS_PLANAR) {
+            if (state->fixed_function.foggen == FOGGEN_ABS_PLANAR) {
                 mstring_append(body, "  fogDistance = abs(fogDistance);\n");
             }
             break;
@@ -414,81 +466,38 @@ GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz
             mstring_append(body, "  float fogDistance = fogCoord;\n");
             break;
         default:
-            assert(false);
+            assert(!"Invalid foggen mode");
             break;
         }
 
     }
 
     /* If skinning is off the composite matrix already includes the MV matrix */
-    if (state->skinning == SKINNING_OFF) {
+    if (state->fixed_function.skinning == SKINNING_OFF) {
         mstring_append(body, "  tPosition = position;\n");
     }
 
     mstring_append(body,
-    "   oPos = tPosition * compositeMat;\n"
-    "   oPos.w = clampAwayZeroInf(oPos.w);\n"
-    "   oPos = invViewport * oPos;\n"
+    "  oPos = tPosition * compositeMat;\n"
+    "  oPos.z = oPos.z / clipRange.y;\n"
+    "  oPos.w = clampAwayZeroInf(oPos.w);\n"
+    "  oPos.xy /= oPos.w;\n"
+    "  oPos.xy += c[" stringify(NV_IGRAPH_XF_XFCTX_VPOFF) "].xy;\n"
+    "  oPos.xy = roundScreenCoords(oPos.xy);\n"
+    "  oPos.xy = (2.0f * oPos.xy - surfaceSize) / surfaceSize;\n"
+    "  oPos.xy *= oPos.w;\n"
     );
 
-    if (state->vulkan) {
-        mstring_append(body, "   oPos.y *= -1;\n");
-    }
-
     /* FIXME: Testing */
     if (state->point_params_enable) {
-        mstring_append_fmt(
+        mstring_append(
             body,
             "  float d_e = length(position * modelViewMat0);\n"
-            "  oPts.x = 1/sqrt(%f + %f*d_e + %f*d_e*d_e) + %f;\n",
-            state->point_params[0], state->point_params[1], state->point_params[2],
-            state->point_params[6]);
-        mstring_append_fmt(body, "  oPts.x = min(oPts.x*%f + %f, 64.0) * %d;\n",
-                           state->point_params[3], state->point_params[7],
+            "  oPts.x = 1/sqrt(pointParams[0] + pointParams[1] * d_e + pointParams[2] * d_e * d_e) + pointParams[6];\n");
+        mstring_append_fmt(body, "  oPts.x = min(oPts.x * pointParams[3] + pointParams[7], 64.0) * %d;\n",
                            state->surface_scale_factor);
     } else {
         mstring_append_fmt(body, "  oPts.x = %f * %d;\n", state->point_size,
                            state->surface_scale_factor);
     }
 }
-
-static void append_skinning_code(MString* str, bool mix,
-                                 unsigned int count, const char* type,
-                                 const char* output, const char* input,
-                                 const char* matrix, const char* swizzle)
-{
-    if (count == 0) {
-        mstring_append_fmt(str, "%s %s = (%s * %s0).%s;\n",
-                           type, output, input, matrix, swizzle);
-    } else {
-        mstring_append_fmt(str, "%s %s = %s(0.0);\n", type, output, type);
-        if (mix) {
-            /* Generated final weight (like GL_WEIGHT_SUM_UNITY_ARB) */
-            mstring_append(str, "{\n"
-                                "  float weight_i;\n"
-                                "  float weight_n = 1.0;\n");
-            int i;
-            for (i = 0; i < count; i++) {
-                if (i < (count - 1)) {
-                    char c = "xyzw"[i];
-                    mstring_append_fmt(str, "  weight_i = weight.%c;\n"
-                                            "  weight_n -= weight_i;\n",
-                                       c);
-                } else {
-                    mstring_append(str, "  weight_i = weight_n;\n");
-                }
-                mstring_append_fmt(str, "  %s += (%s * %s%d).%s * weight_i;\n",
-                                   output, input, matrix, i, swizzle);
-            }
-            mstring_append(str, "}\n");
-        } else {
-            /* Individual weights */
-            int i;
-            for (i = 0; i < count; i++) {
-                char c = "xyzw"[i];
-                mstring_append_fmt(str, "%s += (%s * %s%d).%s * weight.%c;\n",
-                                   output, input, matrix, i, swizzle, c);
-            }
-        }
-    }
-}

hw/xbox/nv2a/pgraph/glsl/vsh-ff.h

@@ -3,7 +3,7 @@
  *
  * Copyright (c) 2015 espes
  * Copyright (c) 2015 Jannik Vogel
- * Copyright (c) 2020-2024 Matt Borgerson
+ * Copyright (c) 2020-2025 Matt Borgerson
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
@@ -23,9 +23,9 @@
 #define HW_XBOX_NV2A_PGRAPH_GLSL_VSH_FF_H
 
 #include "qemu/mstring.h"
-#include "hw/xbox/nv2a/pgraph/shaders.h"
+#include "vsh.h"
 
-void pgraph_gen_vsh_ff_glsl(const ShaderState *state, MString *header,
-                            MString *body, MString *uniforms);
+void pgraph_glsl_gen_vsh_ff(const VshState *state, MString *header,
+                            MString *body);
 
 #endif

hw/xbox/nv2a/pgraph/glsl/vsh-prog.c

@@ -3,6 +3,7 @@
  *
  * Copyright (c) 2014 Jannik Vogel
  * Copyright (c) 2012 espes
+ * Copyright (c) 2025 Matt Borgerson
  *
  * Based on:
  * Cxbx, VertexShader.cpp
@@ -32,66 +33,10 @@
 #include <stdbool.h>
 #include <assert.h>
 
-#include "hw/xbox/nv2a/pgraph/vsh.h"
 #include "common.h"
+#include "vsh.h"
 #include "vsh-prog.h"
 
-#define VSH_D3DSCM_CORRECTION 96
-
-
-typedef enum {
-    PARAM_UNKNOWN = 0,
-    PARAM_R,
-    PARAM_V,
-    PARAM_C
-} VshParameterType;
-
-typedef enum {
-    OUTPUT_C = 0,
-    OUTPUT_O
-} VshOutputType;
-
-typedef enum {
-    OMUX_MAC = 0,
-    OMUX_ILU
-} VshOutputMux;
-
-typedef enum {
-    ILU_NOP = 0,
-    ILU_MOV,
-    ILU_RCP,
-    ILU_RCC,
-    ILU_RSQ,
-    ILU_EXP,
-    ILU_LOG,
-    ILU_LIT
-} VshILU;
-
-typedef enum {
-    MAC_NOP,
-    MAC_MOV,
-    MAC_MUL,
-    MAC_ADD,
-    MAC_MAD,
-    MAC_DP3,
-    MAC_DPH,
-    MAC_DP4,
-    MAC_DST,
-    MAC_MIN,
-    MAC_MAX,
-    MAC_SLT,
-    MAC_SGE,
-    MAC_ARL
-} VshMAC;
-
-typedef enum {
-    SWIZZLE_X = 0,
-    SWIZZLE_Y,
-    SWIZZLE_Z,
-    SWIZZLE_W
-} VshSwizzle;
-
-
 typedef struct VshFieldMapping {
     VshFieldName field_name;
     uint8_t subtoken;
@@ -143,7 +88,6 @@ static const VshFieldMapping field_mapping[] = {
     {  FLD_FINAL,            3,    0,     1 }
 };
 
-
 typedef struct VshOpcodeParams {
     bool A;
     bool B;
@@ -286,8 +230,6 @@ static const char* out_reg_name[] = {
     "A0.x",
 };
 
-
-
 // Retrieves a number of bits in the instruction token
 static int vsh_get_from_token(const uint32_t *shader_token,
                               uint8_t subtoken,
@@ -306,7 +248,6 @@ uint8_t vsh_get_field(const uint32_t *shader_token, VshFieldName field_name)
                                         field_mapping[field_name].bit_length));
 }
 
-
 // Converts the C register address to disassembly format
 static int16_t convert_c_register(const int16_t c_reg)
 {
@@ -315,9 +256,7 @@ static int16_t convert_c_register(const int16_t c_reg)
     return r; //FIXME: = c_reg?!
 }
 
-
-
-static MString* decode_swizzle(const uint32_t *shader_token,
+static MString *decode_swizzle(const uint32_t *shader_token,
                                VshFieldName swizzle_field)
 {
     const char* swizzle_str = "xyzw";
@@ -355,10 +294,9 @@ static MString* decode_swizzle(const uint32_t *shader_token,
     }
 }
 
-static MString* decode_opcode_input(const uint32_t *shader_token,
+static MString *decode_opcode_input(const uint32_t *shader_token,
                                     VshParameterType param,
-                                    VshFieldName neg_field,
-                                    int reg_num)
+                                    VshFieldName neg_field, int reg_num)
 {
     /* This function decodes a vertex shader opcode parameter into a string.
      * Input A, B or C is controlled via the Param and NEG fieldnames,
@@ -408,13 +346,10 @@ static MString* decode_opcode_input(const uint32_t *shader_token,
     return ret_str;
 }
 
-
-static MString* decode_opcode(const uint32_t *shader_token,
-                              VshOutputMux out_mux,
-                              uint32_t mask,
-                              const char *opcode,
-                              const char *inputs,
-                              MString** suffix)
+static MString *decode_opcode(const uint32_t *shader_token,
+                              VshOutputMux out_mux, uint32_t mask,
+                              const char *opcode, const char *inputs,
+                              MString **suffix)
 {
     MString *ret = mstring_new();
     int reg_num = vsh_get_field(shader_token, FLD_OUT_R);
@@ -496,8 +431,7 @@ static MString* decode_opcode(const uint32_t *shader_token,
     return ret;
 }
 
-
-static MString* decode_token(const uint32_t *shader_token)
+static MString *decode_token(const uint32_t *shader_token)
 {
     MString *ret;
 
@@ -639,7 +573,7 @@ static const char* vsh_header =
     // Unfortunately mix() falls victim to the same handling of exceptional
     // (inf/NaN) handling as a multiply, so per-component comparisons are used
     // to guarantee HW behavior (anything * 0 must == 0).
-    "  vec4 zero_components = sign(src0) * sign(src1);\n"
+    "  vec4 zero_components = sign(NaNToOne(src0)) * sign(NaNToOne(src1));\n"
     "  vec4 ret = src0 * src1;\n"
     "  if (zero_components.x == 0.0) { ret.x = 0.0; }\n"
     "  if (zero_components.y == 0.0) { ret.y = 0.0; }\n"
@@ -789,11 +723,9 @@ static const char* vsh_header =
     "  return t;\n"
     "}\n";
 
-void pgraph_gen_vsh_prog_glsl(uint16_t version,
-                   const uint32_t *tokens,
-                   unsigned int length,
-                   bool vulkan,
-                   MString *header, MString *body)
+void pgraph_glsl_gen_vsh_prog(uint16_t version, const uint32_t *tokens,
+                              unsigned int length, MString *header,
+                              MString *body)
 {
 
     mstring_append(header, vsh_header);
@@ -805,12 +737,10 @@ void pgraph_gen_vsh_prog_glsl(uint16_t version,
         const uint32_t* cur_token = &tokens[slot * VSH_TOKEN_SIZE];
         MString *token_str = decode_token(cur_token);
         mstring_append_fmt(body,
-                           "  /* Slot %d: 0x%08X 0x%08X 0x%08X 0x%08X */",
-                           slot,
-                           cur_token[0],cur_token[1],cur_token[2],cur_token[3]);
-        mstring_append(body, "\n");
-        mstring_append(body, mstring_get_str(token_str));
-        mstring_append(body, "\n");
+                           "  /* Slot %d: 0x%08X 0x%08X 0x%08X 0x%08X */\n"
+                           "  %s\n",
+                           slot, cur_token[0], cur_token[1], cur_token[2],
+                           cur_token[3], mstring_get_str(token_str));
         mstring_unref(token_str);
 
         if (vsh_get_field(cur_token, FLD_FINAL)) {
@@ -821,22 +751,12 @@ void pgraph_gen_vsh_prog_glsl(uint16_t version,
     assert(has_final);
 
     mstring_append(body,
-        /* the shaders leave the result in screen space, while
-         * opengl expects it in clip space.
-         * TODO: the pixel-center co-ordinate differences should handled
+        /* The shaders leave the result in screen space, while OpenGL expects it
+         * in clip space.
          */
-        "  oPos.x = 2.0 * (oPos.x - surfaceSize.x * 0.5) / surfaceSize.x;\n"
-        );
+        "  oPos.xy = roundScreenCoords(oPos.xy);\n"
+        "  oPos.xy = (2.0f * oPos.xy - surfaceSize) / surfaceSize;\n"
 
-    if (vulkan) {
-        mstring_append(body,
-                       "  oPos.y = 2.0 * oPos.y / surfaceSize.y - 1.0;\n");
-    } else {
-        mstring_append(body, "  oPos.y = -2.0 * (oPos.y - surfaceSize.y * 0.5) "
-                             "/ surfaceSize.y;\n");
-    }
-
-    mstring_append(body,
         "  oPos.z = oPos.z / clipRange.y;\n"
         "  oPos.w = clampAwayZeroInf(oPos.w);\n"
 

hw/xbox/nv2a/pgraph/glsl/vsh-prog.h

@@ -3,13 +3,7 @@
  *
  * Copyright (c) 2014 Jannik Vogel
  * Copyright (c) 2012 espes
- *
- * Based on:
- * Cxbx, VertexShader.cpp
- * Copyright (c) 2004 Aaron Robinson <caustik@caustik.com>
- *                    Kingofc <kingofc@freenet.de>
- * Dxbx, uPushBuffer.pas
- * Copyright (c) 2007 Shadow_tj, PatrickvL
+ * Copyright (c) 2025 Matt Borgerson
  *
  * This program is free software; you can redistribute it and/or
  * modify it under the terms of the GNU General Public License as
@@ -28,8 +22,10 @@
 #ifndef HW_XBOX_NV2A_PGRAPH_GLSL_VSH_PROG_H
 #define HW_XBOX_NV2A_PGRAPH_GLSL_VSH_PROG_H
 
-void pgraph_gen_vsh_prog_glsl(uint16_t version, const uint32_t *tokens,
-                              unsigned int length,
-                              bool vulkan, MString *header, MString *body);
+#include "qemu/mstring.h"
+
+void pgraph_glsl_gen_vsh_prog(uint16_t version, const uint32_t *tokens,
+                              unsigned int length, MString *header,
+                              MString *body);
 
 #endif

hw/xbox/nv2a/pgraph/glsl/vsh.c

@@ -3,7 +3,7 @@
  *
  * Copyright (c) 2015 espes
  * Copyright (c) 2015 Jannik Vogel
- * Copyright (c) 2020-2024 Matt Borgerson
+ * Copyright (c) 2020-2025 Matt Borgerson
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
@@ -20,39 +20,176 @@
  */
 
 #include "qemu/osdep.h"
-#include "hw/xbox/nv2a/pgraph/shaders.h"
-#include "common.h"
+#include "hw/xbox/nv2a/pgraph/pgraph.h"
 #include "vsh.h"
 #include "vsh-ff.h"
 #include "vsh-prog.h"
-#include <stdbool.h>
 
-MString *pgraph_gen_vsh_glsl(const ShaderState *state, bool prefix_outputs)
+DEF_UNIFORM_INFO_ARR(VshUniform, VSH_UNIFORM_DECL_X)
+
+static void set_fixed_function_vsh_state(PGRAPHState *pg,
+                                         FixedFunctionVshState *state)
 {
-    int i;
-    MString *output = mstring_new();
-    mstring_append_fmt(output, "#version %d\n\n", state->vulkan ? 450 : 400);
+    state->skinning = (enum VshSkinning)GET_MASK(
+        pgraph_reg_r(pg, NV_PGRAPH_CSV0_D), NV_PGRAPH_CSV0_D_SKIN);
+    state->normalization = pgraph_reg_r(pg, NV_PGRAPH_CSV0_C) &
+                           NV_PGRAPH_CSV0_C_NORMALIZATION_ENABLE;
+    state->local_eye =
+        GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CSV0_C), NV_PGRAPH_CSV0_C_LOCALEYE);
 
-    MString *header = mstring_from_str("");
-    MString *uniforms = mstring_from_str("");
+    state->emission_src = (enum MaterialColorSource)GET_MASK(
+        pgraph_reg_r(pg, NV_PGRAPH_CSV0_C), NV_PGRAPH_CSV0_C_EMISSION);
+    state->ambient_src = (enum MaterialColorSource)GET_MASK(
+        pgraph_reg_r(pg, NV_PGRAPH_CSV0_C), NV_PGRAPH_CSV0_C_AMBIENT);
+    state->diffuse_src = (enum MaterialColorSource)GET_MASK(
+        pgraph_reg_r(pg, NV_PGRAPH_CSV0_C), NV_PGRAPH_CSV0_C_DIFFUSE);
+    state->specular_src = (enum MaterialColorSource)GET_MASK(
+        pgraph_reg_r(pg, NV_PGRAPH_CSV0_C), NV_PGRAPH_CSV0_C_SPECULAR);
 
-    const char *u = state->vulkan ? "" : "uniform "; // FIXME: Remove
+    for (int i = 0; i < 4; i++) {
+        state->texture_matrix_enable[i] = pg->texture_matrix_enable[i];
+    }
 
-    mstring_append_fmt(uniforms,
-        "%svec4 clipRange;\n"
-        "%svec2 surfaceSize;\n"
-        "%svec4 c[" stringify(NV2A_VERTEXSHADER_CONSTANTS) "];\n"
-        "%svec2 fogParam;\n",
-        u, u, u, u
-        );
+    for (int i = 0; i < 4; i++) {
+        unsigned int reg = (i < 2) ? NV_PGRAPH_CSV1_A : NV_PGRAPH_CSV1_B;
+        for (int j = 0; j < 4; j++) {
+            unsigned int masks[] = {
+                (i % 2) ? NV_PGRAPH_CSV1_A_T1_S : NV_PGRAPH_CSV1_A_T0_S,
+                (i % 2) ? NV_PGRAPH_CSV1_A_T1_T : NV_PGRAPH_CSV1_A_T0_T,
+                (i % 2) ? NV_PGRAPH_CSV1_A_T1_R : NV_PGRAPH_CSV1_A_T0_R,
+                (i % 2) ? NV_PGRAPH_CSV1_A_T1_Q : NV_PGRAPH_CSV1_A_T0_Q
+            };
+            state->texgen[i][j] =
+                (enum VshTexgen)GET_MASK(pgraph_reg_r(pg, reg), masks[j]);
+        }
+    }
 
-    mstring_append(header,
+    state->lighting =
+        GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CSV0_C), NV_PGRAPH_CSV0_C_LIGHTING);
+    if (state->lighting) {
+        for (int i = 0; i < NV2A_MAX_LIGHTS; i++) {
+            state->light[i] =
+                (enum VshLight)GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CSV0_D),
+                                        NV_PGRAPH_CSV0_D_LIGHT0 << (i * 2));
+        }
+    }
+
+    if (pgraph_reg_r(pg, NV_PGRAPH_CONTROL_3) & NV_PGRAPH_CONTROL_3_FOGENABLE) {
+        state->foggen = (enum VshFoggen)GET_MASK(
+            pgraph_reg_r(pg, NV_PGRAPH_CSV0_D), NV_PGRAPH_CSV0_D_FOGGENMODE);
+    }
+}
+
+static void set_programmable_vsh_state(PGRAPHState *pg,
+                                       ProgrammableVshState *prog)
+{
+    int program_start = GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CSV0_C),
+                                 NV_PGRAPH_CSV0_C_CHEOPS_PROGRAM_START);
+
+    prog->program_length = 0;
+    for (int i = program_start; i < NV2A_MAX_TRANSFORM_PROGRAM_LENGTH; i++) {
+        uint32_t *cur_token = (uint32_t *)&pg->program_data[i];
+        memcpy(&prog->program_data[prog->program_length], cur_token,
+               VSH_TOKEN_SIZE * sizeof(uint32_t));
+        prog->program_length++;
+
+        if (vsh_get_field(cur_token, FLD_FINAL)) {
+            break;
+        }
+    }
+}
+
+void pgraph_glsl_set_vsh_state(PGRAPHState *pg, VshState *vsh)
+{
+    bool vertex_program = GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CSV0_D),
+                                   NV_PGRAPH_CSV0_D_MODE) == 2;
+
+    bool fixed_function = GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CSV0_D),
+                                   NV_PGRAPH_CSV0_D_MODE) == 0;
+
+    assert(vertex_program || fixed_function);
+
+    vsh->surface_scale_factor = pg->surface_scale_factor; // FIXME
+
+    vsh->compressed_attrs = pg->compressed_attrs;
+    vsh->uniform_attrs = pg->uniform_attrs;
+    vsh->swizzle_attrs = pg->swizzle_attrs;
+
+    vsh->specular_enable = GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CSV0_C),
+                                    NV_PGRAPH_CSV0_C_SPECULAR_ENABLE);
+    vsh->separate_specular = GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CSV0_C),
+                                      NV_PGRAPH_CSV0_C_SEPARATE_SPECULAR);
+    vsh->ignore_specular_alpha =
+        !GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CSV0_C),
+                  NV_PGRAPH_CSV0_C_ALPHA_FROM_MATERIAL_SPECULAR);
+    vsh->specular_power = pg->specular_power;
+    vsh->specular_power_back = pg->specular_power_back;
+
+    vsh->z_perspective = pgraph_reg_r(pg, NV_PGRAPH_CONTROL_0) &
+                         NV_PGRAPH_CONTROL_0_Z_PERSPECTIVE_ENABLE;
+
+    vsh->point_params_enable = GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CSV0_D),
+                                        NV_PGRAPH_CSV0_D_POINTPARAMSENABLE);
+    vsh->point_size = GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_POINTSIZE),
+                               NV097_SET_POINT_SIZE_V) /
+                      8.0f;
+    if (vsh->point_params_enable) {
+        for (int i = 0; i < 8; i++) {
+            vsh->point_params[i] = pg->point_params[i];
+        }
+    }
+
+    vsh->smooth_shading = GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CONTROL_3),
+                                   NV_PGRAPH_CONTROL_3_SHADEMODE) ==
+                          NV_PGRAPH_CONTROL_3_SHADEMODE_SMOOTH;
+
+    vsh->fog_enable =
+        pgraph_reg_r(pg, NV_PGRAPH_CONTROL_3) & NV_PGRAPH_CONTROL_3_FOGENABLE;
+    if (vsh->fog_enable) {
+        /*FIXME: Use CSV0_D? */
+        vsh->fog_mode =
+            (enum VshFogMode)GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CONTROL_3),
+                                      NV_PGRAPH_CONTROL_3_FOG_MODE);
+    }
+
+    vsh->is_fixed_function = fixed_function;
+    if (fixed_function) {
+        set_fixed_function_vsh_state(pg, &vsh->fixed_function);
+    } else {
+        set_programmable_vsh_state(pg, &vsh->programmable);
+    }
+}
+
+MString *pgraph_glsl_gen_vsh(const VshState *state, GenVshGlslOptions opts)
+{
+    MString *uniforms = mstring_new();
+    const char *u = opts.vulkan ? "" : "uniform ";
+    for (int i = 0; i < ARRAY_SIZE(VshUniformInfo); i++) {
+        const UniformInfo *info = &VshUniformInfo[i];
+        const char *type_str = uniform_element_type_to_str[info->type];
+        if (i == VshUniform_inlineValue &&
+            (!state->uniform_attrs ||
+             opts.use_push_constants_for_uniform_attrs)) {
+            continue;
+        }
+        if (info->count == 1) {
+            mstring_append_fmt(uniforms, "%s%s %s;\n", u, type_str,
+                               info->name);
+        } else {
+            mstring_append_fmt(uniforms, "%s%s %s[%zd];\n", u, type_str,
+                               info->name, info->count);
+        }
+    }
+
+    MString *header = mstring_from_str(
         GLSL_DEFINE(fogPlane, GLSL_C(NV_IGRAPH_XF_XFCTX_FOG))
         GLSL_DEFINE(texMat0, GLSL_C_MAT4(NV_IGRAPH_XF_XFCTX_T0MAT))
         GLSL_DEFINE(texMat1, GLSL_C_MAT4(NV_IGRAPH_XF_XFCTX_T1MAT))
         GLSL_DEFINE(texMat2, GLSL_C_MAT4(NV_IGRAPH_XF_XFCTX_T2MAT))
         GLSL_DEFINE(texMat3, GLSL_C_MAT4(NV_IGRAPH_XF_XFCTX_T3MAT))
 
+        "\n"
+        "#define FLOAT_MAX uintBitsToFloat(0x7F7FFFFFu)\n"
         "\n"
         "vec4 oPos = vec4(0.0,0.0,0.0,1.0);\n"
         "vec4 oD0 = vec4(0.0,0.0,0.0,1.0);\n"
@@ -81,12 +218,23 @@ MString *pgraph_gen_vsh_glsl(const ShaderState *state, bool prefix_outputs)
         "    t = clamp(t, uintBitsToFloat(0xDF800000), uintBitsToFloat(0x9F800000));\n"
         "  }\n"
         "  return t;\n"
+        "}\n"
+        "\n"
+        "vec4 NaNToOne(vec4 src) {\n"
+        "  return mix(src, vec4(1.0), isnan(src));\n"
+        "}\n"
+        "\n"
+        // Xbox NV2A rasterizer appears to have 4 bit precision fixed-point
+        // fractional part and to convert floating-point coordinates by
+        // by truncating (not flooring).
+        "vec2 roundScreenCoords(vec2 pos) {\n"
+        "  return trunc(pos * 16.0f) / 16.0f;\n"
         "}\n");
 
-    pgraph_get_glsl_vtx_header(header, state->vulkan, state->smooth_shading,
-                             false, prefix_outputs, false);
+    pgraph_glsl_get_vtx_header(header, opts.vulkan, state->smooth_shading,
+                               false, opts.prefix_outputs, false);
 
-    if (prefix_outputs) {
+    if (opts.prefix_outputs) {
         mstring_append(header,
                        "#define vtxD0 v_vtxD0\n"
                        "#define vtxD1 v_vtxD1\n"
@@ -103,7 +251,7 @@ MString *pgraph_gen_vsh_glsl(const ShaderState *state, bool prefix_outputs)
 
     int num_uniform_attrs = 0;
 
-    for (i = 0; i < NV2A_VERTEXSHADER_ATTRIBUTES; i++) {
+    for (int i = 0; i < NV2A_VERTEXSHADER_ATTRIBUTES; i++) {
         bool is_uniform = state->uniform_attrs & (1 << i);
         bool is_swizzled = state->swizzle_attrs & (1 << i);
         bool is_compressed = state->compressed_attrs & (1 << i);
@@ -128,11 +276,12 @@ MString *pgraph_gen_vsh_glsl(const ShaderState *state, bool prefix_outputs)
             }
         }
     }
+
     mstring_append(header, "\n");
 
     MString *body = mstring_from_str("void main() {\n");
 
-    for (i = 0; i < NV2A_VERTEXSHADER_ATTRIBUTES; i++) {
+    for (int i = 0; i < NV2A_VERTEXSHADER_ATTRIBUTES; i++) {
         if (state->compressed_attrs & (1 << i)) {
             mstring_append_fmt(
                 body, "vec4 v%d = decompress_11_11_10(v%d_cmp);\n", i, i);
@@ -144,23 +293,19 @@ MString *pgraph_gen_vsh_glsl(const ShaderState *state, bool prefix_outputs)
 
     }
 
-    if (state->fixed_function) {
-        pgraph_gen_vsh_ff_glsl(state, header, body, uniforms);
-    } else if (state->vertex_program) {
-        pgraph_gen_vsh_prog_glsl(VSH_VERSION_XVS,
-                                 (uint32_t *)state->program_data,
-                                 state->program_length,
-                                 state->vulkan, header, body);
+    if (state->is_fixed_function) {
+        pgraph_glsl_gen_vsh_ff(state, header, body);
     } else {
-        assert(false);
+        pgraph_glsl_gen_vsh_prog(
+            VSH_VERSION_XVS, (uint32_t *)state->programmable.program_data,
+            state->programmable.program_length, header, body);
     }
 
-
-    /* Fog */
-
-    if (state->fog_enable) {
-
-        if (state->vertex_program) {
+    if (!state->fog_enable) {
+        /* FIXME: Is the fog still calculated / passed somehow?! */
+        mstring_append(body, "  oFog = vec4(1.0);\n");
+    } else {
+        if (!state->is_fixed_function) {
             /* FIXME: Does foggen do something here? Let's do some tracking..
              *
              *   "RollerCoaster Tycoon" has
@@ -221,7 +366,6 @@ MString *pgraph_gen_vsh_glsl(const ShaderState *state, bool prefix_outputs)
             assert(false);
             break;
         }
-        /* Calculate absolute for the modes which need it */
         switch (state->fog_mode) {
         case FOG_MODE_LINEAR_ABS:
         case FOG_MODE_EXP_ABS:
@@ -232,17 +376,18 @@ MString *pgraph_gen_vsh_glsl(const ShaderState *state, bool prefix_outputs)
             break;
         }
 
-        mstring_append(body, "  oFog.xyzw = vec4(fogFactor);\n");
-    } else {
-        /* FIXME: Is the fog still calculated / passed somehow?!
+        /* Fog is clamped to min/max normal float values here to match HW
+         * interpolation. It is then clamped to [0,1] in the pixel shader.
          */
-        mstring_append(body, "  oFog.xyzw = vec4(1.0);\n");
+        // clang-format off
+        mstring_append(body,
+                       "  oFog = clamp(NaNToOne(vec4(fogFactor)), -FLOAT_MAX, FLOAT_MAX);\n");
+        // clang-format on
     }
 
-    /* Set outputs */
     mstring_append(body, "\n"
-                   "  vtxD0 = clamp(oD0, 0.0, 1.0);\n"
-                   "  vtxB0 = clamp(oB0, 0.0, 1.0);\n"
+                   "  vtxD0 = clamp(NaNToOne(oD0), 0.0, 1.0);\n"
+                   "  vtxB0 = clamp(NaNToOne(oB0), 0.0, 1.0);\n"
                    "  vtxFog = oFog.x;\n"
                    "  vtxT0 = oT0;\n"
                    "  vtxT1 = oT1;\n"
@@ -253,9 +398,16 @@ MString *pgraph_gen_vsh_glsl(const ShaderState *state, bool prefix_outputs)
 
     if (state->specular_enable) {
         mstring_append(body,
-                       "  vtxD1 = clamp(oD1, 0.0, 1.0);\n"
-                       "  vtxB1 = clamp(oB1, 0.0, 1.0);\n"
+                       "  vtxD1 = clamp(NaNToOne(oD1), 0.0, 1.0);\n"
+                       "  vtxB1 = clamp(NaNToOne(oB1), 0.0, 1.0);\n"
         );
+
+        if (state->ignore_specular_alpha) {
+            mstring_append(body,
+                           "  vtxD1.w = 1.0;\n"
+                           "  vtxB1.w = 1.0;\n"
+            );
+        }
     } else {
         mstring_append(body,
                        "  vtxD1 = vec4(0.0, 0.0, 0.0, 1.0);\n"
@@ -263,7 +415,7 @@ MString *pgraph_gen_vsh_glsl(const ShaderState *state, bool prefix_outputs)
         );
     }
 
-    if (state->vulkan) {
+    if (opts.vulkan) {
         mstring_append(body,
                    "  gl_Position = oPos;\n"
         );
@@ -276,25 +428,25 @@ MString *pgraph_gen_vsh_glsl(const ShaderState *state, bool prefix_outputs)
     mstring_append(body, "}\n");
 
     /* Return combined header + source */
-    if (state->vulkan) {
+    MString *output =
+        mstring_from_fmt("#version %d\n\n", opts.vulkan ? 450 : 400);
+
+    if (opts.vulkan) {
         // FIXME: Optimize uniforms
-        if (num_uniform_attrs > 0) {
-            if (state->use_push_constants_for_uniform_attrs) {
-                mstring_append_fmt(output,
-                    "layout(push_constant) uniform PushConstants {\n"
-                    "    vec4 inlineValue[%d];\n"
-                    "};\n\n", num_uniform_attrs);
-            } else {
-                mstring_append_fmt(uniforms, "    vec4 inlineValue[%d];\n",
-                                   num_uniform_attrs);
-            }
+        if (num_uniform_attrs > 0 &&
+            opts.use_push_constants_for_uniform_attrs) {
+            mstring_append_fmt(output,
+                               "layout(push_constant) uniform PushConstants {\n"
+                               "    vec4 inlineValue[%d];\n"
+                               "};\n\n",
+                               num_uniform_attrs);
         }
         mstring_append_fmt(
             output,
             "layout(binding = %d, std140) uniform VshUniforms {\n"
             "%s"
             "};\n\n",
-            VSH_UBO_BINDING, mstring_get_str(uniforms));
+            opts.ubo_binding, mstring_get_str(uniforms));
     } else {
         mstring_append(
             output, mstring_get_str(uniforms));
@@ -305,5 +457,110 @@ MString *pgraph_gen_vsh_glsl(const ShaderState *state, bool prefix_outputs)
 
     mstring_append(output, mstring_get_str(body));
     mstring_unref(body);
+
     return output;
 }
+
+void pgraph_glsl_set_vsh_uniform_values(PGRAPHState *pg, const VshState *state,
+                                        const VshUniformLocs locs,
+                                        VshUniformValues *values)
+{
+    if (locs[VshUniform_c] != -1) {
+        QEMU_BUILD_BUG_MSG(sizeof(values->c) != sizeof(pg->vsh_constants),
+                           "Uniform value size inconsistency");
+        memcpy(values->c, pg->vsh_constants, sizeof(pg->vsh_constants));
+    }
+
+    if (locs[VshUniform_clipRange] != -1) {
+        pgraph_glsl_set_clip_range_uniform_value(pg, values->clipRange[0]);
+    }
+
+    if (locs[VshUniform_fogParam] != -1) {
+        uint32_t param_0 = pgraph_reg_r(pg, NV_PGRAPH_FOGPARAM0);
+        uint32_t param_1 = pgraph_reg_r(pg, NV_PGRAPH_FOGPARAM1);
+        values->fogParam[0][0] = *(float *)&param_0;
+        values->fogParam[0][1] = *(float *)&param_1;
+    }
+
+    if (locs[VshUniform_pointParams] != -1) {
+        QEMU_BUILD_BUG_MSG(sizeof(values->pointParams) !=
+                               sizeof(pg->point_params),
+                           "Uniform value size inconsistency");
+        memcpy(values->pointParams, pg->point_params, sizeof(pg->point_params));
+    }
+
+    if (locs[VshUniform_material_alpha] != -1) {
+        values->material_alpha[0] = pg->material_alpha;
+    }
+
+    if (locs[VshUniform_inlineValue] != -1) {
+        pgraph_get_inline_values(pg, state->uniform_attrs, values->inlineValue,
+                                 NULL);
+    }
+
+    if (locs[VshUniform_surfaceSize] != -1) {
+        unsigned int aa_width = 1, aa_height = 1;
+        pgraph_apply_anti_aliasing_factor(pg, &aa_width, &aa_height);
+        float width = (float)pg->surface_binding_dim.width / aa_width;
+        float height = (float)pg->surface_binding_dim.height / aa_height;
+        values->surfaceSize[0][0] = width;
+        values->surfaceSize[0][1] = height;
+    }
+
+    if (state->is_fixed_function) {
+        if (locs[VshUniform_ltctxa] != -1) {
+            QEMU_BUILD_BUG_MSG(sizeof(values->ltctxa) != sizeof(pg->ltctxa),
+                               "Uniform value size inconsistency");
+            memcpy(values->ltctxa, pg->ltctxa, sizeof(pg->ltctxa));
+        }
+
+        if (locs[VshUniform_ltctxb] != -1) {
+            QEMU_BUILD_BUG_MSG(sizeof(values->ltctxb) != sizeof(pg->ltctxb),
+                               "Uniform value size inconsistency");
+            memcpy(values->ltctxb, pg->ltctxb, sizeof(pg->ltctxb));
+        }
+
+        if (locs[VshUniform_ltc1] != -1) {
+            QEMU_BUILD_BUG_MSG(sizeof(values->ltc1) != sizeof(pg->ltc1),
+                               "Uniform value size inconsistency");
+            memcpy(values->ltc1, pg->ltc1, sizeof(pg->ltc1));
+        }
+
+        if (locs[VshUniform_lightInfiniteHalfVector] != -1) {
+            QEMU_BUILD_BUG_MSG(sizeof(values->lightInfiniteHalfVector) !=
+                                   sizeof(pg->light_infinite_half_vector),
+                               "Uniform value size inconsistency");
+            memcpy(values->lightInfiniteHalfVector,
+                   pg->light_infinite_half_vector,
+                   sizeof(pg->light_infinite_half_vector));
+        }
+
+        if (locs[VshUniform_lightInfiniteDirection] != -1) {
+            QEMU_BUILD_BUG_MSG(sizeof(values->lightInfiniteDirection) !=
+                                   sizeof(pg->light_infinite_direction),
+                               "Uniform value size inconsistency");
+            memcpy(values->lightInfiniteDirection, pg->light_infinite_direction,
+                   sizeof(pg->light_infinite_direction));
+        }
+
+        if (locs[VshUniform_lightLocalPosition] != -1) {
+            QEMU_BUILD_BUG_MSG(sizeof(values->lightLocalPosition) !=
+                                   sizeof(pg->light_local_position),
+                               "Uniform value size inconsistency");
+            memcpy(values->lightLocalPosition, pg->light_local_position,
+                   sizeof(pg->light_local_position));
+        }
+
+        if (locs[VshUniform_lightLocalAttenuation] != -1) {
+            QEMU_BUILD_BUG_MSG(sizeof(values->lightLocalAttenuation) !=
+                                   sizeof(pg->light_local_attenuation),
+                               "Uniform value size inconsistency");
+            memcpy(values->lightLocalAttenuation, pg->light_local_attenuation,
+                   sizeof(pg->light_local_attenuation));
+        }
+
+        if (locs[VshUniform_specularPower] != -1) {
+            values->specularPower[0] = pg->specular_power;
+        }
+    }
+}

hw/xbox/nv2a/pgraph/glsl/vsh.h

@@ -3,7 +3,7 @@
  *
  * Copyright (c) 2015 espes
  * Copyright (c) 2015 Jannik Vogel
- * Copyright (c) 2020-2024 Matt Borgerson
+ * Copyright (c) 2020-2025 Matt Borgerson
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
@@ -22,12 +22,92 @@
 #ifndef HW_XBOX_NV2A_PGRAPH_GLSL_VSH_H
 #define HW_XBOX_NV2A_PGRAPH_GLSL_VSH_H
 
-#include "qemu/mstring.h"
-#include "hw/xbox/nv2a/pgraph/shaders.h"
+#include "common.h"
+#include "hw/xbox/nv2a/pgraph/vsh_regs.h"
 
-// FIXME: Move to struct
-#define VSH_UBO_BINDING 0
+typedef struct PGRAPHState PGRAPHState;
 
-MString *pgraph_gen_vsh_glsl(const ShaderState *state, bool prefix_outputs);
+typedef struct FixedFunctionVshState {
+    bool normalization;
+    bool texture_matrix_enable[4];
+    enum VshTexgen texgen[4][4];
+    enum VshFoggen foggen;
+    enum VshSkinning skinning;
+    bool lighting;
+    enum VshLight light[NV2A_MAX_LIGHTS];
+    enum MaterialColorSource emission_src;
+    enum MaterialColorSource ambient_src;
+    enum MaterialColorSource diffuse_src;
+    enum MaterialColorSource specular_src;
+    bool local_eye;
+} FixedFunctionVshState;
+
+typedef struct ProgrammableVshState {
+    uint32_t program_data[NV2A_MAX_TRANSFORM_PROGRAM_LENGTH][VSH_TOKEN_SIZE];
+    int program_length;
+} ProgrammableVshState;
+
+typedef struct {
+    unsigned int surface_scale_factor;  // FIXME: Remove
+
+    uint16_t compressed_attrs;
+    uint16_t uniform_attrs;
+    uint16_t swizzle_attrs;
+
+    bool fog_enable;
+    enum VshFogMode fog_mode;
+
+    bool specular_enable;
+    bool separate_specular;
+    bool ignore_specular_alpha;
+    float specular_power;
+    float specular_power_back;
+
+    bool point_params_enable;
+    float point_size;
+    float point_params[8];
+
+    bool smooth_shading;
+    bool z_perspective;
+
+    bool is_fixed_function;
+    FixedFunctionVshState fixed_function;
+    ProgrammableVshState programmable;
+} VshState;
+
+void pgraph_glsl_set_vsh_state(PGRAPHState *pg, VshState *state);
+
+#define VSH_UNIFORM_DECL_X(S, DECL)                          \
+    DECL(S, c, vec4, NV2A_VERTEXSHADER_CONSTANTS)            \
+    DECL(S, clipRange, vec4, 1)                              \
+    DECL(S, fogParam, vec2, 1)                               \
+    DECL(S, inlineValue, vec4, NV2A_VERTEXSHADER_ATTRIBUTES) \
+    DECL(S, lightInfiniteDirection, vec3, NV2A_MAX_LIGHTS)   \
+    DECL(S, lightInfiniteHalfVector, vec3, NV2A_MAX_LIGHTS)  \
+    DECL(S, lightLocalAttenuation, vec3, NV2A_MAX_LIGHTS)    \
+    DECL(S, lightLocalPosition, vec3, NV2A_MAX_LIGHTS)       \
+    DECL(S, ltc1, vec4, NV2A_LTC1_COUNT)                     \
+    DECL(S, ltctxa, vec4, NV2A_LTCTXA_COUNT)                 \
+    DECL(S, ltctxb, vec4, NV2A_LTCTXB_COUNT)                 \
+    DECL(S, material_alpha, float, 1)                        \
+    DECL(S, pointParams, float, 8)                           \
+    DECL(S, specularPower, float, 1)                         \
+    DECL(S, surfaceSize, vec2, 1)
+
+DECL_UNIFORM_TYPES(VshUniform, VSH_UNIFORM_DECL_X)
+
+typedef struct GenVshGlslOptions {
+    bool vulkan;
+    bool prefix_outputs;
+    bool use_push_constants_for_uniform_attrs;
+    int ubo_binding;
+} GenVshGlslOptions;
+
+MString *pgraph_glsl_gen_vsh(const VshState *state,
+                             GenVshGlslOptions glsl_opts);
+
+void pgraph_glsl_set_vsh_uniform_values(PGRAPHState *pg, const VshState *state,
+                                        const VshUniformLocs locs,
+                                        VshUniformValues *values);
 
 #endif

hw/xbox/nv2a/pgraph/meson.build

@@ -3,7 +3,6 @@ specific_ss.add(files(
 	'profile.c',
 	'rdi.c',
 	's3tc.c',
-	'shaders.c',
 	'swizzle.c',
 	'texture.c',
 	'vertex.c',

hw/xbox/nv2a/pgraph/methods.h.inc

@@ -27,6 +27,7 @@ DEF_METHOD(NV097, SET_COMBINER_SPECULAR_FOG_CW0)
 DEF_METHOD(NV097, SET_COMBINER_SPECULAR_FOG_CW1)
 DEF_METHOD_CASE_4(NV097, SET_TEXTURE_ADDRESS, 64)
 DEF_METHOD(NV097, SET_CONTROL0)
+DEF_METHOD(NV097, SET_LIGHT_CONTROL)
 DEF_METHOD(NV097, SET_COLOR_MATERIAL)
 DEF_METHOD(NV097, SET_FOG_MODE)
 DEF_METHOD(NV097, SET_FOG_GEN_MODE)
@@ -95,6 +96,7 @@ DEF_METHOD_RANGE(NV097, SET_FOG_PARAMS, 3)
 DEF_METHOD_RANGE(NV097, SET_TEXGEN_PLANE_S, 4*4*4)
 DEF_METHOD(NV097, SET_TEXGEN_VIEW_MODEL)
 DEF_METHOD_RANGE(NV097, SET_FOG_PLANE, 4)
+DEF_METHOD_RANGE(NV097, SET_SPECULAR_PARAMS, 6)
 DEF_METHOD_RANGE(NV097, SET_SCENE_AMBIENT_COLOR, 3)
 DEF_METHOD_RANGE(NV097, SET_VIEWPORT_OFFSET, 4)
 DEF_METHOD_RANGE(NV097, SET_POINT_PARAMS, 8)
@@ -103,6 +105,7 @@ DEF_METHOD_RANGE(NV097, SET_COMBINER_FACTOR0, 8)
 DEF_METHOD_RANGE(NV097, SET_COMBINER_FACTOR1, 8)
 DEF_METHOD_RANGE(NV097, SET_COMBINER_ALPHA_OCW, 8)
 DEF_METHOD_RANGE(NV097, SET_COMBINER_COLOR_ICW, 8)
+DEF_METHOD_RANGE(NV097, SET_COLOR_KEY_COLOR, 4)
 DEF_METHOD_RANGE(NV097, SET_VIEWPORT_SCALE, 4)
 DEF_METHOD_RANGE(NV097, SET_TRANSFORM_PROGRAM, 32)
 DEF_METHOD_RANGE(NV097, SET_TRANSFORM_CONSTANT, 32)
@@ -134,6 +137,11 @@ DEF_METHOD_RANGE(NV097, SET_TEXCOORD3_2F, 2)
 DEF_METHOD_RANGE(NV097, SET_TEXCOORD3_4F, 4)
 DEF_METHOD_RANGE(NV097, SET_TEXCOORD3_2S, 1)
 DEF_METHOD_RANGE(NV097, SET_TEXCOORD3_4S, 2)
+DEF_METHOD(NV097, SET_FOG_COORD)
+DEF_METHOD(NV097, SET_WEIGHT1F)
+DEF_METHOD_RANGE(NV097, SET_WEIGHT2F, 2)
+DEF_METHOD_RANGE(NV097, SET_WEIGHT3F, 3)
+DEF_METHOD_RANGE(NV097, SET_WEIGHT4F, 4)
 DEF_METHOD_RANGE(NV097, SET_VERTEX_DATA_ARRAY_FORMAT, 16)
 DEF_METHOD_RANGE(NV097, SET_VERTEX_DATA_ARRAY_OFFSET, 16)
 DEF_METHOD(NV097, SET_LOGIC_OP_ENABLE)
@@ -177,6 +185,7 @@ DEF_METHOD(NV097, CLEAR_SURFACE)
 DEF_METHOD(NV097, SET_CLEAR_RECT_HORIZONTAL)
 DEF_METHOD(NV097, SET_CLEAR_RECT_VERTICAL)
 DEF_METHOD_RANGE(NV097, SET_SPECULAR_FOG_FACTOR, 2)
+DEF_METHOD_RANGE(NV097, SET_SPECULAR_PARAMS_BACK, 6)
 DEF_METHOD(NV097, SET_SHADER_CLIP_PLANE_MODE)
 DEF_METHOD_RANGE(NV097, SET_COMBINER_COLOR_OCW, 8)
 DEF_METHOD(NV097, SET_COMBINER_CONTROL)

hw/xbox/nv2a/pgraph/pgraph.c

@@ -19,6 +19,8 @@
  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  */
 
+#include <math.h>
+
 #include "hw/xbox/nv2a/nv2a_int.h"
 #include "ui/xemu-notifications.h"
 #include "ui/xemu-settings.h"
@@ -223,6 +225,8 @@ void pgraph_init(NV2AState *d)
     qemu_event_init(&pg->sync_complete, false);
     qemu_event_init(&pg->flush_complete, false);
     qemu_cond_init(&pg->framebuffer_released);
+    qemu_event_init(&pg->renderer_switch_complete, false);
+    pg->renderer_switch_phase = PGRAPH_RENDERER_SWITCH_PHASE_IDLE;
 
     pg->frame_time = 0;
     pg->draw_time = 0;
@@ -1075,6 +1079,18 @@ DEF_METHOD(NV097, SET_CONTROL0)
              z_perspective);
 }
 
+DEF_METHOD(NV097, SET_LIGHT_CONTROL)
+{
+    PG_SET_MASK(NV_PGRAPH_CSV0_C, NV_PGRAPH_CSV0_C_SEPARATE_SPECULAR,
+             (parameter & NV097_SET_LIGHT_CONTROL_SEPARATE_SPECULAR) != 0);
+
+    PG_SET_MASK(NV_PGRAPH_CSV0_C, NV_PGRAPH_CSV0_C_LOCALEYE,
+             (parameter & NV097_SET_LIGHT_CONTROL_LOCALEYE) != 0);
+
+    PG_SET_MASK(NV_PGRAPH_CSV0_C, NV_PGRAPH_CSV0_C_ALPHA_FROM_MATERIAL_SPECULAR,
+             (parameter & NV097_SET_LIGHT_CONTROL_ALPHA_FROM_MATERIAL_SPECULAR) != 0);
+}
+
 DEF_METHOD(NV097, SET_COLOR_MATERIAL)
 {
     PG_SET_MASK(NV_PGRAPH_CSV0_C, NV_PGRAPH_CSV0_C_EMISSION,
@@ -1397,8 +1413,10 @@ DEF_METHOD(NV097, SET_BLEND_EQUATION)
 
 DEF_METHOD(NV097, SET_DEPTH_FUNC)
 {
-    PG_SET_MASK(NV_PGRAPH_CONTROL_0, NV_PGRAPH_CONTROL_0_ZFUNC,
-             parameter & 0xF);
+    if (parameter >= 0x200 && parameter <= 0x207) {
+        PG_SET_MASK(NV_PGRAPH_CONTROL_0, NV_PGRAPH_CONTROL_0_ZFUNC,
+                    parameter & 0xF);
+    }
 }
 
 DEF_METHOD(NV097, SET_COLOR_MASK)
@@ -1791,6 +1809,113 @@ DEF_METHOD_INC(NV097, SET_FOG_PLANE)
     pg->vsh_constants_dirty[NV_IGRAPH_XF_XFCTX_FOG] = true;
 }
 
+struct CurveCoefficients {
+  float a;
+  float b;
+  float c;
+};
+
+static const struct CurveCoefficients curve_coefficients[] = {
+  {1.000108475163, -9.838607076280, 54.829089549713},
+  {1.199164441703, -3.292603784852, 7.799987995214},
+  {8.653441252033, 29.189473787191, 43.586027561823},
+  {-531.307758450301, 117.398468683934, 113.155490738338},
+  {-4.662713151292, 1.221108944572, 1.217360986939},
+  {-124.435242105211, 35.401219563514, 35.408114377045},
+  {10672560.259502287954, 21565843.555823743343, 10894794.336297152564},
+  {-51973801.463933646679, -104199997.554352939129, -52225454.356278456748},
+  {972270.324080004124, 2025882.096547174733, 1054898.052467488218},
+};
+
+static const float kCoefficient0StepPoints[] = {
+  -0.022553957999, // power = 1.25
+  -0.421539008617, // power = 4.00
+  -0.678715527058, // power = 9.00
+  -0.838916420937, // power = 20.00
+  -0.961754500866, // power = 90.00
+  -0.990773200989, // power = 375.00
+  -0.994858562946, // power = 650.00
+  -0.996561050415, // power = 1000.00
+  -0.999547004700, // power = 1250.00
+};
+
+static float reconstruct_quadratic(float c0, const struct CurveCoefficients *coefficients) {
+  return coefficients->a + coefficients->b * c0 + coefficients->c * c0 * c0;
+}
+
+static float reconstruct_saturation_growth_rate(float c0, const struct CurveCoefficients *coefficients) {
+  return (coefficients->a * c0) / (coefficients->b + coefficients->c * c0);
+}
+
+static float (* const reconstruct_func_map[])(float, const struct CurveCoefficients *) = {
+  reconstruct_quadratic, // 1.0..1.25 max error 0.01 %
+  reconstruct_quadratic, // 1.25..4.0 max error 2.2 %
+  reconstruct_quadratic, // 4.0..9.0 max error 2.3 %
+  reconstruct_saturation_growth_rate, // 9.0..20.0 max error 1.4 %
+  reconstruct_saturation_growth_rate, // 20.0..90.0 max error 2.1 %
+  reconstruct_saturation_growth_rate, // 90.0..375.0 max error 2.8%
+  reconstruct_quadratic, // 375..650 max error 1.0 %
+  reconstruct_quadratic, // 650..1000 max error 1.7%
+  reconstruct_quadratic, // 1000..1250 max error 1.0%
+};
+
+static float reconstruct_specular_power(const float *params) {
+  // See https://github.com/dracc/xgu/blob/db3172d8c983629f0dc971092981846da22438ae/xgux.h#L279
+
+  // Values < 1.0 will result in a positive c1 and (c2 - c0 * 2) will be very
+  // close to the original value.
+  if (params[1] > 0.0f && params[2] < 1.0f) {
+    return params[2] - (params[0] * 2.0f);
+  }
+
+  float c0 = params[0];
+  float c3 = params[3];
+  // FIXME: This handling is not correct, but is distinct without crashing.
+  // It does not appear possible for a DirectX-generated value to be positive,
+  // so while this differs from hardware behavior, it may be irrelevant in
+  // practice.
+  if (c0 > 0.0f || c3 > 0.0f) {
+    return 0.0001f;
+  }
+
+  float reconstructed_power = 0.f;
+  for (uint32_t i = 0; i < sizeof(kCoefficient0StepPoints) / sizeof(kCoefficient0StepPoints[0]); ++i) {
+    if (c0 > kCoefficient0StepPoints[i]) {
+      reconstructed_power = reconstruct_func_map[i](c0, &curve_coefficients[i]);
+      break;
+    }
+  }
+
+  float reconstructed_half_power = 0.f;
+  for (uint32_t i = 0; i < sizeof(kCoefficient0StepPoints) / sizeof(kCoefficient0StepPoints[0]); ++i) {
+    if (c3 > kCoefficient0StepPoints[i]) {
+      reconstructed_half_power = reconstruct_func_map[i](c3, &curve_coefficients[i]);
+      break;
+    }
+  }
+
+  // The range can be extended beyond 1250 by using the half power params. This
+  // will only work for DirectX generated values, arbitrary params could
+  // erroneously trigger this.
+  //
+  // There are some very low power (~1) values that have inverted powers, but
+  // they are easily identified by comparatively high c0 parameters.
+  if (reconstructed_power == 0.f || (reconstructed_half_power > reconstructed_power && c0 < -0.1f)) {
+    return reconstructed_half_power * 2.f;
+  }
+
+  return reconstructed_power;
+}
+
+DEF_METHOD_INC(NV097, SET_SPECULAR_PARAMS)
+{
+    int slot = (method - NV097_SET_SPECULAR_PARAMS) / 4;
+    pg->specular_params[slot] = *(float *)&parameter;
+    if (slot == 5) {
+        pg->specular_power = reconstruct_specular_power(pg->specular_params);
+    }
+}
+
 DEF_METHOD_INC(NV097, SET_SCENE_AMBIENT_COLOR)
 {
     int slot = (method - NV097_SET_SCENE_AMBIENT_COLOR) / 4;
@@ -1843,6 +1968,12 @@ DEF_METHOD_INC(NV097, SET_COMBINER_COLOR_ICW)
     pgraph_reg_w(pg, NV_PGRAPH_COMBINECOLORI0 + slot*4, parameter);
 }
 
+DEF_METHOD_INC(NV097, SET_COLOR_KEY_COLOR)
+{
+    int slot = (method - NV097_SET_COLOR_KEY_COLOR) / 4;
+    pgraph_reg_w(pg, NV_PGRAPH_COLORKEYCOLOR0 + slot * 4, parameter);
+}
+
 DEF_METHOD_INC(NV097, SET_VIEWPORT_SCALE)
 {
     int slot = (method - NV097_SET_VIEWPORT_SCALE) / 4;
@@ -2010,6 +2141,26 @@ DEF_METHOD_INC(NV097, SET_VERTEX4F)
     }
 }
 
+DEF_METHOD(NV097, SET_FOG_COORD)
+{
+    VertexAttribute *attribute = &pg->vertex_attributes[NV2A_VERTEX_ATTR_FOG];
+    pgraph_allocate_inline_buffer_vertices(pg, NV2A_VERTEX_ATTR_FOG);
+    attribute->inline_value[0] = *(float*)&parameter;
+    attribute->inline_value[1] = attribute->inline_value[0];
+    attribute->inline_value[2] = attribute->inline_value[0];
+    attribute->inline_value[3] = attribute->inline_value[0];
+}
+
+DEF_METHOD(NV097, SET_WEIGHT1F)
+{
+    VertexAttribute *attribute = &pg->vertex_attributes[NV2A_VERTEX_ATTR_WEIGHT];
+    pgraph_allocate_inline_buffer_vertices(pg, NV2A_VERTEX_ATTR_WEIGHT);
+    attribute->inline_value[0] = *(float*)&parameter;
+    attribute->inline_value[1] = 0.f;
+    attribute->inline_value[2] = 0.f;
+    attribute->inline_value[3] = 1.f;
+}
+
 DEF_METHOD_INC(NV097, SET_NORMAL3S)
 {
     int slot = (method - NV097_SET_NORMAL3S) / 4;
@@ -2144,7 +2295,6 @@ DEF_METHOD_INC(NV097, SET_TEXCOORD1_4F)
     SET_VERTEX_ATTRIBUTE_F(NV097_SET_TEXCOORD1_4F, NV2A_VERTEX_ATTR_TEXTURE1);
 }
 
-
 DEF_METHOD_INC(NV097, SET_TEXCOORD2_4F)
 {
     SET_VERTEX_ATTRIBUTE_F(NV097_SET_TEXCOORD2_4F, NV2A_VERTEX_ATTR_TEXTURE2);
@@ -2155,8 +2305,34 @@ DEF_METHOD_INC(NV097, SET_TEXCOORD3_4F)
     SET_VERTEX_ATTRIBUTE_F(NV097_SET_TEXCOORD3_4F, NV2A_VERTEX_ATTR_TEXTURE3);
 }
 
+DEF_METHOD_INC(NV097, SET_WEIGHT4F)
+{
+    SET_VERTEX_ATTRIBUTE_F(NV097_SET_WEIGHT4F, NV2A_VERTEX_ATTR_WEIGHT);
+}
+
 #undef SET_VERTEX_ATTRIBUTE_F
 
+DEF_METHOD_INC(NV097, SET_WEIGHT2F)
+{
+    int slot = (method - NV097_SET_WEIGHT2F) / 4;
+    VertexAttribute *attribute =
+        &pg->vertex_attributes[NV2A_VERTEX_ATTR_WEIGHT];
+    pgraph_allocate_inline_buffer_vertices(pg, NV2A_VERTEX_ATTR_WEIGHT);
+    attribute->inline_value[slot] = *(float*)&parameter;
+    attribute->inline_value[2] = 0.0f;
+    attribute->inline_value[3] = 1.0f;
+}
+
+DEF_METHOD_INC(NV097, SET_WEIGHT3F)
+{
+    int slot = (method - NV097_SET_WEIGHT3F) / 4;
+    VertexAttribute *attribute =
+        &pg->vertex_attributes[NV2A_VERTEX_ATTR_WEIGHT];
+    pgraph_allocate_inline_buffer_vertices(pg, NV2A_VERTEX_ATTR_WEIGHT);
+    attribute->inline_value[slot] = *(float*)&parameter;
+    attribute->inline_value[3] = 1.0f;
+}
+
 #define SET_VERTEX_ATRIBUTE_TEX_2F(command, attr_index)                    \
     do {                                                                   \
         int slot = (method - (command)) / 4;                               \
@@ -2526,7 +2702,11 @@ DEF_METHOD(NV097, DRAW_ARRAYS)
     int32_t count = GET_MASK(parameter, NV097_DRAW_ARRAYS_COUNT) + 1;
 
     if (pg->inline_elements_length) {
-        /* FIXME: Determine HW behavior for overflow case. */
+        /* FIXME: HW throws an exception if the start index is > 0xFFFF. This
+         * would prevent this assert from firing for any reasonable choice of
+         * NV2A_MAX_BATCH_LENGTH (which must be larger to accommodate
+         * NV097_INLINE_ARRAY anyway)
+         */
         assert((pg->inline_elements_length + count) < NV2A_MAX_BATCH_LENGTH);
         assert(!pg->draw_arrays_prevent_connect);
 
@@ -2728,6 +2908,15 @@ DEF_METHOD_INC(NV097, SET_SPECULAR_FOG_FACTOR)
     pgraph_reg_w(pg, NV_PGRAPH_SPECFOGFACTOR0 + slot*4, parameter);
 }
 
+DEF_METHOD_INC(NV097, SET_SPECULAR_PARAMS_BACK)
+{
+    int slot = (method - NV097_SET_SPECULAR_PARAMS_BACK) / 4;
+    pg->specular_params_back[slot] = *(float *)&parameter;
+    if (slot == 5) {
+        pg->specular_power_back = reconstruct_specular_power(pg->specular_params_back);
+    }
+}
+
 DEF_METHOD(NV097, SET_SHADER_CLIP_PLANE_MODE)
 {
     pgraph_reg_w(pg, NV_PGRAPH_SHADERCLIPMODE, parameter);
@@ -2958,12 +3147,31 @@ void pgraph_write_zpass_pixel_cnt_report(NV2AState *d, uint32_t parameter,
     NV2A_DPRINTF("Report result %d @%" HWADDR_PRIx, result, offset);
 }
 
+static void do_wait_for_renderer_switch(CPUState *cpu, run_on_cpu_data data)
+{
+    NV2AState *d = (NV2AState *)data.host_ptr;
+
+    qemu_mutex_lock(&d->pfifo.lock);
+    d->pgraph.renderer_switch_phase = PGRAPH_RENDERER_SWITCH_PHASE_CPU_WAITING;
+    pfifo_kick(d);
+    qemu_mutex_unlock(&d->pfifo.lock);
+    qemu_event_wait(&d->pgraph.renderer_switch_complete);
+}
+
 void pgraph_process_pending(NV2AState *d)
 {
     PGRAPHState *pg = &d->pgraph;
     pg->renderer->ops.process_pending(d);
 
-    if (g_config.display.renderer != pg->renderer->type) {
+    if (g_config.display.renderer != pg->renderer->type &&
+        pg->renderer_switch_phase == PGRAPH_RENDERER_SWITCH_PHASE_IDLE) {
+        pg->renderer_switch_phase = PGRAPH_RENDERER_SWITCH_PHASE_STARTED;
+        qemu_event_reset(&pg->renderer_switch_complete);
+        async_safe_run_on_cpu(qemu_get_cpu(0), do_wait_for_renderer_switch,
+                              RUN_ON_CPU_HOST_PTR(d));
+    }
+
+    if (pg->renderer_switch_phase == PGRAPH_RENDERER_SWITCH_PHASE_CPU_WAITING) {
         qemu_mutex_lock(&d->pgraph.renderer_lock);
         qemu_mutex_unlock(&d->pfifo.lock);
         qemu_mutex_lock(&d->pgraph.lock);
@@ -2975,14 +3183,13 @@ void pgraph_process_pending(NV2AState *d)
             qemu_mutex_lock(&d->pfifo.lock);
             qemu_mutex_unlock(&d->pgraph.lock);
 
-            if (pg->renderer->ops.process_pending) {
-                pg->renderer->ops.process_pending(d);
-            }
+            pg->renderer->ops.process_pending(d);
 
             qemu_mutex_unlock(&d->pfifo.lock);
             qemu_mutex_lock(&d->pgraph.lock);
             while (pg->framebuffer_in_use) {
-                qemu_cond_wait(&d->pgraph.framebuffer_released, &d->pgraph.renderer_lock);
+                qemu_cond_wait(&d->pgraph.framebuffer_released,
+                               &d->pgraph.renderer_lock);
             }
 
             if (pg->renderer->ops.finalize) {
@@ -2995,6 +3202,9 @@ void pgraph_process_pending(NV2AState *d)
         qemu_mutex_unlock(&d->pgraph.renderer_lock);
         qemu_mutex_unlock(&d->pgraph.lock);
         qemu_mutex_lock(&d->pfifo.lock);
+
+        pg->renderer_switch_phase = PGRAPH_RENDERER_SWITCH_PHASE_IDLE;
+        qemu_event_set(&pg->renderer_switch_complete);
     }
 }
 

hw/xbox/nv2a/pgraph/pgraph.h

@@ -29,9 +29,10 @@
 #include "qemu/thread.h"
 #include "cpu.h"
 
-#include "shaders.h"
 #include "surface.h"
+#include "texture.h"
 #include "util.h"
+#include "vsh_regs.h"
 
 typedef struct NV2AState NV2AState;
 typedef struct PGRAPHNullState PGRAPHNullState;
@@ -197,6 +198,11 @@ typedef struct PGRAPHState {
     float light_local_position[NV2A_MAX_LIGHTS][3];
     float light_local_attenuation[NV2A_MAX_LIGHTS][3];
 
+    float specular_params[6];
+    float specular_power;
+    float specular_params_back[6];
+    float specular_power_back;
+
     float point_params[8];
 
     VertexAttribute vertex_attributes[NV2A_VERTEXSHADER_ATTRIBUTES];
@@ -238,6 +244,13 @@ typedef struct PGRAPHState {
     bool framebuffer_in_use;
     QemuCond framebuffer_released;
 
+    enum {
+        PGRAPH_RENDERER_SWITCH_PHASE_IDLE,
+        PGRAPH_RENDERER_SWITCH_PHASE_STARTED,
+        PGRAPH_RENDERER_SWITCH_PHASE_CPU_WAITING,
+    } renderer_switch_phase;
+    QemuEvent renderer_switch_complete;
+
     unsigned int surface_scale_factor;
     uint8_t *scale_buf;
 

hw/xbox/nv2a/pgraph/psh.h

@@ -1,92 +0,0 @@
-/*
- * QEMU Geforce NV2A pixel shader translation
- *
- * Copyright (c) 2013 espes
- *
- * This library is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation; either
- * version 2 of the License, or (at your option) any later version.
- *
- * This library is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * Lesser General Public License for more details.
- *
- * You should have received a copy of the GNU Lesser General Public
- * License along with this library; if not, see <http://www.gnu.org/licenses/>.
- */
-
-#ifndef HW_NV2A_PSH_H
-#define HW_NV2A_PSH_H
-
-#include <stdint.h>
-#include <stdbool.h>
-
-enum PshAlphaFunc {
-    ALPHA_FUNC_NEVER,
-    ALPHA_FUNC_LESS,
-    ALPHA_FUNC_EQUAL,
-    ALPHA_FUNC_LEQUAL,
-    ALPHA_FUNC_GREATER,
-    ALPHA_FUNC_NOTEQUAL,
-    ALPHA_FUNC_GEQUAL,
-    ALPHA_FUNC_ALWAYS,
-};
-
-enum PshShadowDepthFunc {
-    SHADOW_DEPTH_FUNC_NEVER,
-    SHADOW_DEPTH_FUNC_LESS,
-    SHADOW_DEPTH_FUNC_EQUAL,
-    SHADOW_DEPTH_FUNC_LEQUAL,
-    SHADOW_DEPTH_FUNC_GREATER,
-    SHADOW_DEPTH_FUNC_NOTEQUAL,
-    SHADOW_DEPTH_FUNC_GEQUAL,
-    SHADOW_DEPTH_FUNC_ALWAYS,
-};
-
-enum ConvolutionFilter {
-    CONVOLUTION_FILTER_DISABLED,
-    CONVOLUTION_FILTER_QUINCUNX,
-    CONVOLUTION_FILTER_GAUSSIAN,
-};
-
-typedef struct PshState {
-    bool vulkan;
-
-    /* fragment shader - register combiner stuff */
-    uint32_t combiner_control;
-    uint32_t shader_stage_program;
-    uint32_t other_stage_input;
-    uint32_t final_inputs_0;
-    uint32_t final_inputs_1;
-
-    uint32_t rgb_inputs[8], rgb_outputs[8];
-    uint32_t alpha_inputs[8], alpha_outputs[8];
-
-    bool point_sprite;
-    bool rect_tex[4];
-    bool snorm_tex[4];
-    bool compare_mode[4][4];
-    bool alphakill[4];
-    enum ConvolutionFilter conv_tex[4];
-    bool tex_x8y24[4];
-    int dim_tex[4];
-
-    float border_logical_size[4][3];
-    float border_inv_real_size[4][3];
-
-    bool shadow_map[4];
-    enum PshShadowDepthFunc shadow_depth_func;
-
-    bool alpha_test;
-    enum PshAlphaFunc alpha_func;
-
-    bool window_clip_exclusive;
-
-    bool smooth_shading;
-    bool depth_clipping;
-    bool z_perspective;
-} PshState;
-
-#endif

hw/xbox/nv2a/pgraph/psh_regs.h

@@ -0,0 +1,190 @@
+/*
+ * QEMU Geforce NV2A pixel shader translation
+ *
+ * Copyright (c) 2013 espes
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef HW_NV2A_PSH_H
+#define HW_NV2A_PSH_H
+
+#include <stdint.h>
+#include <stdbool.h>
+
+/*
+ * For some background, see the OpenGL extension:
+ * https://www.opengl.org/registry/specs/NV/register_combiners.txt
+ */
+
+enum PS_TEXTUREMODES
+{                                 // valid in stage 0 1 2 3
+    PS_TEXTUREMODES_NONE=                 0x00L, // * * * *
+    PS_TEXTUREMODES_PROJECT2D=            0x01L, // * * * *
+    PS_TEXTUREMODES_PROJECT3D=            0x02L, // * * * *
+    PS_TEXTUREMODES_CUBEMAP=              0x03L, // * * * *
+    PS_TEXTUREMODES_PASSTHRU=             0x04L, // * * * *
+    PS_TEXTUREMODES_CLIPPLANE=            0x05L, // * * * *
+    PS_TEXTUREMODES_BUMPENVMAP=           0x06L, // - * * *
+    PS_TEXTUREMODES_BUMPENVMAP_LUM=       0x07L, // - * * *
+    PS_TEXTUREMODES_BRDF=                 0x08L, // - - * *
+    PS_TEXTUREMODES_DOT_ST=               0x09L, // - - * *
+    PS_TEXTUREMODES_DOT_ZW=               0x0aL, // - - * *
+    PS_TEXTUREMODES_DOT_RFLCT_DIFF=       0x0bL, // - - * -
+    PS_TEXTUREMODES_DOT_RFLCT_SPEC=       0x0cL, // - - - *
+    PS_TEXTUREMODES_DOT_STR_3D=           0x0dL, // - - - *
+    PS_TEXTUREMODES_DOT_STR_CUBE=         0x0eL, // - - - *
+    PS_TEXTUREMODES_DPNDNT_AR=            0x0fL, // - * * *
+    PS_TEXTUREMODES_DPNDNT_GB=            0x10L, // - * * *
+    PS_TEXTUREMODES_DOTPRODUCT=           0x11L, // - * * -
+    PS_TEXTUREMODES_DOT_RFLCT_SPEC_CONST= 0x12L, // - - - *
+    // 0x13-0x1f reserved
+};
+
+enum PS_INPUTMAPPING
+{
+    PS_INPUTMAPPING_UNSIGNED_IDENTITY= 0x00L, // max(0,x)         OK for final combiner
+    PS_INPUTMAPPING_UNSIGNED_INVERT=   0x20L, // 1 - max(0,x)     OK for final combiner
+    PS_INPUTMAPPING_EXPAND_NORMAL=     0x40L, // 2*max(0,x) - 1   invalid for final combiner
+    PS_INPUTMAPPING_EXPAND_NEGATE=     0x60L, // 1 - 2*max(0,x)   invalid for final combiner
+    PS_INPUTMAPPING_HALFBIAS_NORMAL=   0x80L, // max(0,x) - 1/2   invalid for final combiner
+    PS_INPUTMAPPING_HALFBIAS_NEGATE=   0xa0L, // 1/2 - max(0,x)   invalid for final combiner
+    PS_INPUTMAPPING_SIGNED_IDENTITY=   0xc0L, // x                invalid for final combiner
+    PS_INPUTMAPPING_SIGNED_NEGATE=     0xe0L, // -x               invalid for final combiner
+};
+
+enum PS_REGISTER
+{
+    PS_REGISTER_ZERO=              0x00L, // r
+    PS_REGISTER_DISCARD=           0x00L, // w
+    PS_REGISTER_C0=                0x01L, // r
+    PS_REGISTER_C1=                0x02L, // r
+    PS_REGISTER_FOG=               0x03L, // r
+    PS_REGISTER_V0=                0x04L, // r/w
+    PS_REGISTER_V1=                0x05L, // r/w
+    PS_REGISTER_T0=                0x08L, // r/w
+    PS_REGISTER_T1=                0x09L, // r/w
+    PS_REGISTER_T2=                0x0aL, // r/w
+    PS_REGISTER_T3=                0x0bL, // r/w
+    PS_REGISTER_R0=                0x0cL, // r/w
+    PS_REGISTER_R1=                0x0dL, // r/w
+    PS_REGISTER_V1R0_SUM=          0x0eL, // r
+    PS_REGISTER_EF_PROD=           0x0fL, // r
+
+    PS_REGISTER_ONE=               PS_REGISTER_ZERO | PS_INPUTMAPPING_UNSIGNED_INVERT, // OK for final combiner
+    PS_REGISTER_NEGATIVE_ONE=      PS_REGISTER_ZERO | PS_INPUTMAPPING_EXPAND_NORMAL,   // invalid for final combiner
+    PS_REGISTER_ONE_HALF=          PS_REGISTER_ZERO | PS_INPUTMAPPING_HALFBIAS_NEGATE, // invalid for final combiner
+    PS_REGISTER_NEGATIVE_ONE_HALF= PS_REGISTER_ZERO | PS_INPUTMAPPING_HALFBIAS_NORMAL, // invalid for final combiner
+};
+
+enum PS_COMBINERCOUNTFLAGS
+{
+    PS_COMBINERCOUNT_MUX_LSB=     0x0000L, // mux on r0.a lsb
+    PS_COMBINERCOUNT_MUX_MSB=     0x0001L, // mux on r0.a msb
+
+    PS_COMBINERCOUNT_SAME_C0=     0x0000L, // c0 same in each stage
+    PS_COMBINERCOUNT_UNIQUE_C0=   0x0010L, // c0 unique in each stage
+
+    PS_COMBINERCOUNT_SAME_C1=     0x0000L, // c1 same in each stage
+    PS_COMBINERCOUNT_UNIQUE_C1=   0x0100L  // c1 unique in each stage
+};
+
+enum PS_COMBINEROUTPUT
+{
+    PS_COMBINEROUTPUT_IDENTITY=            0x00L, // y = x
+    PS_COMBINEROUTPUT_BIAS=                0x08L, // y = x - 0.5
+    PS_COMBINEROUTPUT_SHIFTLEFT_1=         0x10L, // y = x*2
+    PS_COMBINEROUTPUT_SHIFTLEFT_1_BIAS=    0x18L, // y = (x - 0.5)*2
+    PS_COMBINEROUTPUT_SHIFTLEFT_2=         0x20L, // y = x*4
+    PS_COMBINEROUTPUT_SHIFTRIGHT_1=        0x30L, // y = x/2
+
+    PS_COMBINEROUTPUT_AB_BLUE_TO_ALPHA=    0x80L, // RGB only
+
+    PS_COMBINEROUTPUT_CD_BLUE_TO_ALPHA=    0x40L, // RGB only
+
+    PS_COMBINEROUTPUT_AB_MULTIPLY=         0x00L,
+    PS_COMBINEROUTPUT_AB_DOT_PRODUCT=      0x02L, // RGB only
+
+    PS_COMBINEROUTPUT_CD_MULTIPLY=         0x00L,
+    PS_COMBINEROUTPUT_CD_DOT_PRODUCT=      0x01L, // RGB only
+
+    PS_COMBINEROUTPUT_AB_CD_SUM=           0x00L, // 3rd output is AB+CD
+    PS_COMBINEROUTPUT_AB_CD_MUX=           0x04L, // 3rd output is MUX(AB,CD) based on R0.a
+};
+
+enum PS_CHANNEL
+{
+    PS_CHANNEL_RGB=   0x00, // used as RGB source
+    PS_CHANNEL_BLUE=  0x00, // used as ALPHA source
+    PS_CHANNEL_ALPHA= 0x10, // used as RGB or ALPHA source
+};
+
+
+enum PS_FINALCOMBINERSETTING
+{
+    PS_FINALCOMBINERSETTING_CLAMP_SUM=     0x80, // V1+R0 sum clamped to [0,1]
+
+    PS_FINALCOMBINERSETTING_COMPLEMENT_V1= 0x40, // unsigned invert mapping
+
+    PS_FINALCOMBINERSETTING_COMPLEMENT_R0= 0x20, // unsigned invert mapping
+};
+
+enum PS_DOTMAPPING
+{                              // valid in stage 0 1 2 3
+    PS_DOTMAPPING_ZERO_TO_ONE=         0x00L, // - * * *
+    PS_DOTMAPPING_MINUS1_TO_1_D3D=     0x01L, // - * * *
+    PS_DOTMAPPING_MINUS1_TO_1_GL=      0x02L, // - * * *
+    PS_DOTMAPPING_MINUS1_TO_1=         0x03L, // - * * *
+    PS_DOTMAPPING_HILO_1=              0x04L, // - * * *
+    PS_DOTMAPPING_HILO_HEMISPHERE_D3D= 0x05L, // - * * *
+    PS_DOTMAPPING_HILO_HEMISPHERE_GL=  0x06L, // - * * *
+    PS_DOTMAPPING_HILO_HEMISPHERE=     0x07L, // - * * *
+};
+
+enum PS_COLORKEYMODE {
+    COLOR_KEY_NONE = 0,
+    COLOR_KEY_KILL_ALPHA = 1,
+    COLOR_KEY_KILL_COLOR_AND_ALPHA = 2,
+    COLOR_KEY_DISCARD = 3,
+};
+
+enum PshAlphaFunc {
+    ALPHA_FUNC_NEVER,
+    ALPHA_FUNC_LESS,
+    ALPHA_FUNC_EQUAL,
+    ALPHA_FUNC_LEQUAL,
+    ALPHA_FUNC_GREATER,
+    ALPHA_FUNC_NOTEQUAL,
+    ALPHA_FUNC_GEQUAL,
+    ALPHA_FUNC_ALWAYS,
+};
+
+enum PshShadowDepthFunc {
+    SHADOW_DEPTH_FUNC_NEVER,
+    SHADOW_DEPTH_FUNC_LESS,
+    SHADOW_DEPTH_FUNC_EQUAL,
+    SHADOW_DEPTH_FUNC_LEQUAL,
+    SHADOW_DEPTH_FUNC_GREATER,
+    SHADOW_DEPTH_FUNC_NOTEQUAL,
+    SHADOW_DEPTH_FUNC_GEQUAL,
+    SHADOW_DEPTH_FUNC_ALWAYS,
+};
+
+enum ConvolutionFilter {
+    CONVOLUTION_FILTER_DISABLED,
+    CONVOLUTION_FILTER_QUINCUNX,
+    CONVOLUTION_FILTER_GAUSSIAN,
+};
+
+#endif

hw/xbox/nv2a/pgraph/shaders.c

@@ -1,302 +0,0 @@
-/*
- * Geforce NV2A PGRAPH OpenGL Renderer
- *
- * Copyright (c) 2015 espes
- * Copyright (c) 2015 Jannik Vogel
- * Copyright (c) 2020-2024 Matt Borgerson
- *
- * This library is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation; either
- * version 2 of the License, or (at your option) any later version.
- *
- * This library is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * Lesser General Public License for more details.
- *
- * You should have received a copy of the GNU Lesser General Public
- * License along with this library; if not, see <http://www.gnu.org/licenses/>.
- */
-
-#include "hw/xbox/nv2a/debug.h"
-#include "texture.h"
-#include "pgraph.h"
-#include "shaders.h"
-
-ShaderState pgraph_get_shader_state(PGRAPHState *pg)
-{
-    bool vertex_program = GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CSV0_D),
-                                   NV_PGRAPH_CSV0_D_MODE) == 2;
-
-    bool fixed_function = GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CSV0_D),
-                                   NV_PGRAPH_CSV0_D_MODE) == 0;
-
-    int program_start = GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CSV0_C),
-                                 NV_PGRAPH_CSV0_C_CHEOPS_PROGRAM_START);
-
-    pg->program_data_dirty = false;
-
-    ShaderState state;
-
-    // We will hash it, so make sure any padding is zeroed
-    memset(&state, 0, sizeof(ShaderState));
-
-    state.surface_scale_factor = pg->surface_scale_factor;
-
-    state.compressed_attrs = pg->compressed_attrs;
-    state.uniform_attrs = pg->uniform_attrs;
-    state.swizzle_attrs = pg->swizzle_attrs;
-
-    /* register combiner stuff */
-    state.psh.window_clip_exclusive =
-        pgraph_reg_r(pg, NV_PGRAPH_SETUPRASTER) & NV_PGRAPH_SETUPRASTER_WINDOWCLIPTYPE;
-    state.psh.combiner_control = pgraph_reg_r(pg, NV_PGRAPH_COMBINECTL);
-    state.psh.shader_stage_program = pgraph_reg_r(pg, NV_PGRAPH_SHADERPROG);
-    state.psh.other_stage_input = pgraph_reg_r(pg, NV_PGRAPH_SHADERCTL);
-    state.psh.final_inputs_0 = pgraph_reg_r(pg, NV_PGRAPH_COMBINESPECFOG0);
-    state.psh.final_inputs_1 = pgraph_reg_r(pg, NV_PGRAPH_COMBINESPECFOG1);
-
-    state.psh.alpha_test =
-        pgraph_reg_r(pg, NV_PGRAPH_CONTROL_0) & NV_PGRAPH_CONTROL_0_ALPHATESTENABLE;
-    state.psh.alpha_func = (enum PshAlphaFunc)GET_MASK(
-        pgraph_reg_r(pg, NV_PGRAPH_CONTROL_0), NV_PGRAPH_CONTROL_0_ALPHAFUNC);
-
-    state.psh.point_sprite = pgraph_reg_r(pg, NV_PGRAPH_SETUPRASTER) &
-                             NV_PGRAPH_SETUPRASTER_POINTSMOOTHENABLE;
-
-    state.psh.shadow_depth_func = (enum PshShadowDepthFunc)GET_MASK(
-        pgraph_reg_r(pg, NV_PGRAPH_SHADOWCTL), NV_PGRAPH_SHADOWCTL_SHADOW_ZFUNC);
-
-    state.fixed_function = fixed_function;
-    state.specular_enable = GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CSV0_C),
-                                     NV_PGRAPH_CSV0_C_SPECULAR_ENABLE);
-
-    /* fixed function stuff */
-    if (fixed_function) {
-        state.skinning = (enum VshSkinning)GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CSV0_D),
-                                                    NV_PGRAPH_CSV0_D_SKIN);
-        state.lighting =
-            GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CSV0_C), NV_PGRAPH_CSV0_C_LIGHTING);
-        state.normalization =
-            pgraph_reg_r(pg, NV_PGRAPH_CSV0_C) & NV_PGRAPH_CSV0_C_NORMALIZATION_ENABLE;
-
-        /* color material */
-        state.emission_src = (enum MaterialColorSource)GET_MASK(
-            pgraph_reg_r(pg, NV_PGRAPH_CSV0_C), NV_PGRAPH_CSV0_C_EMISSION);
-        state.ambient_src = (enum MaterialColorSource)GET_MASK(
-            pgraph_reg_r(pg, NV_PGRAPH_CSV0_C), NV_PGRAPH_CSV0_C_AMBIENT);
-        state.diffuse_src = (enum MaterialColorSource)GET_MASK(
-            pgraph_reg_r(pg, NV_PGRAPH_CSV0_C), NV_PGRAPH_CSV0_C_DIFFUSE);
-        state.specular_src = (enum MaterialColorSource)GET_MASK(
-            pgraph_reg_r(pg, NV_PGRAPH_CSV0_C), NV_PGRAPH_CSV0_C_SPECULAR);
-    }
-
-    /* vertex program stuff */
-    state.vertex_program = vertex_program,
-    state.z_perspective = pgraph_reg_r(pg, NV_PGRAPH_CONTROL_0) &
-                          NV_PGRAPH_CONTROL_0_Z_PERSPECTIVE_ENABLE;
-    state.psh.z_perspective = state.z_perspective;
-
-    state.point_params_enable = GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CSV0_D),
-                                         NV_PGRAPH_CSV0_D_POINTPARAMSENABLE);
-    state.point_size =
-        GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_POINTSIZE), NV097_SET_POINT_SIZE_V) / 8.0f;
-    if (state.point_params_enable) {
-        for (int i = 0; i < 8; i++) {
-            state.point_params[i] = pg->point_params[i];
-        }
-    }
-
-    /* geometry shader stuff */
-    state.primitive_mode = (enum ShaderPrimitiveMode)pg->primitive_mode;
-    state.polygon_front_mode = (enum ShaderPolygonMode)GET_MASK(
-        pgraph_reg_r(pg, NV_PGRAPH_SETUPRASTER), NV_PGRAPH_SETUPRASTER_FRONTFACEMODE);
-    state.polygon_back_mode = (enum ShaderPolygonMode)GET_MASK(
-        pgraph_reg_r(pg, NV_PGRAPH_SETUPRASTER), NV_PGRAPH_SETUPRASTER_BACKFACEMODE);
-
-    state.smooth_shading = GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CONTROL_3),
-                                    NV_PGRAPH_CONTROL_3_SHADEMODE) ==
-                           NV_PGRAPH_CONTROL_3_SHADEMODE_SMOOTH;
-    state.psh.smooth_shading = state.smooth_shading;
-
-    state.psh.depth_clipping = GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_ZCOMPRESSOCCLUDE),
-                                        NV_PGRAPH_ZCOMPRESSOCCLUDE_ZCLAMP_EN) ==
-                               NV_PGRAPH_ZCOMPRESSOCCLUDE_ZCLAMP_EN_CULL;
-
-    state.program_length = 0;
-
-    if (vertex_program) {
-        // copy in vertex program tokens
-        for (int i = program_start; i < NV2A_MAX_TRANSFORM_PROGRAM_LENGTH;
-             i++) {
-            uint32_t *cur_token = (uint32_t *)&pg->program_data[i];
-            memcpy(&state.program_data[state.program_length], cur_token,
-                   VSH_TOKEN_SIZE * sizeof(uint32_t));
-            state.program_length++;
-
-            if (vsh_get_field(cur_token, FLD_FINAL)) {
-                break;
-            }
-        }
-    }
-
-    /* Texgen */
-    for (int i = 0; i < 4; i++) {
-        unsigned int reg = (i < 2) ? NV_PGRAPH_CSV1_A : NV_PGRAPH_CSV1_B;
-        for (int j = 0; j < 4; j++) {
-            unsigned int masks[] = {
-                (i % 2) ? NV_PGRAPH_CSV1_A_T1_S : NV_PGRAPH_CSV1_A_T0_S,
-                (i % 2) ? NV_PGRAPH_CSV1_A_T1_T : NV_PGRAPH_CSV1_A_T0_T,
-                (i % 2) ? NV_PGRAPH_CSV1_A_T1_R : NV_PGRAPH_CSV1_A_T0_R,
-                (i % 2) ? NV_PGRAPH_CSV1_A_T1_Q : NV_PGRAPH_CSV1_A_T0_Q
-            };
-            state.texgen[i][j] =
-                (enum VshTexgen)GET_MASK(pgraph_reg_r(pg, reg), masks[j]);
-        }
-    }
-
-    /* Fog */
-    state.fog_enable =
-        pgraph_reg_r(pg, NV_PGRAPH_CONTROL_3) & NV_PGRAPH_CONTROL_3_FOGENABLE;
-    if (state.fog_enable) {
-        /*FIXME: Use CSV0_D? */
-        state.fog_mode = (enum VshFogMode)GET_MASK(
-            pgraph_reg_r(pg, NV_PGRAPH_CONTROL_3), NV_PGRAPH_CONTROL_3_FOG_MODE);
-        state.foggen = (enum VshFoggen)GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CSV0_D),
-                                                NV_PGRAPH_CSV0_D_FOGGENMODE);
-    } else {
-        /* FIXME: Do we still pass the fogmode? */
-        state.fog_mode = (enum VshFogMode)0;
-        state.foggen = (enum VshFoggen)0;
-    }
-
-    /* Texture matrices */
-    for (int i = 0; i < 4; i++) {
-        state.texture_matrix_enable[i] = pg->texture_matrix_enable[i];
-    }
-
-    /* Lighting */
-    if (state.lighting) {
-        for (int i = 0; i < NV2A_MAX_LIGHTS; i++) {
-            state.light[i] = (enum VshLight)GET_MASK(
-                pgraph_reg_r(pg, NV_PGRAPH_CSV0_D), NV_PGRAPH_CSV0_D_LIGHT0 << (i * 2));
-        }
-    }
-
-    /* Copy content of enabled combiner stages */
-    int num_stages = pgraph_reg_r(pg, NV_PGRAPH_COMBINECTL) & 0xFF;
-    for (int i = 0; i < num_stages; i++) {
-        state.psh.rgb_inputs[i] = pgraph_reg_r(pg, NV_PGRAPH_COMBINECOLORI0 + i * 4);
-        state.psh.rgb_outputs[i] = pgraph_reg_r(pg, NV_PGRAPH_COMBINECOLORO0 + i * 4);
-        state.psh.alpha_inputs[i] = pgraph_reg_r(pg, NV_PGRAPH_COMBINEALPHAI0 + i * 4);
-        state.psh.alpha_outputs[i] = pgraph_reg_r(pg, NV_PGRAPH_COMBINEALPHAO0 + i * 4);
-        // constant_0[i] = pgraph_reg_r(pg, NV_PGRAPH_COMBINEFACTOR0 + i * 4);
-        // constant_1[i] = pgraph_reg_r(pg, NV_PGRAPH_COMBINEFACTOR1 + i * 4);
-    }
-
-    for (int i = 0; i < 4; i++) {
-        for (int j = 0; j < 4; j++) {
-            state.psh.compare_mode[i][j] =
-                (pgraph_reg_r(pg, NV_PGRAPH_SHADERCLIPMODE) >> (4 * i + j)) & 1;
-        }
-
-        uint32_t ctl_0 = pgraph_reg_r(pg, NV_PGRAPH_TEXCTL0_0 + i * 4);
-        bool enabled = pgraph_is_texture_stage_active(pg, i) &&
-                       (ctl_0 & NV_PGRAPH_TEXCTL0_0_ENABLE);
-        if (!enabled) {
-            continue;
-        }
-
-        state.psh.alphakill[i] = ctl_0 & NV_PGRAPH_TEXCTL0_0_ALPHAKILLEN;
-
-        uint32_t tex_fmt = pgraph_reg_r(pg, NV_PGRAPH_TEXFMT0 + i * 4);
-        state.psh.dim_tex[i] = GET_MASK(tex_fmt, NV_PGRAPH_TEXFMT0_DIMENSIONALITY);
-
-        unsigned int color_format = GET_MASK(tex_fmt, NV_PGRAPH_TEXFMT0_COLOR);
-        BasicColorFormatInfo f = kelvin_color_format_info_map[color_format];
-        state.psh.rect_tex[i] = f.linear;
-        state.psh.tex_x8y24[i] = color_format == NV097_SET_TEXTURE_FORMAT_COLOR_LU_IMAGE_DEPTH_X8_Y24_FIXED ||
-                                color_format == NV097_SET_TEXTURE_FORMAT_COLOR_LU_IMAGE_DEPTH_X8_Y24_FLOAT;
-
-        uint32_t border_source =
-            GET_MASK(tex_fmt, NV_PGRAPH_TEXFMT0_BORDER_SOURCE);
-        bool cubemap = GET_MASK(tex_fmt, NV_PGRAPH_TEXFMT0_CUBEMAPENABLE);
-        state.psh.border_logical_size[i][0] = 0.0f;
-        state.psh.border_logical_size[i][1] = 0.0f;
-        state.psh.border_logical_size[i][2] = 0.0f;
-        if (border_source != NV_PGRAPH_TEXFMT0_BORDER_SOURCE_COLOR) {
-            if (!f.linear && !cubemap) {
-                // The actual texture will be (at least) double the reported
-                // size and shifted by a 4 texel border but texture coordinates
-                // will still be relative to the reported size.
-                unsigned int reported_width =
-                    1 << GET_MASK(tex_fmt, NV_PGRAPH_TEXFMT0_BASE_SIZE_U);
-                unsigned int reported_height =
-                    1 << GET_MASK(tex_fmt, NV_PGRAPH_TEXFMT0_BASE_SIZE_V);
-                unsigned int reported_depth =
-                    1 << GET_MASK(tex_fmt, NV_PGRAPH_TEXFMT0_BASE_SIZE_P);
-
-                state.psh.border_logical_size[i][0] = reported_width;
-                state.psh.border_logical_size[i][1] = reported_height;
-                state.psh.border_logical_size[i][2] = reported_depth;
-
-                if (reported_width < 8) {
-                    state.psh.border_inv_real_size[i][0] = 0.0625f;
-                } else {
-                    state.psh.border_inv_real_size[i][0] =
-                        1.0f / (reported_width * 2.0f);
-                }
-                if (reported_height < 8) {
-                    state.psh.border_inv_real_size[i][1] = 0.0625f;
-                } else {
-                    state.psh.border_inv_real_size[i][1] =
-                        1.0f / (reported_height * 2.0f);
-                }
-                if (reported_depth < 8) {
-                    state.psh.border_inv_real_size[i][2] = 0.0625f;
-                } else {
-                    state.psh.border_inv_real_size[i][2] =
-                        1.0f / (reported_depth * 2.0f);
-                }
-            } else {
-                NV2A_UNIMPLEMENTED(
-                    "Border source texture with linear %d cubemap %d", f.linear,
-                    cubemap);
-            }
-        }
-
-        /* Keep track of whether texture data has been loaded as signed
-         * normalized integers or not. This dictates whether or not we will need
-         * to re-map in fragment shader for certain texture modes (e.g.
-         * bumpenvmap).
-         *
-         * FIXME: When signed texture data is loaded as unsigned and remapped in
-         * fragment shader, there may be interpolation artifacts. Fix this to
-         * support signed textures more appropriately.
-         */
-#if 0 // FIXME
-        state.psh.snorm_tex[i] = (f.gl_internal_format == GL_RGB8_SNORM)
-                                 || (f.gl_internal_format == GL_RG8_SNORM);
-#endif
-        state.psh.shadow_map[i] = f.depth;
-
-        uint32_t filter = pgraph_reg_r(pg, NV_PGRAPH_TEXFILTER0 + i * 4);
-        unsigned int min_filter = GET_MASK(filter, NV_PGRAPH_TEXFILTER0_MIN);
-        enum ConvolutionFilter kernel = CONVOLUTION_FILTER_DISABLED;
-        /* FIXME: We do not distinguish between min and mag when
-         * performing convolution. Just use it if specified for min (common AA
-         * case).
-         */
-        if (min_filter == NV_PGRAPH_TEXFILTER0_MIN_CONVOLUTION_2D_LOD0) {
-            int k = GET_MASK(filter, NV_PGRAPH_TEXFILTER0_CONVOLUTION_KERNEL);
-            assert(k == NV_PGRAPH_TEXFILTER0_CONVOLUTION_KERNEL_QUINCUNX ||
-                   k == NV_PGRAPH_TEXFILTER0_CONVOLUTION_KERNEL_GAUSSIAN_3);
-            kernel = (enum ConvolutionFilter)k;
-        }
-
-        state.psh.conv_tex[i] = kernel;
-    }
-
-    return state;
-}

hw/xbox/nv2a/pgraph/shaders.h

@@ -1,110 +0,0 @@
-/*
- * QEMU Geforce NV2A shader generator
- *
- * Copyright (c) 2015 espes
- * Copyright (c) 2015 Jannik Vogel
- *
- * This library is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation; either
- * version 2 of the License, or (at your option) any later version.
- *
- * This library is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * Lesser General Public License for more details.
- *
- * You should have received a copy of the GNU Lesser General Public
- * License along with this library; if not, see <http://www.gnu.org/licenses/>.
- */
-
-#ifndef HW_XBOX_NV2A_PGRAPH_SHADERS_H
-#define HW_XBOX_NV2A_PGRAPH_SHADERS_H
-
-#include <stdint.h>
-#include "hw/xbox/nv2a/nv2a_regs.h"
-
-#include "vsh.h"
-#include "psh.h"
-
-enum ShaderPrimitiveMode {
-    PRIM_TYPE_INVALID,
-    PRIM_TYPE_POINTS,
-    PRIM_TYPE_LINES,
-    PRIM_TYPE_LINE_LOOP,
-    PRIM_TYPE_LINE_STRIP,
-    PRIM_TYPE_TRIANGLES,
-    PRIM_TYPE_TRIANGLE_STRIP,
-    PRIM_TYPE_TRIANGLE_FAN,
-    PRIM_TYPE_QUADS,
-    PRIM_TYPE_QUAD_STRIP,
-    PRIM_TYPE_POLYGON,
-};
-
-enum ShaderPolygonMode {
-    POLY_MODE_FILL,
-    POLY_MODE_POINT,
-    POLY_MODE_LINE,
-};
-
-enum MaterialColorSource {
-    MATERIAL_COLOR_SRC_MATERIAL,
-    MATERIAL_COLOR_SRC_DIFFUSE,
-    MATERIAL_COLOR_SRC_SPECULAR,
-};
-
-typedef struct ShaderState {
-    bool vulkan;
-    bool use_push_constants_for_uniform_attrs;
-    unsigned int surface_scale_factor;
-
-    PshState psh;
-    uint16_t compressed_attrs;
-    uint16_t uniform_attrs;
-    uint16_t swizzle_attrs;
-
-    bool texture_matrix_enable[4];
-    enum VshTexgen texgen[4][4];
-
-    bool fog_enable;
-    enum VshFoggen foggen;
-    enum VshFogMode fog_mode;
-
-    enum VshSkinning skinning;
-
-    bool normalization;
-
-    enum MaterialColorSource emission_src;
-    enum MaterialColorSource ambient_src;
-    enum MaterialColorSource diffuse_src;
-    enum MaterialColorSource specular_src;
-
-    bool lighting;
-    enum VshLight light[NV2A_MAX_LIGHTS];
-
-    bool fixed_function;
-    bool specular_enable;
-
-    /* vertex program */
-    bool vertex_program;
-    uint32_t program_data[NV2A_MAX_TRANSFORM_PROGRAM_LENGTH][VSH_TOKEN_SIZE];
-    int program_length;
-    bool z_perspective;
-
-    /* primitive format for geometry shader */
-    enum ShaderPolygonMode polygon_front_mode;
-    enum ShaderPolygonMode polygon_back_mode;
-    enum ShaderPrimitiveMode primitive_mode;
-
-    bool point_params_enable;
-    float point_size;
-    float point_params[8];
-
-    bool smooth_shading;
-} ShaderState;
-
-typedef struct PGRAPHState PGRAPHState;
-
-ShaderState pgraph_get_shader_state(PGRAPHState *pg);
-
-#endif

hw/xbox/nv2a/pgraph/vk/buffer.c

@@ -18,7 +18,6 @@
  */
 
 #include "renderer.h"
-#include <vulkan/vulkan_core.h>
 
 static void create_buffer(PGRAPHState *pg, StorageBuffer *buffer)
 {

hw/xbox/nv2a/pgraph/vk/constants.h

@@ -21,7 +21,7 @@
 #define HW_XBOX_NV2A_PGRAPH_VK_CONSTANTS_H
 
 #include "hw/xbox/nv2a/nv2a_regs.h"
-#include "hw/xbox/nv2a/pgraph/shaders.h"
+#include "hw/xbox/nv2a/pgraph/vsh_regs.h"
 #include <vulkan/vulkan.h>
 
 static const VkFilter pgraph_texture_min_filter_vk_map[] = {

hw/xbox/nv2a/pgraph/vk/debug.c

@@ -25,6 +25,8 @@
 #endif
 
 #ifdef CONFIG_RENDERDOC
+#include "trace/control.h"
+
 #pragma GCC diagnostic ignored "-Wstrict-prototypes"
 #include "thirdparty/renderdoc_app.h"
 #endif
@@ -46,11 +48,21 @@ void pgraph_vk_debug_frame_terminator(void)
 
         PGRAPHVkState *r = g_nv2a->pgraph.vk_renderer_state;
         if (rdoc_api->IsTargetControlConnected()) {
-            if (rdoc_api->IsFrameCapturing()) {
+            bool capturing = rdoc_api->IsFrameCapturing();
+            if (capturing && renderdoc_capture_frames == 0) {
                 rdoc_api->EndFrameCapture(RENDERDOC_DEVICEPOINTER_FROM_VKINSTANCE(r->instance), 0);
+                if (renderdoc_trace_frames) {
+                    trace_enable_events("-nv2a_pgraph_*");
+                    renderdoc_trace_frames = false;
+                }
             }
             if (renderdoc_capture_frames > 0) {
-                rdoc_api->StartFrameCapture(RENDERDOC_DEVICEPOINTER_FROM_VKINSTANCE(r->instance), 0);
+                if (!capturing) {
+                    if (renderdoc_trace_frames) {
+                        trace_enable_events("nv2a_pgraph_*");
+                    }
+                    rdoc_api->StartFrameCapture(RENDERDOC_DEVICEPOINTER_FROM_VKINSTANCE(r->instance), 0);
+                }
                 --renderdoc_capture_frames;
             }
         }

hw/xbox/nv2a/pgraph/vk/draw.c

@@ -1,7 +1,7 @@
 /*
  * Geforce NV2A PGRAPH Vulkan Renderer
  *
- * Copyright (c) 2024 Matt Borgerson
+ * Copyright (c) 2024-2025 Matt Borgerson
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
@@ -20,6 +20,7 @@
 #include "qemu/osdep.h"
 #include "qemu/fast-hash.h"
 #include "renderer.h"
+#include <math.h>
 
 void pgraph_vk_draw_begin(NV2AState *d)
 {
@@ -50,8 +51,8 @@ static VkPrimitiveTopology get_primitive_topology(PGRAPHState *pg)
 {
     PGRAPHVkState *r = pg->vk_renderer_state;
 
-    int polygon_mode = r->shader_binding->state.polygon_front_mode;
-    int primitive_mode = r->shader_binding->state.primitive_mode;
+    int polygon_mode = r->shader_binding->state.geom.polygon_front_mode;
+    int primitive_mode = r->shader_binding->state.geom.primitive_mode;
 
     if (polygon_mode == POLY_MODE_POINT) {
         return VK_PRIMITIVE_TOPOLOGY_POINT_LIST;
@@ -92,7 +93,8 @@ static VkPrimitiveTopology get_primitive_topology(PGRAPHState *pg)
     }
 }
 
-static void pipeline_cache_entry_init(Lru *lru, LruNode *node, void *state)
+static void pipeline_cache_entry_init(Lru *lru, LruNode *node,
+                                      const void *state)
 {
     PipelineBinding *snode = container_of(node, PipelineBinding, node);
     snode->layout = VK_NULL_HANDLE;
@@ -116,7 +118,8 @@ static void pipeline_cache_entry_post_evict(Lru *lru, LruNode *node)
     snode->layout = VK_NULL_HANDLE;
 }
 
-static bool pipeline_cache_entry_compare(Lru *lru, LruNode *node, void *key)
+static bool pipeline_cache_entry_compare(Lru *lru, LruNode *node,
+                                         const void *key)
 {
     PipelineBinding *snode = container_of(node, PipelineBinding, node);
     return memcmp(&snode->key, key, sizeof(PipelineKey));
@@ -745,15 +748,15 @@ static void create_pipeline(PGRAPHState *pg)
         (VkPipelineShaderStageCreateInfo){
             .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
             .stage = VK_SHADER_STAGE_VERTEX_BIT,
-            .module = r->shader_binding->vertex->module,
+            .module = r->shader_binding->vsh.module_info->module,
             .pName = "main",
         };
-    if (r->shader_binding->geometry) {
+    if (r->shader_binding->geom.module_info) {
         shader_stages[num_active_shader_stages++] =
             (VkPipelineShaderStageCreateInfo){
                 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
                 .stage = VK_SHADER_STAGE_GEOMETRY_BIT,
-                .module = r->shader_binding->geometry->module,
+                .module = r->shader_binding->geom.module_info->module,
                 .pName = "main",
             };
     }
@@ -761,7 +764,7 @@ static void create_pipeline(PGRAPHState *pg)
         (VkPipelineShaderStageCreateInfo){
             .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
             .stage = VK_SHADER_STAGE_FRAGMENT_BIT,
-            .module = r->shader_binding->fragment->module,
+            .module = r->shader_binding->psh.module_info->module,
             .pName = "main",
         };
 
@@ -815,7 +818,7 @@ static void create_pipeline(PGRAPHState *pg)
         .depthClampEnable = VK_TRUE,
         .rasterizerDiscardEnable = VK_FALSE,
         .polygonMode = pgraph_polygon_mode_vk_map[r->shader_binding->state
-                                                      .polygon_front_mode],
+                                                      .geom.polygon_front_mode],
         .lineWidth = 1.0f,
         .frontFace = (pgraph_reg_r(pg, NV_PGRAPH_SETUPRASTER) &
                       NV_PGRAPH_SETUPRASTER_FRONTFACE) ?
@@ -945,12 +948,23 @@ static void create_pipeline(PGRAPHState *pg)
         .blendConstants[3] = blend_constant[3],
     };
 
-    VkDynamicState dynamic_states[2] = { VK_DYNAMIC_STATE_VIEWPORT,
+    VkDynamicState dynamic_states[3] = { VK_DYNAMIC_STATE_VIEWPORT,
                                          VK_DYNAMIC_STATE_SCISSOR };
+    int num_dynamic_states = 2;
+
+    snode->has_dynamic_line_width =
+        (r->enabled_physical_device_features.wideLines == VK_TRUE) &&
+        (r->shader_binding->state.geom.polygon_front_mode == POLY_MODE_LINE ||
+         r->shader_binding->state.geom.primitive_mode == PRIM_TYPE_LINES ||
+         r->shader_binding->state.geom.primitive_mode == PRIM_TYPE_LINE_LOOP ||
+         r->shader_binding->state.geom.primitive_mode == PRIM_TYPE_LINE_STRIP);
+    if (snode->has_dynamic_line_width) {
+        dynamic_states[num_dynamic_states++] = VK_DYNAMIC_STATE_LINE_WIDTH;
+    }
 
     VkPipelineDynamicStateCreateInfo dynamic_state = {
         .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
-        .dynamicStateCount = ARRAY_SIZE(dynamic_states),
+        .dynamicStateCount = num_dynamic_states,
         .pDynamicStates = dynamic_states,
     };
 
@@ -1000,9 +1014,9 @@ static void create_pipeline(PGRAPHState *pg)
     };
 
     VkPushConstantRange push_constant_range;
-    if (r->shader_binding->state.use_push_constants_for_uniform_attrs) {
+    if (r->use_push_constants_for_uniform_attrs) {
         int num_uniform_attributes =
-            __builtin_popcount(r->shader_binding->state.uniform_attrs);
+            __builtin_popcount(r->shader_binding->state.vsh.uniform_attrs);
         if (num_uniform_attributes) {
             push_constant_range = (VkPushConstantRange){
                 .stageFlags = VK_SHADER_STAGE_VERTEX_BIT,
@@ -1055,7 +1069,7 @@ static void push_vertex_attr_values(PGRAPHState *pg)
 {
     PGRAPHVkState *r = pg->vk_renderer_state;
 
-    if (!r->shader_binding->state.use_push_constants_for_uniform_attrs) {
+    if (!r->use_push_constants_for_uniform_attrs) {
         return;
     }
 
@@ -1064,8 +1078,8 @@ static void push_vertex_attr_values(PGRAPHState *pg)
     float values[NV2A_VERTEXSHADER_ATTRIBUTES][4];
     int num_uniform_attrs = 0;
 
-    pgraph_get_inline_values(pg, r->shader_binding->state.uniform_attrs, values,
-                             &num_uniform_attrs);
+    pgraph_get_inline_values(pg, r->shader_binding->state.vsh.uniform_attrs,
+                             values, &num_uniform_attrs);
 
     if (num_uniform_attrs > 0) {
         vkCmdPushConstants(r->command_buffer, r->pipeline_binding->layout,
@@ -1420,6 +1434,21 @@ static void begin_pre_draw(PGRAPHState *pg)
     pgraph_vk_ensure_command_buffer(pg);
 }
 
+static float clamp_line_width_to_device_limits(PGRAPHState *pg, float width)
+{
+    PGRAPHVkState *r = pg->vk_renderer_state;
+
+    float min_width = r->device_props.limits.lineWidthRange[0];
+    float max_width = r->device_props.limits.lineWidthRange[1];
+    float granularity = r->device_props.limits.lineWidthGranularity;
+
+    if (granularity != 0.0f) {
+        float steps = roundf((width - min_width) / granularity);
+        width = min_width + steps * granularity;
+    }
+    return fminf(fmaxf(min_width, width), max_width);
+}
+
 static void begin_draw(PGRAPHState *pg)
 {
     PGRAPHVkState *r = pg->vk_renderer_state;
@@ -1491,6 +1520,12 @@ static void begin_draw(PGRAPHState *pg)
             .extent.height = scissor_height,
         };
         vkCmdSetScissor(r->command_buffer, 0, 1, &scissor);
+
+        if (r->pipeline_binding->has_dynamic_line_width) {
+            float line_width =
+                clamp_line_width_to_device_limits(pg, pg->surface_scale_factor);
+            vkCmdSetLineWidth(r->command_buffer, line_width);
+        }
     }
 
     if (!pg->clearing) {

hw/xbox/nv2a/pgraph/vk/glsl.c

@@ -1,7 +1,7 @@
 /*
  * Geforce NV2A PGRAPH Vulkan Renderer
  *
- * Copyright (c) 2024 Matt Borgerson
+ * Copyright (c) 2024-2025 Matt Borgerson
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
@@ -269,12 +269,24 @@ static void block_to_uniforms(const SpvReflectBlockVariable *block, ShaderUnifor
 
         assert(member->array.dims_count < 2);
 
+        int dim = 1;
+        for (int i = 0; i < member->array.dims_count; i++) {
+            dim *= member->array.dims[i];
+        }
+        int stride = MAX(member->array.stride, member->numeric.matrix.stride);
+        if (member->numeric.matrix.column_count) {
+            dim *= member->numeric.matrix.column_count;
+            if (member->array.stride) {
+                stride =
+                    member->array.stride / member->numeric.matrix.column_count;
+            }
+        }
         layout->uniforms[k] = (ShaderUniform){
             .name = strdup(member->name),
             .offset = member->offset,
             .dim_v = MAX(1, member->numeric.vector.component_count),
-            .dim_a = MAX(member->array.dims_count ? member->array.dims[0] : 1, member->numeric.matrix.column_count),
-            .stride = MAX(member->array.stride, member->numeric.matrix.stride),
+            .dim_a = dim,
+            .stride = stride,
         };
 
         // fprintf(stderr, "<%s offset=%zd dim_v=%zd dim_a=%zd stride=%zd>\n",
@@ -356,6 +368,7 @@ ShaderModuleInfo *pgraph_vk_create_shader_module_from_glsl(
     PGRAPHVkState *r, VkShaderStageFlagBits stage, const char *glsl)
 {
     ShaderModuleInfo *info = g_malloc0(sizeof(*info));
+    info->refcnt = 0;
     info->glsl = strdup(glsl);
     info->spirv = pgraph_vk_compile_glsl_to_spv(
         vk_shader_stage_to_glslang_stage(stage), glsl);
@@ -374,8 +387,24 @@ static void finalize_uniform_layout(ShaderUniformLayout *layout)
     }
 }
 
+void pgraph_vk_ref_shader_module(ShaderModuleInfo *info)
+{
+    info->refcnt++;
+}
+
+void pgraph_vk_unref_shader_module(PGRAPHVkState *r, ShaderModuleInfo *info)
+{
+    assert(info->refcnt >= 1);
+
+    info->refcnt--;
+    if (info->refcnt == 0) {
+        pgraph_vk_destroy_shader_module(r, info);
+    }
+}
+
 void pgraph_vk_destroy_shader_module(PGRAPHVkState *r, ShaderModuleInfo *info)
 {
+    assert(info->refcnt == 0);
     if (info->glsl) {
         free(info->glsl);
     }

hw/xbox/nv2a/pgraph/vk/glsl.h

@@ -1,7 +1,7 @@
 /*
  * Geforce NV2A PGRAPH Vulkan Renderer
  *
- * Copyright (c) 2024 Matt Borgerson
+ * Copyright (c) 2024-2025 Matt Borgerson
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
@@ -63,7 +63,7 @@ static inline void uniform_std140(ShaderUniformLayout *layout)
 			align = size;
 			stride = 0;
 		}
-		
+
 		offset = ROUND_UP(offset, align);
 
 		u->align = align;
@@ -87,7 +87,7 @@ static inline void uniform_std430(ShaderUniformLayout *layout)
 		size *= u->dim_v;
 		size_t align = size;
 		size *= u->dim_a;
-		
+
 		offset = ROUND_UP(offset, align);
 
 		u->align = align;
@@ -120,10 +120,10 @@ void *uniform_ptr(ShaderUniformLayout *layout, int idx)
     return (char *)layout->allocation + layout->uniforms[idx - 1].offset;
 }
 
-static inline
-void uniform_copy(ShaderUniformLayout *layout, int idx, void *values, size_t value_size, size_t count)
+static inline void uniform_copy(ShaderUniformLayout *layout, int idx,
+                                void *values, size_t value_size, size_t count)
 {
-	assert(idx > 0 && "invalid uniform index");
+    assert(idx > 0 && "invalid uniform index");
 
     ShaderUniform *u = &layout->uniforms[idx - 1];
     const size_t element_size = value_size * u->dim_v;
@@ -135,14 +135,14 @@ void uniform_copy(ShaderUniformLayout *layout, int idx, void *values, size_t val
 
     int index = 0;
     while (bytes_remaining) {
-    	assert(p_out < p_max);
-    	assert(index < u->dim_a);
+        assert((p_out + element_size) <= p_max);
+        assert(index < u->dim_a);
         memcpy(p_out, p_in, element_size);
         bytes_remaining -= element_size;
         p_out += u->stride;
         p_in += element_size;
         index += 1;
-	}
+    }
 }
 
 static inline
@@ -202,4 +202,10 @@ void uniform4i(ShaderUniformLayout *layout, int idx, int v0, int v1, int v2, int
 	uniform1iv(layout, idx, 4, values);
 }
 
+static inline void uniform1uiv(ShaderUniformLayout *layout, int idx,
+                               size_t count, uint32_t *values)
+{
+    uniform_copy(layout, idx, values, sizeof(uint32_t), count);
+}
+
 #endif

hw/xbox/nv2a/pgraph/vk/instance.c

@@ -1,7 +1,7 @@
 /*
  * Geforce NV2A PGRAPH Vulkan Renderer
  *
- * Copyright (c) 2024 Matt Borgerson
+ * Copyright (c) 2024-2025 Matt Borgerson
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
@@ -522,36 +522,48 @@ static bool create_logical_device(PGRAPHState *pg, Error **errp)
         .pQueuePriorities = &queuePriority,
     };
 
-    // Ensure device supports required features
-    VkPhysicalDeviceFeatures available_features, enabled_features;
-    vkGetPhysicalDeviceFeatures(r->physical_device, &available_features);
-    memset(&enabled_features, 0, sizeof(enabled_features));
+    // Check device features
+    VkPhysicalDeviceFeatures physical_device_features;
+    vkGetPhysicalDeviceFeatures(r->physical_device, &physical_device_features);
+    memset(&r->enabled_physical_device_features, 0,
+           sizeof(r->enabled_physical_device_features));
 
     struct {
         const char *name;
         VkBool32 available, *enabled;
-    } required_features[] = {
-        #define F(n) { #n, available_features.n, &enabled_features.n }
-        F(shaderClipDistance),
-        F(geometryShader),
-        F(shaderTessellationAndGeometryPointSize),
-        F(depthClamp),
-        F(occlusionQueryPrecise),
+        bool required;
+    } desired_features[] = {
+        // clang-format off
+        #define F(n, req) { \
+            .name = #n, \
+            .available = physical_device_features.n, \
+            .enabled = &r->enabled_physical_device_features.n, \
+            .required = req, \
+        }
+        F(shaderClipDistance, true),
+        F(geometryShader, true),
+        F(shaderTessellationAndGeometryPointSize, true),
+        F(depthClamp, true),
+        F(occlusionQueryPrecise, true),
+        F(fillModeNonSolid, true),
+        F(wideLines, false),
         #undef F
+        // clang-format on
     };
 
-    bool all_features_available = true;
-    for (int i = 0; i < ARRAY_SIZE(required_features); i++) {
-        if (required_features[i].available != VK_TRUE) {
+    bool all_required_features_available = true;
+    for (int i = 0; i < ARRAY_SIZE(desired_features); i++) {
+        if (desired_features[i].required &&
+            desired_features[i].available != VK_TRUE) {
             fprintf(stderr,
                     "Error: Device does not support required feature %s\n",
-                    required_features[i].name);
-            all_features_available = false;
+                    desired_features[i].name);
+            all_required_features_available = false;
         }
-        *required_features[i].enabled = VK_TRUE;
+        *desired_features[i].enabled = desired_features[i].available;
     }
 
-    if (!all_features_available) {
+    if (!all_required_features_available) {
         error_setg(errp, "Device does not support required features");
         return false;
     }
@@ -584,7 +596,7 @@ static bool create_logical_device(PGRAPHState *pg, Error **errp)
         .sType = VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO,
         .queueCreateInfoCount = 1,
         .pQueueCreateInfos = &queue_create_info,
-        .pEnabledFeatures = &enabled_features,
+        .pEnabledFeatures = &r->enabled_physical_device_features,
         .enabledExtensionCount = enabled_extension_names->len,
         .ppEnabledExtensionNames =
             &g_array_index(enabled_extension_names, const char *, 0),

hw/xbox/nv2a/pgraph/vk/renderer.h

@@ -1,7 +1,7 @@
 /*
  * Geforce NV2A PGRAPH Vulkan Renderer
  *
- * Copyright (c) 2024 Matt Borgerson
+ * Copyright (c) 2024-2025 Matt Borgerson
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
@@ -29,7 +29,7 @@
 #include "hw/xbox/nv2a/nv2a_regs.h"
 #include "hw/xbox/nv2a/pgraph/surface.h"
 #include "hw/xbox/nv2a/pgraph/texture.h"
-#include "hw/xbox/nv2a/pgraph/shaders.h"
+#include "hw/xbox/nv2a/pgraph/glsl/shaders.h"
 
 #include <vulkan/vulkan.h>
 #include <glslang/Include/glslang_c_interface.h>
@@ -77,6 +77,7 @@ typedef struct PipelineBinding {
     VkPipeline pipeline;
     VkRenderPass render_pass;
     unsigned int draw_time;
+    bool has_dynamic_line_width;
 } PipelineBinding;
 
 enum Buffer {
@@ -145,6 +146,7 @@ typedef struct SurfaceBinding {
 } SurfaceBinding;
 
 typedef struct ShaderModuleInfo {
+    int refcnt;
     char *glsl;
     GByteArray *spirv;
     VkShaderModule module;
@@ -154,48 +156,44 @@ typedef struct ShaderModuleInfo {
     ShaderUniformLayout push_constants;
 } ShaderModuleInfo;
 
+typedef struct ShaderModuleCacheKey {
+    VkShaderStageFlagBits kind;
+    union {
+        struct {
+            VshState state;
+            GenVshGlslOptions glsl_opts;
+        } vsh;
+        struct {
+            GeomState state;
+            GenGeomGlslOptions glsl_opts;
+        } geom;
+        struct {
+            PshState state;
+            GenPshGlslOptions glsl_opts;
+        } psh;
+    };
+} ShaderModuleCacheKey;
+
+typedef struct ShaderModuleCacheEntry {
+    LruNode node;
+    ShaderModuleCacheKey key;
+    ShaderModuleInfo *module_info;
+} ShaderModuleCacheEntry;
+
 typedef struct ShaderBinding {
     LruNode node;
-    bool initialized;
-
     ShaderState state;
-    ShaderModuleInfo *geometry;
-    ShaderModuleInfo *vertex;
-    ShaderModuleInfo *fragment;
-
-    int psh_constant_loc[9][2];
-    int alpha_ref_loc;
-
-    int bump_mat_loc[NV2A_MAX_TEXTURES];
-    int bump_scale_loc[NV2A_MAX_TEXTURES];
-    int bump_offset_loc[NV2A_MAX_TEXTURES];
-    int tex_scale_loc[NV2A_MAX_TEXTURES];
-
-    int surface_size_loc;
-    int clip_range_loc;
-    int clip_range_floc;
-    int depth_offset_loc;
-
-    int vsh_constant_loc;
-    uint32_t vsh_constants[NV2A_VERTEXSHADER_CONSTANTS][4];
-
-    int inv_viewport_loc;
-    int ltctxa_loc;
-    int ltctxb_loc;
-    int ltc1_loc;
-
-    int fog_color_loc;
-    int fog_param_loc;
-    int light_infinite_half_vector_loc[NV2A_MAX_LIGHTS];
-    int light_infinite_direction_loc[NV2A_MAX_LIGHTS];
-    int light_local_position_loc[NV2A_MAX_LIGHTS];
-    int light_local_attenuation_loc[NV2A_MAX_LIGHTS];
-
-    int clip_region_loc;
-
-    int material_alpha_loc;
-
-    int uniform_attrs_loc;
+    struct {
+        ShaderModuleInfo *module_info;
+        VshUniformLocs uniform_locs;
+    } vsh;
+    struct {
+        ShaderModuleInfo *module_info;
+    } geom;
+    struct {
+        ShaderModuleInfo *module_info;
+        PshUniformLocs uniform_locs;
+    } psh;
 } ShaderBinding;
 
 typedef struct TextureKey {
@@ -331,6 +329,7 @@ typedef struct PGRAPHVkState {
     bool memory_budget_extension_enabled;
 
     VkPhysicalDevice physical_device;
+    VkPhysicalDeviceFeatures enabled_physical_device_features;
     VkPhysicalDeviceProperties device_props;
     VkDevice device;
     VmaAllocator allocator;
@@ -405,6 +404,10 @@ typedef struct PGRAPHVkState {
     ShaderBinding *shader_binding;
     ShaderModuleInfo *quad_vert_module, *solid_frag_module;
     bool shader_bindings_changed;
+    bool use_push_constants_for_uniform_attrs;
+
+    Lru shader_module_cache;
+    ShaderModuleCacheEntry *shader_module_cache_entries;
 
     // FIXME: Merge these into a structure
     uint64_t uniform_buffer_hashes[2];
@@ -461,6 +464,8 @@ VkShaderModule pgraph_vk_create_shader_module_from_spv(PGRAPHVkState *r,
                                                        GByteArray *spv);
 ShaderModuleInfo *pgraph_vk_create_shader_module_from_glsl(
     PGRAPHVkState *r, VkShaderStageFlagBits stage, const char *glsl);
+void pgraph_vk_ref_shader_module(ShaderModuleInfo *info);
+void pgraph_vk_unref_shader_module(PGRAPHVkState *r, ShaderModuleInfo *info);
 void pgraph_vk_destroy_shader_module(PGRAPHVkState *r, ShaderModuleInfo *info);
 
 // buffer.c
@@ -550,7 +555,6 @@ void pgraph_vk_init_shaders(PGRAPHState *pg);
 void pgraph_vk_finalize_shaders(PGRAPHState *pg);
 void pgraph_vk_update_descriptor_sets(PGRAPHState *pg);
 void pgraph_vk_bind_shaders(PGRAPHState *pg);
-void pgraph_vk_update_shader_uniforms(PGRAPHState *pg);
 
 // reports.c
 void pgraph_vk_init_reports(PGRAPHState *pg);

hw/xbox/nv2a/pgraph/vk/shaders.c

@@ -1,13 +1,7 @@
 /*
  * Geforce NV2A PGRAPH Vulkan Renderer
  *
- * Copyright (c) 2024 Matt Borgerson
- *
- * Based on GL implementation:
- *
- * Copyright (c) 2015 espes
- * Copyright (c) 2015 Jannik Vogel
- * Copyright (c) 2018-2024 Matt Borgerson
+ * Copyright (c) 2024-2025 Matt Borgerson
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
@@ -24,15 +18,13 @@
  */
 
 #include "qemu/osdep.h"
-#include "hw/xbox/nv2a/pgraph/shaders.h"
-#include "hw/xbox/nv2a/pgraph/util.h"
-#include "hw/xbox/nv2a/pgraph/glsl/geom.h"
-#include "hw/xbox/nv2a/pgraph/glsl/vsh.h"
-#include "hw/xbox/nv2a/pgraph/glsl/psh.h"
 #include "qemu/fast-hash.h"
 #include "qemu/mstring.h"
 #include "renderer.h"
-#include <locale.h>
+
+#define VSH_UBO_BINDING 0
+#define PSH_UBO_BINDING 1
+#define PSH_TEX_BINDING 2
 
 const size_t MAX_UNIFORM_ATTR_VALUES_SIZE = NV2A_VERTEXSHADER_ATTRIBUTES * 4 * sizeof(float);
 
@@ -159,8 +151,8 @@ void pgraph_vk_update_descriptor_sets(PGRAPHState *pg)
     }
 
     ShaderBinding *binding = r->shader_binding;
-    ShaderUniformLayout *layouts[] = { &binding->vertex->uniforms,
-                                       &binding->fragment->uniforms };
+    ShaderUniformLayout *layouts[] = { &binding->vsh.module_info->uniforms,
+                                       &binding->psh.module_info->uniforms };
     VkDeviceSize ubo_buffer_total_size = 0;
     for (int i = 0; i < ARRAY_SIZE(layouts); i++) {
         ubo_buffer_total_size += layouts[i]->total_size;
@@ -236,90 +228,72 @@ void pgraph_vk_update_descriptor_sets(PGRAPHState *pg)
     r->descriptor_set_index++;
 }
 
-static void update_shader_constant_locations(ShaderBinding *binding)
+static void update_shader_uniform_locs(ShaderBinding *binding)
 {
-    char tmp[64];
-
-    /* lookup fragment shader uniforms */
-    for (int i = 0; i < 9; i++) {
-        for (int j = 0; j < 2; j++) {
-            snprintf(tmp, sizeof(tmp), "c%d_%d", j, i);
-            binding->psh_constant_loc[i][j] =
-                uniform_index(&binding->fragment->uniforms, tmp);
-        }
-    }
-    binding->alpha_ref_loc =
-        uniform_index(&binding->fragment->uniforms, "alphaRef");
-    binding->fog_color_loc =
-        uniform_index(&binding->fragment->uniforms, "fogColor");
-    for (int i = 1; i < NV2A_MAX_TEXTURES; i++) {
-        snprintf(tmp, sizeof(tmp), "bumpMat%d", i);
-        binding->bump_mat_loc[i] =
-            uniform_index(&binding->fragment->uniforms, tmp);
-        snprintf(tmp, sizeof(tmp), "bumpScale%d", i);
-        binding->bump_scale_loc[i] =
-            uniform_index(&binding->fragment->uniforms, tmp);
-        snprintf(tmp, sizeof(tmp), "bumpOffset%d", i);
-        binding->bump_offset_loc[i] =
-            uniform_index(&binding->fragment->uniforms, tmp);
+    for (int i = 0; i < ARRAY_SIZE(binding->vsh.uniform_locs); i++) {
+        binding->vsh.uniform_locs[i] = uniform_index(
+            &binding->vsh.module_info->uniforms, VshUniformInfo[i].name);
     }
 
-    for (int i = 0; i < NV2A_MAX_TEXTURES; i++) {
-        snprintf(tmp, sizeof(tmp), "texScale%d", i);
-        binding->tex_scale_loc[i] =
-            uniform_index(&binding->fragment->uniforms, tmp);
+    for (int i = 0; i < ARRAY_SIZE(binding->psh.uniform_locs); i++) {
+        binding->psh.uniform_locs[i] = uniform_index(
+            &binding->psh.module_info->uniforms, PshUniformInfo[i].name);
     }
-
-    /* lookup vertex shader uniforms */
-    binding->vsh_constant_loc = uniform_index(&binding->vertex->uniforms, "c");
-    binding->surface_size_loc =
-        uniform_index(&binding->vertex->uniforms, "surfaceSize");
-    binding->clip_range_loc =
-        uniform_index(&binding->vertex->uniforms, "clipRange");
-    binding->clip_range_floc =
-        uniform_index(&binding->fragment->uniforms, "clipRange");
-    binding->depth_offset_loc =
-        uniform_index(&binding->fragment->uniforms, "depthOffset");
-    binding->fog_param_loc =
-        uniform_index(&binding->vertex->uniforms, "fogParam");
-
-    binding->inv_viewport_loc =
-        uniform_index(&binding->vertex->uniforms, "invViewport");
-    binding->ltctxa_loc = uniform_index(&binding->vertex->uniforms, "ltctxa");
-    binding->ltctxb_loc = uniform_index(&binding->vertex->uniforms, "ltctxb");
-    binding->ltc1_loc = uniform_index(&binding->vertex->uniforms, "ltc1");
-
-    for (int i = 0; i < NV2A_MAX_LIGHTS; i++) {
-        snprintf(tmp, sizeof(tmp), "lightInfiniteHalfVector%d", i);
-        binding->light_infinite_half_vector_loc[i] =
-            uniform_index(&binding->vertex->uniforms, tmp);
-        snprintf(tmp, sizeof(tmp), "lightInfiniteDirection%d", i);
-        binding->light_infinite_direction_loc[i] =
-            uniform_index(&binding->vertex->uniforms, tmp);
-
-        snprintf(tmp, sizeof(tmp), "lightLocalPosition%d", i);
-        binding->light_local_position_loc[i] =
-            uniform_index(&binding->vertex->uniforms, tmp);
-        snprintf(tmp, sizeof(tmp), "lightLocalAttenuation%d", i);
-        binding->light_local_attenuation_loc[i] =
-            uniform_index(&binding->vertex->uniforms, tmp);
-    }
-
-    binding->clip_region_loc =
-        uniform_index(&binding->fragment->uniforms, "clipRegion");
-
-    binding->material_alpha_loc =
-        uniform_index(&binding->vertex->uniforms, "material_alpha");
-
-    binding->uniform_attrs_loc =
-        uniform_index(&binding->vertex->uniforms, "inlineValue");
 }
 
-static void shader_cache_entry_init(Lru *lru, LruNode *node, void *state)
+static ShaderModuleInfo *
+get_and_ref_shader_module_for_key(PGRAPHVkState *r,
+                                  const ShaderModuleCacheKey *key)
 {
-    ShaderBinding *snode = container_of(node, ShaderBinding, node);
-    memcpy(&snode->state, state, sizeof(ShaderState));
-    snode->initialized = false;
+    uint64_t hash = fast_hash((void *)key, sizeof(ShaderModuleCacheKey));
+    LruNode *node = lru_lookup(&r->shader_module_cache, hash, key);
+    ShaderModuleCacheEntry *module =
+        container_of(node, ShaderModuleCacheEntry, node);
+    pgraph_vk_ref_shader_module(module->module_info);
+    return module->module_info;
+}
+
+static void shader_cache_entry_init(Lru *lru, LruNode *node, const void *state)
+{
+    PGRAPHVkState *r = container_of(lru, PGRAPHVkState, shader_cache);
+    ShaderBinding *binding = container_of(node, ShaderBinding, node);
+    memcpy(&binding->state, state, sizeof(ShaderState));
+
+    NV2A_VK_DPRINTF("cache miss");
+    nv2a_profile_inc_counter(NV2A_PROF_SHADER_GEN);
+
+    ShaderModuleCacheKey key;
+
+    bool need_geometry_shader = pgraph_glsl_need_geom(&binding->state.geom);
+    if (need_geometry_shader) {
+        memset(&key, 0, sizeof(key));
+        key.kind = VK_SHADER_STAGE_GEOMETRY_BIT;
+        key.geom.state = binding->state.geom;
+        key.geom.glsl_opts.vulkan = true;
+        binding->geom.module_info = get_and_ref_shader_module_for_key(r, &key);
+    } else {
+        binding->geom.module_info = NULL;
+    }
+
+    memset(&key, 0, sizeof(key));
+    key.kind = VK_SHADER_STAGE_VERTEX_BIT;
+    key.vsh.state = binding->state.vsh;
+    key.vsh.glsl_opts.vulkan = true;
+    key.vsh.glsl_opts.prefix_outputs = need_geometry_shader;
+    key.vsh.glsl_opts.use_push_constants_for_uniform_attrs =
+        r->use_push_constants_for_uniform_attrs;
+    key.vsh.glsl_opts.ubo_binding = VSH_UBO_BINDING;
+    binding->vsh.module_info = get_and_ref_shader_module_for_key(r, &key);
+
+    memset(&key, 0, sizeof(key));
+    key.kind = VK_SHADER_STAGE_FRAGMENT_BIT;
+    key.psh.state = binding->state.psh;
+    key.psh.glsl_opts.vulkan = true;
+    key.psh.glsl_opts.ubo_binding = PSH_UBO_BINDING;
+    key.psh.glsl_opts.tex_binding = PSH_TEX_BINDING;
+    binding->psh.module_info = get_and_ref_shader_module_for_key(r, &key);
+
+    update_shader_uniform_locs(binding);
 }
 
 static void shader_cache_entry_post_evict(Lru *lru, LruNode *node)
@@ -328,25 +302,74 @@ static void shader_cache_entry_post_evict(Lru *lru, LruNode *node)
     ShaderBinding *snode = container_of(node, ShaderBinding, node);
 
     ShaderModuleInfo *modules[] = {
-        snode->geometry,
-        snode->vertex,
-        snode->fragment,
+        snode->vsh.module_info,
+        snode->geom.module_info,
+        snode->psh.module_info,
     };
     for (int i = 0; i < ARRAY_SIZE(modules); i++) {
         if (modules[i]) {
-            pgraph_vk_destroy_shader_module(r, modules[i]);
+            pgraph_vk_unref_shader_module(r, modules[i]);
         }
     }
-
-    snode->initialized = false;
 }
 
-static bool shader_cache_entry_compare(Lru *lru, LruNode *node, void *key)
+static bool shader_cache_entry_compare(Lru *lru, LruNode *node, const void *key)
 {
     ShaderBinding *snode = container_of(node, ShaderBinding, node);
     return memcmp(&snode->state, key, sizeof(ShaderState));
 }
 
+static void shader_module_cache_entry_init(Lru *lru, LruNode *node,
+                                           const void *key)
+{
+    PGRAPHVkState *r = container_of(lru, PGRAPHVkState, shader_module_cache);
+    ShaderModuleCacheEntry *module =
+        container_of(node, ShaderModuleCacheEntry, node);
+    memcpy(&module->key, key, sizeof(ShaderModuleCacheKey));
+
+    MString *code;
+
+    switch (module->key.kind) {
+    case VK_SHADER_STAGE_VERTEX_BIT:
+        code = pgraph_glsl_gen_vsh(&module->key.vsh.state,
+                                   module->key.vsh.glsl_opts);
+        break;
+    case VK_SHADER_STAGE_GEOMETRY_BIT:
+        code = pgraph_glsl_gen_geom(&module->key.geom.state,
+                                    module->key.geom.glsl_opts);
+        break;
+    case VK_SHADER_STAGE_FRAGMENT_BIT:
+        code = pgraph_glsl_gen_psh(&module->key.psh.state,
+                                   module->key.psh.glsl_opts);
+        break;
+    default:
+        assert(!"Invalid shader module kind");
+        code = NULL;
+    }
+
+    module->module_info = pgraph_vk_create_shader_module_from_glsl(
+        r, module->key.kind, mstring_get_str(code));
+    pgraph_vk_ref_shader_module(module->module_info);
+    mstring_unref(code);
+}
+
+static void shader_module_cache_entry_post_evict(Lru *lru, LruNode *node)
+{
+    PGRAPHVkState *r = container_of(lru, PGRAPHVkState, shader_module_cache);
+    ShaderModuleCacheEntry *module =
+        container_of(node, ShaderModuleCacheEntry, node);
+    pgraph_vk_unref_shader_module(r, module->module_info);
+    module->module_info = NULL;
+}
+
+static bool shader_module_cache_entry_compare(Lru *lru, LruNode *node,
+                                              const void *key)
+{
+    ShaderModuleCacheEntry *module =
+        container_of(node, ShaderModuleCacheEntry, node);
+    return memcmp(&module->key, key, sizeof(ShaderModuleCacheKey));
+}
+
 static void shader_cache_init(PGRAPHState *pg)
 {
     PGRAPHVkState *r = pg->vk_renderer_state;
@@ -361,6 +384,22 @@ static void shader_cache_init(PGRAPHState *pg)
     r->shader_cache.init_node = shader_cache_entry_init;
     r->shader_cache.compare_nodes = shader_cache_entry_compare;
     r->shader_cache.post_node_evict = shader_cache_entry_post_evict;
+
+    /* FIXME: Make this configurable */
+    const size_t shader_module_cache_size = 50 * 1024;
+    lru_init(&r->shader_module_cache);
+    r->shader_module_cache_entries =
+        g_malloc_n(shader_module_cache_size, sizeof(ShaderModuleCacheEntry));
+    assert(r->shader_module_cache_entries != NULL);
+    for (int i = 0; i < shader_module_cache_size; i++) {
+        lru_add_free(&r->shader_module_cache,
+                     &r->shader_module_cache_entries[i].node);
+    }
+
+    r->shader_module_cache.init_node = shader_module_cache_entry_init;
+    r->shader_module_cache.compare_nodes = shader_module_cache_entry_compare;
+    r->shader_module_cache.post_node_evict =
+        shader_module_cache_entry_post_evict;
 }
 
 static void shader_cache_finalize(PGRAPHState *pg)
@@ -370,475 +409,78 @@ static void shader_cache_finalize(PGRAPHState *pg)
     lru_flush(&r->shader_cache);
     g_free(r->shader_cache_entries);
     r->shader_cache_entries = NULL;
+
+    lru_flush(&r->shader_module_cache);
+    g_free(r->shader_module_cache_entries);
+    r->shader_module_cache_entries = NULL;
 }
 
-static ShaderBinding *gen_shaders(PGRAPHState *pg, ShaderState *state)
+static ShaderBinding *get_shader_binding_for_state(PGRAPHVkState *r,
+                                                   const ShaderState *state)
 {
-    PGRAPHVkState *r = pg->vk_renderer_state;
-
     uint64_t hash = fast_hash((void *)state, sizeof(*state));
     LruNode *node = lru_lookup(&r->shader_cache, hash, state);
-    ShaderBinding *snode = container_of(node, ShaderBinding, node);
-
-    NV2A_VK_DPRINTF("shader state hash: %016" PRIx64 " %p", hash, snode);
-
-    if (!snode->initialized) {
-        NV2A_VK_DPRINTF("cache miss");
-        nv2a_profile_inc_counter(NV2A_PROF_SHADER_GEN);
-
-        char *previous_numeric_locale = setlocale(LC_NUMERIC, NULL);
-        if (previous_numeric_locale) {
-            previous_numeric_locale = g_strdup(previous_numeric_locale);
-        }
-
-        /* Ensure numeric values are printed with '.' radix, no grouping */
-        setlocale(LC_NUMERIC, "C");
-
-        MString *geometry_shader_code = pgraph_gen_geom_glsl(
-            state->polygon_front_mode, state->polygon_back_mode,
-            state->primitive_mode, state->smooth_shading, true);
-        if (geometry_shader_code) {
-            NV2A_VK_DPRINTF("geometry shader: \n%s",
-                            mstring_get_str(geometry_shader_code));
-            snode->geometry = pgraph_vk_create_shader_module_from_glsl(
-                r, VK_SHADER_STAGE_GEOMETRY_BIT,
-                mstring_get_str(geometry_shader_code));
-            mstring_unref(geometry_shader_code);
-        } else {
-            snode->geometry = NULL;
-        }
-
-        MString *vertex_shader_code =
-            pgraph_gen_vsh_glsl(state, geometry_shader_code != NULL);
-        NV2A_VK_DPRINTF("vertex shader: \n%s",
-                        mstring_get_str(vertex_shader_code));
-        snode->vertex = pgraph_vk_create_shader_module_from_glsl(
-            r, VK_SHADER_STAGE_VERTEX_BIT,
-            mstring_get_str(vertex_shader_code));
-        mstring_unref(vertex_shader_code);
-
-        MString *fragment_shader_code = pgraph_gen_psh_glsl(state->psh);
-        NV2A_VK_DPRINTF("fragment shader: \n%s",
-                        mstring_get_str(fragment_shader_code));
-        snode->fragment = pgraph_vk_create_shader_module_from_glsl(
-            r, VK_SHADER_STAGE_FRAGMENT_BIT,
-            mstring_get_str(fragment_shader_code));
-        mstring_unref(fragment_shader_code);
-
-        if (previous_numeric_locale) {
-            setlocale(LC_NUMERIC, previous_numeric_locale);
-            g_free(previous_numeric_locale);
-        }
-
-        update_shader_constant_locations(snode);
-
-        snode->initialized = true;
-    }
-
-    return snode;
+    ShaderBinding *binding = container_of(node, ShaderBinding, node);
+    NV2A_VK_DPRINTF("shader state hash: %016" PRIx64 " %p", hash, binding);
+    return binding;
 }
 
-static void update_uniform_attr_values(PGRAPHState *pg, ShaderBinding *binding)
+static void apply_uniform_updates(ShaderUniformLayout *layout,
+                                  const UniformInfo *info, int *locs,
+                                  void *values, size_t count)
 {
-    float values[NV2A_VERTEXSHADER_ATTRIBUTES][4];
-    int num_uniform_attrs = 0;
-
-    pgraph_get_inline_values(pg, binding->state.uniform_attrs, values,
-                             &num_uniform_attrs);
-
-    if (num_uniform_attrs > 0) {
-        uniform1fv(&binding->vertex->uniforms, binding->uniform_attrs_loc,
-                   num_uniform_attrs * 4, &values[0][0]);
+    for (int i = 0; i < count; i++) {
+        if (locs[i] != -1) {
+            uniform_copy(layout, locs[i], (char*)values + info[i].val_offs,
+                         4, (info[i].size * info[i].count) / 4);
+        }
     }
 }
 
-// FIXME: Move to common
-static void shader_update_constants(PGRAPHState *pg, ShaderBinding *binding,
-                                    bool binding_changed, bool vertex_program,
-                                    bool fixed_function)
-{
-    ShaderState *state = &binding->state;
-
-    /* update combiner constants */
-    for (int i = 0; i < 9; i++) {
-        uint32_t constant[2];
-        if (i == 8) {
-            /* final combiner */
-            constant[0] = pgraph_reg_r(pg, NV_PGRAPH_SPECFOGFACTOR0);
-            constant[1] = pgraph_reg_r(pg, NV_PGRAPH_SPECFOGFACTOR1);
-        } else {
-            constant[0] = pgraph_reg_r(pg, NV_PGRAPH_COMBINEFACTOR0 + i * 4);
-            constant[1] = pgraph_reg_r(pg, NV_PGRAPH_COMBINEFACTOR1 + i * 4);
-        }
-
-        for (int j = 0; j < 2; j++) {
-            GLint loc = binding->psh_constant_loc[i][j];
-            if (loc != -1) {
-                float value[4];
-                pgraph_argb_pack32_to_rgba_float(constant[j], value);
-                uniform1fv(&binding->fragment->uniforms, loc, 4, value);
-            }
-        }
-    }
-    if (binding->alpha_ref_loc != -1) {
-        int alpha_ref = GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_CONTROL_0),
-                                   NV_PGRAPH_CONTROL_0_ALPHAREF);
-        uniform1i(&binding->fragment->uniforms, binding->alpha_ref_loc,
-                         alpha_ref);
-    }
-
-
-    /* For each texture stage */
-    for (int i = 0; i < NV2A_MAX_TEXTURES; i++) {
-        int loc;
-
-        /* Bump luminance only during stages 1 - 3 */
-        if (i > 0) {
-            loc = binding->bump_mat_loc[i];
-            if (loc != -1) {
-                uint32_t m_u32[4];
-                m_u32[0] = pgraph_reg_r(pg, NV_PGRAPH_BUMPMAT00 + 4 * (i - 1));
-                m_u32[1] = pgraph_reg_r(pg, NV_PGRAPH_BUMPMAT01 + 4 * (i - 1));
-                m_u32[2] = pgraph_reg_r(pg, NV_PGRAPH_BUMPMAT10 + 4 * (i - 1));
-                m_u32[3] = pgraph_reg_r(pg, NV_PGRAPH_BUMPMAT11 + 4 * (i - 1));
-                float m[4];
-                m[0] = *(float*)&m_u32[0];
-                m[1] = *(float*)&m_u32[1];
-                m[2] = *(float*)&m_u32[2];
-                m[3] = *(float*)&m_u32[3];
-                uniformMatrix2fv(&binding->fragment->uniforms, loc, m);
-            }
-            loc = binding->bump_scale_loc[i];
-            if (loc != -1) {
-                uint32_t v =
-                    pgraph_reg_r(pg, NV_PGRAPH_BUMPSCALE1 + (i - 1) * 4);
-                uniform1f(&binding->fragment->uniforms, loc,
-                                 *(float *)&v);
-            }
-            loc = binding->bump_offset_loc[i];
-            if (loc != -1) {
-                uint32_t v =
-                    pgraph_reg_r(pg, NV_PGRAPH_BUMPOFFSET1 + (i - 1) * 4);
-                uniform1f(&binding->fragment->uniforms, loc,
-                                 *(float *)&v);
-            }
-        }
-
-        loc = binding->tex_scale_loc[i];
-        if (loc != -1) {
-            assert(pg->vk_renderer_state->texture_bindings[i] != NULL);
-            float scale = pg->vk_renderer_state->texture_bindings[i]->key.scale;
-            BasicColorFormatInfo f_basic = kelvin_color_format_info_map[pg->vk_renderer_state->texture_bindings[i]->key.state.color_format];
-            if (!f_basic.linear) {
-                scale = 1.0;
-            }
-            uniform1f(&binding->fragment->uniforms, loc, scale);
-        }
-    }
-
-    if (binding->fog_color_loc != -1) {
-        uint32_t fog_color = pgraph_reg_r(pg, NV_PGRAPH_FOGCOLOR);
-        uniform4f(&binding->fragment->uniforms, binding->fog_color_loc,
-                         GET_MASK(fog_color, NV_PGRAPH_FOGCOLOR_RED) / 255.0,
-                         GET_MASK(fog_color, NV_PGRAPH_FOGCOLOR_GREEN) / 255.0,
-                         GET_MASK(fog_color, NV_PGRAPH_FOGCOLOR_BLUE) / 255.0,
-                         GET_MASK(fog_color, NV_PGRAPH_FOGCOLOR_ALPHA) / 255.0);
-    }
-    if (binding->fog_param_loc != -1) {
-        uint32_t v[2];
-        v[0] = pgraph_reg_r(pg, NV_PGRAPH_FOGPARAM0);
-        v[1] = pgraph_reg_r(pg, NV_PGRAPH_FOGPARAM1);
-        uniform2f(&binding->vertex->uniforms,
-                         binding->fog_param_loc, *(float *)&v[0],
-                         *(float *)&v[1]);
-    }
-
-    float zmax;
-    switch (pg->surface_shape.zeta_format) {
-    case NV097_SET_SURFACE_FORMAT_ZETA_Z16:
-        zmax = pg->surface_shape.z_format ? f16_max : (float)0xFFFF;
-        break;
-    case NV097_SET_SURFACE_FORMAT_ZETA_Z24S8:
-        zmax = pg->surface_shape.z_format ? f24_max : (float)0xFFFFFF;
-        break;
-    default:
-        assert(0);
-    }
-
-    if (fixed_function) {
-        /* update lighting constants */
-        struct {
-            uint32_t *v;
-            int locs;
-            size_t len;
-        } lighting_arrays[] = {
-            { &pg->ltctxa[0][0], binding->ltctxa_loc, NV2A_LTCTXA_COUNT },
-            { &pg->ltctxb[0][0], binding->ltctxb_loc, NV2A_LTCTXB_COUNT },
-            { &pg->ltc1[0][0], binding->ltc1_loc, NV2A_LTC1_COUNT },
-        };
-
-        for (int i = 0; i < ARRAY_SIZE(lighting_arrays); i++) {
-            uniform1iv(
-                &binding->vertex->uniforms, lighting_arrays[i].locs,
-                lighting_arrays[i].len * 4, (void *)lighting_arrays[i].v);
-        }
-
-        for (int i = 0; i < NV2A_MAX_LIGHTS; i++) {
-            int loc = binding->light_infinite_half_vector_loc[i];
-            if (loc != -1) {
-                uniform1fv(&binding->vertex->uniforms, loc, 3,
-                                 pg->light_infinite_half_vector[i]);
-            }
-            loc = binding->light_infinite_direction_loc[i];
-            if (loc != -1) {
-                uniform1fv(&binding->vertex->uniforms, loc, 3,
-                                 pg->light_infinite_direction[i]);
-            }
-
-            loc = binding->light_local_position_loc[i];
-            if (loc != -1) {
-                uniform1fv(&binding->vertex->uniforms, loc, 3,
-                                 pg->light_local_position[i]);
-            }
-            loc = binding->light_local_attenuation_loc[i];
-            if (loc != -1) {
-                uniform1fv(&binding->vertex->uniforms, loc, 3,
-                                 pg->light_local_attenuation[i]);
-            }
-        }
-
-        /* estimate the viewport by assuming it matches the surface ... */
-        unsigned int aa_width = 1, aa_height = 1;
-        pgraph_apply_anti_aliasing_factor(pg, &aa_width, &aa_height);
-
-        float m11 = 0.5 * (pg->surface_binding_dim.width / aa_width);
-        float m22 = -0.5 * (pg->surface_binding_dim.height / aa_height);
-        float m33 = zmax;
-        float m41 = *(float *)&pg->vsh_constants[NV_IGRAPH_XF_XFCTX_VPOFF][0];
-        float m42 = *(float *)&pg->vsh_constants[NV_IGRAPH_XF_XFCTX_VPOFF][1];
-
-        float invViewport[16] = {
-            1.0 / m11, 0,  0, 0,         0, 1.0 / m22,        0,
-            0,         0,  0, 1.0 / m33, 0, -1.0 + m41 / m11, 1.0 + m42 / m22,
-            0,         1.0
-        };
-
-        if (binding->inv_viewport_loc != -1) {
-            uniformMatrix4fv(&binding->vertex->uniforms,
-                                    binding->inv_viewport_loc, &invViewport[0]);
-        }
-    }
-
-    /* update vertex program constants */
-    uniform1iv(&binding->vertex->uniforms, binding->vsh_constant_loc,
-               NV2A_VERTEXSHADER_CONSTANTS * 4, (void *)pg->vsh_constants);
-
-    if (binding->surface_size_loc != -1) {
-        unsigned int aa_width = 1, aa_height = 1;
-        pgraph_apply_anti_aliasing_factor(pg, &aa_width, &aa_height);
-        uniform2f(&binding->vertex->uniforms, binding->surface_size_loc,
-                         pg->surface_binding_dim.width / aa_width,
-                         pg->surface_binding_dim.height / aa_height);
-    }
-
-    if (binding->clip_range_loc != -1 || binding->clip_range_floc != -1) {
-        uint32_t v[2];
-        v[0] = pgraph_reg_r(pg, NV_PGRAPH_ZCLIPMIN);
-        v[1] = pgraph_reg_r(pg, NV_PGRAPH_ZCLIPMAX);
-        float zclip_min = *(float *)&v[0];
-        float zclip_max = *(float *)&v[1];
-
-        if (binding->clip_range_loc != -1) {
-            uniform4f(&binding->vertex->uniforms, binding->clip_range_loc, 0,
-                      zmax, zclip_min, zclip_max);
-        }
-        if (binding->clip_range_floc != -1) {
-            uniform4f(&binding->fragment->uniforms, binding->clip_range_floc, 0,
-                      zmax, zclip_min, zclip_max);
-        }
-    }
-
-    if (binding->depth_offset_loc != -1) {
-        float zbias = 0.0f;
-
-        if (pgraph_reg_r(pg, NV_PGRAPH_SETUPRASTER) &
-            (NV_PGRAPH_SETUPRASTER_POFFSETFILLENABLE |
-             NV_PGRAPH_SETUPRASTER_POFFSETLINEENABLE |
-             NV_PGRAPH_SETUPRASTER_POFFSETPOINTENABLE)) {
-            uint32_t zbias_u32 = pgraph_reg_r(pg, NV_PGRAPH_ZOFFSETBIAS);
-            zbias = *(float *)&zbias_u32;
-
-            if (pgraph_reg_r(pg, NV_PGRAPH_ZOFFSETFACTOR) != 0 &&
-                (pgraph_reg_r(pg, NV_PGRAPH_CONTROL_0) &
-                 NV_PGRAPH_CONTROL_0_Z_PERSPECTIVE_ENABLE)) {
-                /* TODO: emulate zfactor when z_perspective true, i.e.
-                 * w-buffering. Perhaps calculate an additional offset based on
-                 * triangle orientation in geometry shader and pass the result
-                 * to fragment shader and add it to gl_FragDepth as well.
-                 */
-                NV2A_UNIMPLEMENTED("NV_PGRAPH_ZOFFSETFACTOR for w-buffering");
-            }
-        }
-
-        uniform1f(&binding->fragment->uniforms, binding->depth_offset_loc,
-                  zbias);
-    }
-
-    /* Clipping regions */
-    unsigned int max_gl_width = pg->surface_binding_dim.width;
-    unsigned int max_gl_height = pg->surface_binding_dim.height;
-    pgraph_apply_scaling_factor(pg, &max_gl_width, &max_gl_height);
-
-    uint32_t clip_regions[8][4];
-
-    for (int i = 0; i < 8; i++) {
-        uint32_t x = pgraph_reg_r(pg, NV_PGRAPH_WINDOWCLIPX0 + i * 4);
-        unsigned int x_min = GET_MASK(x, NV_PGRAPH_WINDOWCLIPX0_XMIN);
-        unsigned int x_max = GET_MASK(x, NV_PGRAPH_WINDOWCLIPX0_XMAX) + 1;
-        uint32_t y = pgraph_reg_r(pg, NV_PGRAPH_WINDOWCLIPY0 + i * 4);
-        unsigned int y_min = GET_MASK(y, NV_PGRAPH_WINDOWCLIPY0_YMIN);
-        unsigned int y_max = GET_MASK(y, NV_PGRAPH_WINDOWCLIPY0_YMAX) + 1;
-        pgraph_apply_anti_aliasing_factor(pg, &x_min, &y_min);
-        pgraph_apply_anti_aliasing_factor(pg, &x_max, &y_max);
-
-        pgraph_apply_scaling_factor(pg, &x_min, &y_min);
-        pgraph_apply_scaling_factor(pg, &x_max, &y_max);
-
-        clip_regions[i][0] = x_min;
-        clip_regions[i][1] = y_min;
-        clip_regions[i][2] = x_max;
-        clip_regions[i][3] = y_max;
-    }
-    uniform1iv(&binding->fragment->uniforms, binding->clip_region_loc,
-                     8 * 4, (void *)clip_regions);
-
-    if (binding->material_alpha_loc != -1) {
-        uniform1f(&binding->vertex->uniforms, binding->material_alpha_loc,
-                         pg->material_alpha);
-    }
-
-    if (!state->use_push_constants_for_uniform_attrs && state->uniform_attrs) {
-        update_uniform_attr_values(pg, binding);
-    }
-}
-
-// Quickly check PGRAPH state to see if any registers have changed that
-// necessitate a full shader state inspection.
-static bool check_shaders_dirty(PGRAPHState *pg)
-{
-    PGRAPHVkState *r = pg->vk_renderer_state;
-
-    if (!r->shader_binding) {
-        return true;
-    }
-    if (pg->program_data_dirty) {
-        return true;
-    }
-
-    int num_stages = pgraph_reg_r(pg, NV_PGRAPH_COMBINECTL) & 0xFF;
-    for (int i = 0; i < num_stages; i++) {
-        if (pgraph_is_reg_dirty(pg, NV_PGRAPH_COMBINEALPHAI0 + i * 4) ||
-            pgraph_is_reg_dirty(pg, NV_PGRAPH_COMBINEALPHAO0 + i * 4) ||
-            pgraph_is_reg_dirty(pg, NV_PGRAPH_COMBINECOLORI0 + i * 4) ||
-            pgraph_is_reg_dirty(pg, NV_PGRAPH_COMBINECOLORO0 + i * 4)) {
-            return true;
-        }
-    }
-
-    unsigned int regs[] = {
-        NV_PGRAPH_COMBINECTL,
-        NV_PGRAPH_COMBINESPECFOG0,
-        NV_PGRAPH_COMBINESPECFOG1,
-        NV_PGRAPH_CONTROL_0,
-        NV_PGRAPH_CONTROL_3,
-        NV_PGRAPH_CSV0_C,
-        NV_PGRAPH_CSV0_D,
-        NV_PGRAPH_CSV1_A,
-        NV_PGRAPH_CSV1_B,
-        NV_PGRAPH_POINTSIZE,
-        NV_PGRAPH_SETUPRASTER,
-        NV_PGRAPH_SHADERCLIPMODE,
-        NV_PGRAPH_SHADERCTL,
-        NV_PGRAPH_SHADERPROG,
-        NV_PGRAPH_SHADOWCTL,
-        NV_PGRAPH_ZCOMPRESSOCCLUDE,
-    };
-    for (int i = 0; i < ARRAY_SIZE(regs); i++) {
-        if (pgraph_is_reg_dirty(pg, regs[i])) {
-            return true;
-        }
-    }
-
-    ShaderState *state = &r->shader_binding->state;
-    if (pg->uniform_attrs != state->uniform_attrs ||
-        pg->swizzle_attrs != state->swizzle_attrs ||
-        pg->compressed_attrs != state->compressed_attrs ||
-        pg->primitive_mode != state->primitive_mode ||
-        pg->surface_scale_factor != state->surface_scale_factor) {
-        return true;
-    }
-
-    // Textures
-    for (int i = 0; i < 4; i++) {
-        if (pg->texture_matrix_enable[i] != pg->vk_renderer_state->shader_binding->state.texture_matrix_enable[i] ||
-            pgraph_is_reg_dirty(pg, NV_PGRAPH_TEXCTL0_0 + i * 4) ||
-            pgraph_is_reg_dirty(pg, NV_PGRAPH_TEXFILTER0 + i * 4) ||
-            pgraph_is_reg_dirty(pg, NV_PGRAPH_TEXFMT0 + i * 4)) {
-            return true;
-        }
-    }
-
-    nv2a_profile_inc_counter(NV2A_PROF_SHADER_BIND_NOTDIRTY);
-
-    return false;
-}
-
-void pgraph_vk_bind_shaders(PGRAPHState *pg)
+// FIXME: Dirty tracking
+static void update_shader_uniforms(PGRAPHState *pg)
 {
     NV2A_VK_DGROUP_BEGIN("%s", __func__);
 
     PGRAPHVkState *r = pg->vk_renderer_state;
-
-    r->shader_bindings_changed = false;
-
-    if (check_shaders_dirty(pg)) {
-        ShaderState new_state;
-        memset(&new_state, 0, sizeof(ShaderState));
-        new_state = pgraph_get_shader_state(pg);
-        new_state.vulkan = true;
-        new_state.psh.vulkan = true;
-        new_state.use_push_constants_for_uniform_attrs =
-            (r->device_props.limits.maxPushConstantsSize >=
-             MAX_UNIFORM_ATTR_VALUES_SIZE);
-
-        if (!r->shader_binding || memcmp(&r->shader_binding->state, &new_state, sizeof(ShaderState))) {
-            r->shader_binding = gen_shaders(pg, &new_state);
-            r->shader_bindings_changed = true;
-        }
-    }
-
-    // FIXME: Use dirty bits
-    pgraph_vk_update_shader_uniforms(pg);
-
-    NV2A_VK_DGROUP_END();
-}
-
-void pgraph_vk_update_shader_uniforms(PGRAPHState *pg)
-{
-    PGRAPHVkState *r = pg->vk_renderer_state;
-    NV2A_VK_DGROUP_BEGIN("%s", __func__);
     nv2a_profile_inc_counter(NV2A_PROF_SHADER_BIND);
 
     assert(r->shader_binding);
     ShaderBinding *binding = r->shader_binding;
-    ShaderUniformLayout *layouts[] = { &binding->vertex->uniforms,
-                                        &binding->fragment->uniforms };
-    shader_update_constants(pg, r->shader_binding, true,
-                            r->shader_binding->state.vertex_program,
-                            r->shader_binding->state.fixed_function);
+    ShaderUniformLayout *layouts[] = { &binding->vsh.module_info->uniforms,
+                                       &binding->psh.module_info->uniforms };
+
+    VshUniformValues vsh_values;
+    pgraph_glsl_set_vsh_uniform_values(pg, &binding->state.vsh,
+                                  binding->vsh.uniform_locs, &vsh_values);
+    apply_uniform_updates(&binding->vsh.module_info->uniforms, VshUniformInfo,
+                          binding->vsh.uniform_locs, &vsh_values,
+                          VshUniform__COUNT);
+
+    PshUniformValues psh_values;
+    pgraph_glsl_set_psh_uniform_values(pg, binding->psh.uniform_locs,
+                                       &psh_values);
+    for (int i = 0; i < 4; i++) {
+        assert(r->texture_bindings[i] != NULL);
+        float scale = r->texture_bindings[i]->key.scale;
+
+        BasicColorFormatInfo f_basic =
+            kelvin_color_format_info_map[pg->vk_renderer_state
+                                             ->texture_bindings[i]
+                                             ->key.state.color_format];
+        if (!f_basic.linear) {
+            scale = 1.0;
+        }
+
+        psh_values.texScale[i] = scale;
+    }
+    apply_uniform_updates(&binding->psh.module_info->uniforms, PshUniformInfo,
+                          binding->psh.uniform_locs, &psh_values,
+                          PshUniform__COUNT);
 
     for (int i = 0; i < ARRAY_SIZE(layouts); i++) {
-        uint64_t hash = fast_hash(layouts[i]->allocation, layouts[i]->total_size);
+        uint64_t hash =
+            fast_hash(layouts[i]->allocation, layouts[i]->total_size);
         r->uniforms_changed |= (hash != r->uniform_buffer_hashes[i]);
         r->uniform_buffer_hashes[i] = hash;
     }
@@ -850,13 +492,44 @@ void pgraph_vk_update_shader_uniforms(PGRAPHState *pg)
     NV2A_VK_DGROUP_END();
 }
 
+void pgraph_vk_bind_shaders(PGRAPHState *pg)
+{
+    NV2A_VK_DGROUP_BEGIN("%s", __func__);
+
+    PGRAPHVkState *r = pg->vk_renderer_state;
+
+    r->shader_bindings_changed = false;
+
+    if (!r->shader_binding ||
+        pgraph_glsl_check_shader_state_dirty(pg, &r->shader_binding->state)) {
+        ShaderState new_state = pgraph_glsl_get_shader_state(pg);
+        if (!r->shader_binding || memcmp(&r->shader_binding->state, &new_state,
+                                         sizeof(ShaderState))) {
+            r->shader_binding = get_shader_binding_for_state(r, &new_state);
+            r->shader_bindings_changed = true;
+        }
+    } else {
+        nv2a_profile_inc_counter(NV2A_PROF_SHADER_BIND_NOTDIRTY);
+    }
+
+    update_shader_uniforms(pg);
+
+    NV2A_VK_DGROUP_END();
+}
+
 void pgraph_vk_init_shaders(PGRAPHState *pg)
 {
+    PGRAPHVkState *r = pg->vk_renderer_state;
+
     pgraph_vk_init_glsl_compiler();
     create_descriptor_pool(pg);
     create_descriptor_set_layout(pg);
     create_descriptor_sets(pg);
     shader_cache_init(pg);
+
+    r->use_push_constants_for_uniform_attrs =
+        (r->device_props.limits.maxPushConstantsSize >=
+         MAX_UNIFORM_ATTR_VALUES_SIZE);
 }
 
 void pgraph_vk_finalize_shaders(PGRAPHState *pg)

hw/xbox/nv2a/pgraph/vk/surface-compute.c

@@ -524,7 +524,8 @@ void pgraph_vk_unpack_depth_stencil(PGRAPHState *pg, SurfaceBinding *surface,
     pgraph_vk_end_debug_marker(r, cmd);
 }
 
-static void pipeline_cache_entry_init(Lru *lru, LruNode *node, void *state)
+static void pipeline_cache_entry_init(Lru *lru, LruNode *node,
+                                      const void *state)
 {
     PGRAPHVkState *r = container_of(lru, PGRAPHVkState, compute.pipeline_cache);
     ComputePipeline *snode = container_of(node, ComputePipeline, node);
@@ -556,7 +557,8 @@ static void pipeline_cache_entry_post_evict(Lru *lru, LruNode *node)
     pipeline_cache_release_node_resources(r, snode);
 }
 
-static bool pipeline_cache_entry_compare(Lru *lru, LruNode *node, void *key)
+static bool pipeline_cache_entry_compare(Lru *lru, LruNode *node,
+                                         const void *key)
 {
     ComputePipeline *snode = container_of(node, ComputePipeline, node);
     return memcmp(&snode->key, key, sizeof(ComputePipelineKey));

hw/xbox/nv2a/pgraph/vk/surface.c

@@ -122,17 +122,22 @@ static void memcpy_image(void *dst, void const *src, int dst_stride,
     }
 }
 
-void pgraph_vk_download_surfaces_in_range_if_dirty(PGRAPHState *pg, hwaddr start, hwaddr size)
+static bool check_surface_overlaps_range(const SurfaceBinding *surface,
+                                         hwaddr range_start, hwaddr range_len)
+{
+    hwaddr surface_end = surface->vram_addr + surface->size;
+    hwaddr range_end = range_start + range_len;
+    return !(surface->vram_addr >= range_end || range_start >= surface_end);
+}
+
+void pgraph_vk_download_surfaces_in_range_if_dirty(PGRAPHState *pg,
+                                                   hwaddr start, hwaddr size)
 {
     PGRAPHVkState *r = pg->vk_renderer_state;
     SurfaceBinding *surface;
 
-    hwaddr end = start + size - 1;
-
     QTAILQ_FOREACH(surface, &r->surfaces, entry) {
-        hwaddr surf_end = surface->vram_addr + surface->size - 1;
-        bool overlapping = !(surface->vram_addr >= end || start >= surf_end);
-        if (overlapping) {
+        if (check_surface_overlaps_range(surface, start, size)) {
             pgraph_vk_surface_download_if_dirty(
                 container_of(pg, NV2AState, pgraph), surface);
         }
@@ -527,33 +532,54 @@ void pgraph_vk_download_dirty_surfaces(NV2AState *d)
 static void surface_access_callback(void *opaque, MemoryRegion *mr, hwaddr addr,
                                     hwaddr len, bool write)
 {
-    SurfaceBinding *e = opaque;
-    assert(addr >= e->vram_addr);
-    hwaddr offset = addr - e->vram_addr;
-    assert(offset < e->size);
+    NV2AState *d = (NV2AState *)opaque;
+    qemu_mutex_lock(&d->pgraph.lock);
 
-    if (qatomic_read(&e->draw_dirty)) {
-        trace_nv2a_pgraph_surface_cpu_access(e->vram_addr, offset);
-        pgraph_vk_wait_for_surface_download(e);
+    PGRAPHVkState *r = d->pgraph.vk_renderer_state;
+    bool wait_for_downloads = false;
+
+    SurfaceBinding *surface;
+    QTAILQ_FOREACH(surface, &r->surfaces, entry) {
+        if (!check_surface_overlaps_range(surface, addr, len)) {
+            continue;
+        }
+
+        hwaddr offset = addr - surface->vram_addr;
+
+        if (write) {
+            trace_nv2a_pgraph_surface_cpu_write(surface->vram_addr, offset);
+        } else {
+            trace_nv2a_pgraph_surface_cpu_read(surface->vram_addr, offset);
+        }
+
+        if (surface->draw_dirty) {
+            surface->download_pending = true;
+            wait_for_downloads = true;
+        }
+
+        if (write) {
+            surface->upload_pending = true;
+        }
     }
 
-    if (write && !qatomic_read(&e->upload_pending)) {
-        trace_nv2a_pgraph_surface_cpu_access(e->vram_addr, offset);
-        qatomic_set(&e->upload_pending, true);
+    qemu_mutex_unlock(&d->pgraph.lock);
+
+    if (wait_for_downloads) {
+        qemu_mutex_lock(&d->pfifo.lock);
+        qemu_event_reset(&r->downloads_complete);
+        qatomic_set(&r->downloads_pending, true);
+        pfifo_kick(d);
+        qemu_mutex_unlock(&d->pfifo.lock);
+        qemu_event_wait(&r->downloads_complete);
     }
 }
 
 static void register_cpu_access_callback(NV2AState *d, SurfaceBinding *surface)
 {
     if (tcg_enabled()) {
-        qemu_mutex_unlock(&d->pgraph.lock);
-        bql_lock();
-        mem_access_callback_insert(qemu_get_cpu(0),
-            d->vram, surface->vram_addr, surface->size,
-            &surface->access_cb, &surface_access_callback,
-            surface);
-        bql_unlock();
-        qemu_mutex_lock(&d->pgraph.lock);
+        surface->access_cb = mem_access_callback_insert(
+            qemu_get_cpu(0), d->vram, surface->vram_addr, surface->size,
+            &surface_access_callback, d);
     }
 }
 
@@ -561,11 +587,7 @@ static void unregister_cpu_access_callback(NV2AState *d,
                                            SurfaceBinding const *surface)
 {
     if (tcg_enabled()) {
-        qemu_mutex_unlock(&d->pgraph.lock);
-        bql_lock();
         mem_access_callback_remove_by_ref(qemu_get_cpu(0), surface->access_cb);
-        bql_unlock();
-        qemu_mutex_lock(&d->pgraph.lock);
     }
 }
 
@@ -627,24 +649,26 @@ static void invalidate_surface(NV2AState *d, SurfaceBinding *surface)
     QTAILQ_INSERT_HEAD(&r->invalid_surfaces, surface, entry);
 }
 
+static bool check_surfaces_overlap(const SurfaceBinding *surface,
+                                   const SurfaceBinding *other_surface)
+{
+    return check_surface_overlaps_range(surface, other_surface->vram_addr,
+                                        other_surface->size);
+}
+
 static void invalidate_overlapping_surfaces(NV2AState *d,
                                             SurfaceBinding const *surface)
 {
     PGRAPHVkState *r = d->pgraph.vk_renderer_state;
 
-    uintptr_t e_end = surface->vram_addr + surface->size - 1;
-
-    SurfaceBinding *s, *next;
-    QTAILQ_FOREACH_SAFE(s, &r->surfaces, entry, next) {
-        uintptr_t s_end = s->vram_addr + s->size - 1;
-        bool overlapping =
-            !(s->vram_addr > e_end || surface->vram_addr > s_end);
-        if (overlapping) {
+    SurfaceBinding *other_surface, *next_surface;
+    QTAILQ_FOREACH_SAFE (other_surface, &r->surfaces, entry, next_surface) {
+        if (check_surfaces_overlap(surface, other_surface)) {
             trace_nv2a_pgraph_surface_evict_overlapping(
-                s->vram_addr, s->width, s->height,
-                s->pitch);
-            pgraph_vk_surface_download_if_dirty(d, s);
-            invalidate_surface(d, s);
+                other_surface->vram_addr, other_surface->width,
+                other_surface->height, other_surface->pitch);
+            pgraph_vk_surface_download_if_dirty(d, other_surface);
+            invalidate_surface(d, other_surface);
         }
     }
 }

hw/xbox/nv2a/pgraph/vk/texture.c

@@ -1089,12 +1089,9 @@ static void create_texture(PGRAPHState *pg, int texture_idx)
     BasicColorFormatInfo f_basic = kelvin_color_format_info_map[state.color_format];
 
     const hwaddr texture_vram_offset = pgraph_get_texture_phys_addr(pg, texture_idx);
-    size_t texture_palette_data_size;
-    const hwaddr texture_palette_vram_offset =
-        pgraph_get_texture_palette_phys_addr_length(pg, texture_idx,
-                                                    &texture_palette_data_size);
-
     size_t texture_length = pgraph_get_texture_length(pg, &state);
+    hwaddr texture_palette_vram_offset = 0;
+    size_t texture_palette_data_size = 0;
 
     uint32_t filter =
         pgraph_reg_r(pg, NV_PGRAPH_TEXFILTER0 + texture_idx * 4);
@@ -1102,14 +1099,21 @@ static void create_texture(PGRAPHState *pg, int texture_idx)
         pgraph_reg_r(pg, NV_PGRAPH_TEXADDRESS0 + texture_idx * 4);
     uint32_t border_color_pack32 =
         pgraph_reg_r(pg, NV_PGRAPH_BORDERCOLOR0 + texture_idx * 4);
+    bool is_indexed = (state.color_format ==
+            NV097_SET_TEXTURE_FORMAT_COLOR_SZ_I8_A8R8G8B8);
 
     TextureKey key;
     memset(&key, 0, sizeof(key));
     key.state = state;
     key.texture_vram_offset = texture_vram_offset;
     key.texture_length = texture_length;
-    key.palette_vram_offset = texture_palette_vram_offset;
-    key.palette_length = texture_palette_data_size;
+    if (is_indexed) {
+        texture_palette_vram_offset =
+            pgraph_get_texture_palette_phys_addr_length(
+                pg, texture_idx, &texture_palette_data_size);
+        key.palette_vram_offset = texture_palette_vram_offset;
+        key.palette_length = texture_palette_data_size;
+    }
     key.scale = 1;
 
     // FIXME: Separate sampler from texture
@@ -1117,9 +1121,6 @@ static void create_texture(PGRAPHState *pg, int texture_idx)
     key.address = address;
     key.border_color = border_color_pack32;
 
-    bool is_indexed = (state.color_format ==
-            NV097_SET_TEXTURE_FORMAT_COLOR_SZ_I8_A8R8G8B8);
-
     bool possibly_dirty = false;
     bool possibly_dirty_checked = false;
     bool surface_to_texture = false;
@@ -1432,7 +1433,7 @@ void pgraph_vk_bind_textures(NV2AState *d)
     NV2A_VK_DGROUP_END();
 }
 
-static void texture_cache_entry_init(Lru *lru, LruNode *node, void *state)
+static void texture_cache_entry_init(Lru *lru, LruNode *node, const void *state)
 {
     TextureBinding *snode = container_of(node, TextureBinding, node);
 
@@ -1485,7 +1486,8 @@ static void texture_cache_entry_post_evict(Lru *lru, LruNode *node)
     texture_cache_release_node_resources(r, snode);
 }
 
-static bool texture_cache_entry_compare(Lru *lru, LruNode *node, void *key)
+static bool texture_cache_entry_compare(Lru *lru, LruNode *node,
+                                        const void *key)
 {
     TextureBinding *snode = container_of(node, TextureBinding, node);
     return memcmp(&snode->key, key, sizeof(TextureKey));

hw/xbox/nv2a/pgraph/vsh_regs.h

@@ -20,8 +20,8 @@
 #ifndef HW_NV2A_VSH_H
 #define HW_NV2A_VSH_H
 
-#include <stdbool.h>
-#include "qemu/mstring.h"
+#include "qemu/osdep.h"
+#include "hw/xbox/nv2a/nv2a_regs.h"
 
 enum VshLight {
     LIGHT_OFF,
@@ -55,8 +55,6 @@ enum VshFoggen {
     FOGGEN_RADIAL,
     FOGGEN_PLANAR,
     FOGGEN_ABS_PLANAR,
-    FOGGEN_ERROR4,
-    FOGGEN_ERROR5,
     FOGGEN_FOG_X
 };
 
@@ -84,6 +82,60 @@ enum VshSkinning {
 
 #define VSH_TOKEN_SIZE 4
 
+#define VSH_D3DSCM_CORRECTION 96
+
+typedef enum {
+    PARAM_UNKNOWN = 0,
+    PARAM_R,
+    PARAM_V,
+    PARAM_C
+} VshParameterType;
+
+typedef enum {
+    OUTPUT_C = 0,
+    OUTPUT_O
+} VshOutputType;
+
+typedef enum {
+    OMUX_MAC = 0,
+    OMUX_ILU
+} VshOutputMux;
+
+typedef enum {
+    ILU_NOP = 0,
+    ILU_MOV,
+    ILU_RCP,
+    ILU_RCC,
+    ILU_RSQ,
+    ILU_EXP,
+    ILU_LOG,
+    ILU_LIT
+} VshILU;
+
+typedef enum {
+    MAC_NOP,
+    MAC_MOV,
+    MAC_MUL,
+    MAC_ADD,
+    MAC_MAD,
+    MAC_DP3,
+    MAC_DPH,
+    MAC_DP4,
+    MAC_DST,
+    MAC_MIN,
+    MAC_MAX,
+    MAC_SLT,
+    MAC_SGE,
+    MAC_ARL
+} VshMAC;
+
+typedef enum {
+    SWIZZLE_X = 0,
+    SWIZZLE_Y,
+    SWIZZLE_Z,
+    SWIZZLE_W
+} VshSwizzle;
+
 typedef enum {
     FLD_ILU = 0,
     FLD_MAC,
@@ -130,4 +182,30 @@ typedef enum {
 
 uint8_t vsh_get_field(const uint32_t *shader_token, VshFieldName field_name);
 
+enum ShaderPrimitiveMode {
+    PRIM_TYPE_INVALID,
+    PRIM_TYPE_POINTS,
+    PRIM_TYPE_LINES,
+    PRIM_TYPE_LINE_LOOP,
+    PRIM_TYPE_LINE_STRIP,
+    PRIM_TYPE_TRIANGLES,
+    PRIM_TYPE_TRIANGLE_STRIP,
+    PRIM_TYPE_TRIANGLE_FAN,
+    PRIM_TYPE_QUADS,
+    PRIM_TYPE_QUAD_STRIP,
+    PRIM_TYPE_POLYGON,
+};
+
+enum ShaderPolygonMode {
+    POLY_MODE_FILL,
+    POLY_MODE_POINT,
+    POLY_MODE_LINE,
+};
+
+enum MaterialColorSource {
+    MATERIAL_COLOR_SRC_MATERIAL,
+    MATERIAL_COLOR_SRC_DIFFUSE,
+    MATERIAL_COLOR_SRC_SPECULAR,
+};
+
 #endif

hw/xbox/nv2a/trace-events

@@ -11,7 +11,8 @@ nv2a_pgraph_method(uint32_t subchannel, uint32_t graphics_class, uint32_t method
 nv2a_pgraph_method_abbrev(uint32_t subchannel, uint32_t graphics_class, uint32_t method, const char *name, unsigned int count) "%d: 0x%"PRIx32" -> 0x%04"PRIx32" %s * %d"
 nv2a_pgraph_method_unhandled(uint32_t subchannel, uint32_t graphics_class, uint32_t method, uint32_t parameter) "%d: 0x%"PRIx32" -> 0x%04"PRIx32" 0x%"PRIx32
 nv2a_pgraph_surface_compare_mismatch(const char *field, long int a, long int b) "%20s -- %8ld vs %8ld"
-nv2a_pgraph_surface_cpu_access(uint32_t addr, uint32_t offset) "0x%08"PRIx32"+0x%"PRIx32
+nv2a_pgraph_surface_cpu_read(uint32_t addr, uint32_t offset) "0x%08"PRIx32"+0x%"PRIx32
+nv2a_pgraph_surface_cpu_write(uint32_t addr, uint32_t offset) "0x%08"PRIx32"+0x%"PRIx32
 nv2a_pgraph_surface_create_color(uint32_t addr, uint32_t width, uint32_t height, const char *layout, uint32_t anti_aliasing, uint32_t clip_x, uint32_t clip_width, uint32_t clip_y, uint32_t clip_height, uint32_t pitch) "Create: [COLOR @ 0x%08" PRIx32 " (%dx%d)] (%s) aa:%d, clip:x=%d,w=%d,y=%d,h=%d,p=%d"
 nv2a_pgraph_surface_create_zeta(uint32_t addr, uint32_t width, uint32_t height, const char *layout, uint32_t anti_aliasing, uint32_t clip_x, uint32_t clip_width, uint32_t clip_y, uint32_t clip_height, uint32_t pitch) " Create: [ZETA  @ 0x%08" PRIx32 " (%dx%d)] (%s) aa:%d, clip:x=%d,w=%d,y=%d,h=%d,p=%d"
 nv2a_pgraph_surface_download(const char *binding, const char *layout, uint32_t addr, uint32_t width, uint32_t height, uint32_t pitch, uint32_t bytes_per_pixel) "[GPU->RAM] %s (%s) surface @ 0x%08" PRIx32 " (w=%d,h=%d,p=%d,bpp=%d)"