ShuriZma
/
xenia-canary
mirror of https://github.com/xenia-canary/xenia-canary.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'upstream/master' into canary

This commit is contained in:
Margen67 2019-10-21 22:23:17 -07:00
parent c53009906f 1005651855
commit d0e3281741
33 changed files with 2130 additions and 1554 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
docs/building.md
View File

@ -72,7 +72,7 @@ xb genspirv
#### Debugging
VS behaves oddly with the debug paths. Open the xenia project properties
VS behaves oddly with the debug paths. Open the 'xenia-app' project properties
and set the 'Command' to `$(SolutionDir)$(TargetPath)` and the
'Working Directory' to `$(SolutionDir)..\..`. You can specify flags and
the file to run in the 'Command Arguments' field (or use `--flagfile=flags.txt`).

51
src/xenia/base/bit_field.h
View File

@ -1,51 +0,0 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2017 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#ifndef XENIA_BASE_BIT_FIELD_H_
#define XENIA_BASE_BIT_FIELD_H_
#include <cstdint>
#include <cstdlib>
#include <type_traits>
namespace xe {
// Bitfield, where position starts at the LSB.
template <typename T, size_t position, size_t n_bits>
struct bf {
// For enum values, we strip them down to an underlying type.
typedef
typename std::conditional<std::is_enum<T>::value, std::underlying_type<T>,
std::remove_reference<T>>::type::type
value_type;
bf() = default;
inline operator T() const { return value(); }
inline T value() const {
auto value = (storage & mask()) >> position;
if (std::is_signed<value_type>::value) {
// If the value is signed, sign-extend it.
value_type sign_mask = value_type(1) << (n_bits - 1);
value = (sign_mask ^ value) - sign_mask;
}
return static_cast<T>(value);
}
inline value_type mask() const {
return ((value_type(1) << n_bits) - 1) << position;
}
value_type storage;
};
} // namespace xe
#endif // XENIA_BASE_BIT_FIELD_H_

14
src/xenia/gpu/command_processor.cc
View File

@ -20,7 +20,6 @@
#include "xenia/base/ring_buffer.h"
#include "xenia/gpu/gpu_flags.h"
#include "xenia/gpu/graphics_system.h"
#include "xenia/gpu/registers.h"
#include "xenia/gpu/sampler_info.h"
#include "xenia/gpu/texture_info.h"
#include "xenia/gpu/xenos.h"
@ -351,20 +350,20 @@ void CommandProcessor::MakeCoherent() {
// https://cgit.freedesktop.org/xorg/driver/xf86-video-radeonhd/tree/src/r6xx_accel.c?id=3f8b6eccd9dba116cc4801e7f80ce21a879c67d2#n454
RegisterFile* regs = register_file_;
auto status_host = regs->values[XE_GPU_REG_COHER_STATUS_HOST].u32;
auto& status_host = regs->Get<reg::COHER_STATUS_HOST>();
auto base_host = regs->values[XE_GPU_REG_COHER_BASE_HOST].u32;
auto size_host = regs->values[XE_GPU_REG_COHER_SIZE_HOST].u32;
if (!(status_host & 0x80000000ul)) {
if (!status_host.status) {
return;
}
const char* action = "N/A";
if ((status_host & 0x03000000) == 0x03000000) {
if (status_host.vc_action_ena && status_host.tc_action_ena) {
action = "VC | TC";
} else if (status_host & 0x02000000) {
} else if (status_host.tc_action_ena) {
action = "TC";
} else if (status_host & 0x01000000) {
} else if (status_host.vc_action_ena) {
action = "VC";
}
@ -373,8 +372,7 @@ void CommandProcessor::MakeCoherent() {
base_host + size_host, size_host, action);
// Mark coherent.
status_host &= ~0x80000000ul;
regs->values[XE_GPU_REG_COHER_STATUS_HOST].u32 = status_host;
status_host.status = 0;
}
void CommandProcessor::PrepareForWait() { trace_writer_.Flush(); }

2
src/xenia/gpu/command_processor.h
View File

@ -151,7 +151,7 @@ class CommandProcessor {
protected:
struct IndexBufferInfo {
IndexFormat format = IndexFormat::kInt16;
Endian endianness = Endian::kUnspecified;
Endian endianness = Endian::kNone;
uint32_t count = 0;
uint32_t guest_base = 0;
size_t length = 0;

456
src/xenia/gpu/d3d12/d3d12_command_processor.cc
View File

@ -1142,8 +1142,7 @@ bool D3D12CommandProcessor::IssueDraw(PrimitiveType primitive_type,
SCOPE_profile_cpu_f("gpu");
#endif // FINE_GRAINED_DRAW_SCOPES
auto enable_mode = static_cast<xenos::ModeControl>(
regs[XE_GPU_REG_RB_MODECONTROL].u32 & 0x7);
xenos::ModeControl enable_mode = regs.Get<reg::RB_MODECONTROL>().edram_mode;
if (enable_mode == xenos::ModeControl::kIgnore) {
// Ignored.
return true;
@ -1153,7 +1152,7 @@ bool D3D12CommandProcessor::IssueDraw(PrimitiveType primitive_type,
return IssueCopy();
}
if ((regs[XE_GPU_REG_RB_SURFACE_INFO].u32 & 0x3FFF) == 0) {
if (regs.Get<reg::RB_SURFACE_INFO>().surface_pitch == 0) {
// Doesn't actually draw.
// TODO(Triang3l): Do something so memexport still works in this case maybe?
// Unlikely that zero would even really be legal though.
@ -1164,7 +1163,8 @@ bool D3D12CommandProcessor::IssueDraw(PrimitiveType primitive_type,
bool tessellated;
if (uint32_t(primitive_type) >=
uint32_t(PrimitiveType::kExplicitMajorModeForceStart)) {
tessellated = (regs[XE_GPU_REG_VGT_OUTPUT_PATH_CNTL].u32 & 0x3) == 0x1;
tessellated = regs.Get<reg::VGT_OUTPUT_PATH_CNTL>().path_select ==
xenos::VGTOutputPath::kTessellationEnable;
} else {
tessellated = false;
}
@ -1202,8 +1202,9 @@ bool D3D12CommandProcessor::IssueDraw(PrimitiveType primitive_type,
bool memexport_used = memexport_used_vertex || memexport_used_pixel;
bool primitive_two_faced = IsPrimitiveTwoFaced(tessellated, primitive_type);
auto pa_su_sc_mode_cntl = regs.Get<reg::PA_SU_SC_MODE_CNTL>();
if (!memexport_used_vertex && primitive_two_faced &&
(regs[XE_GPU_REG_PA_SU_SC_MODE_CNTL].u32 & 0x3) == 0x3) {
pa_su_sc_mode_cntl.cull_front && pa_su_sc_mode_cntl.cull_back) {
// Both sides are culled - can't be expressed in the pipeline state.
return true;
}
@ -1223,9 +1224,10 @@ bool D3D12CommandProcessor::IssueDraw(PrimitiveType primitive_type,
// tessellation factors (as floats) instead of control point indices.
bool adaptive_tessellation;
if (tessellated) {
TessellationMode tessellation_mode =
TessellationMode(regs[XE_GPU_REG_VGT_HOS_CNTL].u32 & 0x3);
adaptive_tessellation = tessellation_mode == TessellationMode::kAdaptive;
xenos::TessellationMode tessellation_mode =
regs.Get<reg::VGT_HOS_CNTL>().tess_mode;
adaptive_tessellation =
tessellation_mode == xenos::TessellationMode::kAdaptive;
if (adaptive_tessellation &&
(!indexed || index_buffer_info->format != IndexFormat::kInt32)) {
return false;
@ -1235,7 +1237,7 @@ bool D3D12CommandProcessor::IssueDraw(PrimitiveType primitive_type,
// passed to vertex shader registers, especially if patches are drawn with
// an index buffer.
// https://www.slideshare.net/blackdevilvikas/next-generation-graphics-programming-on-xbox-360
if (tessellation_mode != TessellationMode::kAdaptive) {
if (tessellation_mode != xenos::TessellationMode::kAdaptive) {
XELOGE(
"Tessellation mode %u is not implemented yet, only adaptive is "
"partially available now - report the game to Xenia developers!",
@ -1309,20 +1311,16 @@ bool D3D12CommandProcessor::IssueDraw(PrimitiveType primitive_type,
vertex_shader->GetUsedTextureMask(),
pixel_shader != nullptr ? pixel_shader->GetUsedTextureMask() : 0);
// Check if early depth/stencil can be enabled explicitly by RB_DEPTHCONTROL
// or implicitly when alpha test and alpha to coverage are disabled.
uint32_t rb_depthcontrol = regs[XE_GPU_REG_RB_DEPTHCONTROL].u32;
uint32_t rb_colorcontrol = regs[XE_GPU_REG_RB_COLORCONTROL].u32;
bool early_z = false;
if (pixel_shader == nullptr) {
// Check if early depth/stencil can be enabled.
bool early_z;
if (pixel_shader) {
auto rb_colorcontrol = regs.Get<reg::RB_COLORCONTROL>();
early_z = pixel_shader->implicit_early_z_allowed() &&
(!rb_colorcontrol.alpha_test_enable ||
rb_colorcontrol.alpha_func == CompareFunction::kAlways) &&
!rb_colorcontrol.alpha_to_mask_enable;
} else {
early_z = true;
} else if (!pixel_shader->writes_depth()) {
if (rb_depthcontrol & 0x8) {
early_z = true;
} else if (pixel_shader->implicit_early_z_allowed()) {
early_z = (!(rb_colorcontrol & 0x8) || (rb_colorcontrol & 0x7) == 0x7) &&
!(rb_colorcontrol & 0x10);
}
}
// Create the pipeline if needed and bind it.
@ -1347,7 +1345,7 @@ bool D3D12CommandProcessor::IssueDraw(PrimitiveType primitive_type,
// Update system constants before uploading them.
UpdateSystemConstantValues(
memexport_used, primitive_two_faced, line_loop_closing_index,
indexed ? index_buffer_info->endianness : Endian::kUnspecified,
indexed ? index_buffer_info->endianness : Endian::kNone,
adaptive_tessellation ? (index_buffer_info->guest_base & 0x1FFFFFFC) : 0,
early_z, GetCurrentColorMask(pixel_shader), pipeline_render_targets);
@ -1366,22 +1364,19 @@ bool D3D12CommandProcessor::IssueDraw(PrimitiveType primitive_type,
(1ull << (vfetch_index & 63))) {
continue;
}
uint32_t vfetch_constant_index =
XE_GPU_REG_SHADER_CONSTANT_FETCH_00_0 + vfetch_index * 2;
if ((regs[vfetch_constant_index].u32 & 0x3) != 3) {
const auto& vfetch_constant = regs.Get<xenos::xe_gpu_vertex_fetch_t>(
XE_GPU_REG_SHADER_CONSTANT_FETCH_00_0 + vfetch_index * 2);
if (vfetch_constant.type != 3) {
XELOGW("Vertex fetch type is not 3 (fetch constant %u is %.8X %.8X)!",
vfetch_index, regs[vfetch_constant_index].u32,
regs[vfetch_constant_index + 1].u32);
vfetch_index, vfetch_constant.dword_0, vfetch_constant.dword_1);
return false;
}
if (!shared_memory_->RequestRange(
regs[vfetch_constant_index].u32 & 0x1FFFFFFC,
regs[vfetch_constant_index + 1].u32 & 0x3FFFFFC)) {
if (!shared_memory_->RequestRange(vfetch_constant.address << 2,
vfetch_constant.size << 2)) {
XELOGE(
"Failed to request vertex buffer at 0x%.8X (size %u) in the shared "
"memory",
regs[vfetch_constant_index].u32 & 0x1FFFFFFC,
regs[vfetch_constant_index + 1].u32 & 0x3FFFFFC);
vfetch_constant.address << 2, vfetch_constant.size << 2);
return false;
}
vertex_buffers_resident[vfetch_index >> 6] |= 1ull << (vfetch_index & 63);
@ -1400,31 +1395,29 @@ bool D3D12CommandProcessor::IssueDraw(PrimitiveType primitive_type,
const std::vector<uint32_t>& memexport_stream_constants_vertex =
vertex_shader->memexport_stream_constants();
for (uint32_t constant_index : memexport_stream_constants_vertex) {
const xenos::xe_gpu_memexport_stream_t* memexport_stream =
reinterpret_cast<const xenos::xe_gpu_memexport_stream_t*>(
&regs[XE_GPU_REG_SHADER_CONSTANT_000_X + constant_index * 4]);
if (memexport_stream->index_count == 0) {
const auto& memexport_stream = regs.Get<xenos::xe_gpu_memexport_stream_t>(
XE_GPU_REG_SHADER_CONSTANT_000_X + constant_index * 4);
if (memexport_stream.index_count == 0) {
continue;
}
uint32_t memexport_format_size =
GetSupportedMemExportFormatSize(memexport_stream->format);
GetSupportedMemExportFormatSize(memexport_stream.format);
if (memexport_format_size == 0) {
XELOGE(
"Unsupported memexport format %s",
FormatInfo::Get(TextureFormat(uint32_t(memexport_stream->format)))
->name);
XELOGE("Unsupported memexport format %s",
FormatInfo::Get(TextureFormat(uint32_t(memexport_stream.format)))
->name);
return false;
}
uint32_t memexport_base_address = memexport_stream->base_address;
uint32_t memexport_size_dwords =
memexport_stream->index_count * memexport_format_size;
memexport_stream.index_count * memexport_format_size;
// Try to reduce the number of shared memory operations when writing
// different elements into the same buffer through different exports
// (happens in Halo 3).
bool memexport_range_reused = false;
for (uint32_t i = 0; i < memexport_range_count; ++i) {
MemExportRange& memexport_range = memexport_ranges[i];
if (memexport_range.base_address_dwords == memexport_base_address) {
if (memexport_range.base_address_dwords ==
memexport_stream.base_address) {
memexport_range.size_dwords =
std::max(memexport_range.size_dwords, memexport_size_dwords);
memexport_range_reused = true;
@ -1435,7 +1428,7 @@ bool D3D12CommandProcessor::IssueDraw(PrimitiveType primitive_type,
if (!memexport_range_reused) {
MemExportRange& memexport_range =
memexport_ranges[memexport_range_count++];
memexport_range.base_address_dwords = memexport_base_address;
memexport_range.base_address_dwords = memexport_stream.base_address;
memexport_range.size_dwords = memexport_size_dwords;
}
}
@ -1444,28 +1437,26 @@ bool D3D12CommandProcessor::IssueDraw(PrimitiveType primitive_type,
const std::vector<uint32_t>& memexport_stream_constants_pixel =
pixel_shader->memexport_stream_constants();
for (uint32_t constant_index : memexport_stream_constants_pixel) {
const xenos::xe_gpu_memexport_stream_t* memexport_stream =
reinterpret_cast<const xenos::xe_gpu_memexport_stream_t*>(
&regs[XE_GPU_REG_SHADER_CONSTANT_256_X + constant_index * 4]);
if (memexport_stream->index_count == 0) {
const auto& memexport_stream = regs.Get<xenos::xe_gpu_memexport_stream_t>(
XE_GPU_REG_SHADER_CONSTANT_256_X + constant_index * 4);
if (memexport_stream.index_count == 0) {
continue;
}
uint32_t memexport_format_size =
GetSupportedMemExportFormatSize(memexport_stream->format);
GetSupportedMemExportFormatSize(memexport_stream.format);
if (memexport_format_size == 0) {
XELOGE(
"Unsupported memexport format %s",
FormatInfo::Get(TextureFormat(uint32_t(memexport_stream->format)))
->name);
XELOGE("Unsupported memexport format %s",
FormatInfo::Get(TextureFormat(uint32_t(memexport_stream.format)))
->name);
return false;
}
uint32_t memexport_base_address = memexport_stream->base_address;
uint32_t memexport_size_dwords =
memexport_stream->index_count * memexport_format_size;
memexport_stream.index_count * memexport_format_size;
bool memexport_range_reused = false;
for (uint32_t i = 0; i < memexport_range_count; ++i) {
MemExportRange& memexport_range = memexport_ranges[i];
if (memexport_range.base_address_dwords == memexport_base_address) {
if (memexport_range.base_address_dwords ==
memexport_stream.base_address) {
memexport_range.size_dwords =
std::max(memexport_range.size_dwords, memexport_size_dwords);
memexport_range_reused = true;
@ -1475,7 +1466,7 @@ bool D3D12CommandProcessor::IssueDraw(PrimitiveType primitive_type,
if (!memexport_range_reused) {
MemExportRange& memexport_range =
memexport_ranges[memexport_range_count++];
memexport_range.base_address_dwords = memexport_base_address;
memexport_range.base_address_dwords = memexport_stream.base_address;
memexport_range.size_dwords = memexport_size_dwords;
}
}
@ -1850,15 +1841,7 @@ void D3D12CommandProcessor::UpdateFixedFunctionState(bool primitive_two_faced) {
// http://ftp.tku.edu.tw/NetBSD/NetBSD-current/xsrc/external/mit/xf86-video-ati/dist/src/r600_reg_auto_r6xx.h
// See r200UpdateWindow:
// https://github.com/freedreno/mesa/blob/master/src/mesa/drivers/dri/r200/r200_state.c
uint32_t pa_sc_window_offset = regs[XE_GPU_REG_PA_SC_WINDOW_OFFSET].u32;
int16_t window_offset_x = pa_sc_window_offset & 0x7FFF;
int16_t window_offset_y = (pa_sc_window_offset >> 16) & 0x7FFF;
if (window_offset_x & 0x4000) {
window_offset_x |= 0x8000;
}
if (window_offset_y & 0x4000) {
window_offset_y |= 0x8000;
}
auto pa_sc_window_offset = regs.Get<reg::PA_SC_WINDOW_OFFSET>();
// Supersampling replacing multisampling due to difficulties of emulating
// EDRAM with multisampling with RTV/DSV (with ROV, there's MSAA), and also
@ -1868,8 +1851,7 @@ void D3D12CommandProcessor::UpdateFixedFunctionState(bool primitive_two_faced) {
pixel_size_x = 1;
pixel_size_y = 1;
} else {
MsaaSamples msaa_samples =
MsaaSamples((regs[XE_GPU_REG_RB_SURFACE_INFO].u32 >> 16) & 0x3);
MsaaSamples msaa_samples = regs.Get<reg::RB_SURFACE_INFO>().msaa_samples;
pixel_size_x = msaa_samples >= MsaaSamples::k4X ? 2 : 1;
pixel_size_y = msaa_samples >= MsaaSamples::k2X ? 2 : 1;
}
@ -1889,30 +1871,30 @@ void D3D12CommandProcessor::UpdateFixedFunctionState(bool primitive_two_faced) {
// box. If it's not, the position is in screen space. Since we can only use
// the NDC in PC APIs, we use a viewport of the largest possible size, and
// divide the position by it in translated shaders.
uint32_t pa_cl_vte_cntl = regs[XE_GPU_REG_PA_CL_VTE_CNTL].u32;
auto pa_cl_vte_cntl = regs.Get<reg::PA_CL_VTE_CNTL>();
float viewport_scale_x =
(pa_cl_vte_cntl & (1 << 0))
pa_cl_vte_cntl.vport_x_scale_ena
? std::abs(regs[XE_GPU_REG_PA_CL_VPORT_XSCALE].f32)
: 1280.0f;
float viewport_scale_y =
(pa_cl_vte_cntl & (1 << 2))
pa_cl_vte_cntl.vport_y_scale_ena
? std::abs(regs[XE_GPU_REG_PA_CL_VPORT_YSCALE].f32)
: 1280.0f;
float viewport_scale_z = (pa_cl_vte_cntl & (1 << 4))
float viewport_scale_z = pa_cl_vte_cntl.vport_z_scale_ena
? regs[XE_GPU_REG_PA_CL_VPORT_ZSCALE].f32
: 1.0f;
float viewport_offset_x = (pa_cl_vte_cntl & (1 << 1))
float viewport_offset_x = pa_cl_vte_cntl.vport_x_offset_ena
? regs[XE_GPU_REG_PA_CL_VPORT_XOFFSET].f32
: std::abs(viewport_scale_x);
float viewport_offset_y = (pa_cl_vte_cntl & (1 << 3))
float viewport_offset_y = pa_cl_vte_cntl.vport_y_offset_ena
? regs[XE_GPU_REG_PA_CL_VPORT_YOFFSET].f32
: std::abs(viewport_scale_y);
float viewport_offset_z = (pa_cl_vte_cntl & (1 << 5))
float viewport_offset_z = pa_cl_vte_cntl.vport_z_offset_ena
? regs[XE_GPU_REG_PA_CL_VPORT_ZOFFSET].f32
: 0.0f;
if (regs[XE_GPU_REG_PA_SU_SC_MODE_CNTL].u32 & (1 << 16)) {
viewport_offset_x += float(window_offset_x);
viewport_offset_y += float(window_offset_y);
if (regs.Get<reg::PA_SU_SC_MODE_CNTL>().vtx_window_offset_enable) {
viewport_offset_x += float(pa_sc_window_offset.window_x_offset);
viewport_offset_y += float(pa_sc_window_offset.window_y_offset);
}
D3D12_VIEWPORT viewport;
viewport.TopLeftX =
@ -1941,21 +1923,22 @@ void D3D12CommandProcessor::UpdateFixedFunctionState(bool primitive_two_faced) {
}
// Scissor.
uint32_t pa_sc_window_scissor_tl =
regs[XE_GPU_REG_PA_SC_WINDOW_SCISSOR_TL].u32;
uint32_t pa_sc_window_scissor_br =
regs[XE_GPU_REG_PA_SC_WINDOW_SCISSOR_BR].u32;
auto pa_sc_window_scissor_tl = regs.Get<reg::PA_SC_WINDOW_SCISSOR_TL>();
auto pa_sc_window_scissor_br = regs.Get<reg::PA_SC_WINDOW_SCISSOR_BR>();
D3D12_RECT scissor;
scissor.left = pa_sc_window_scissor_tl & 0x7FFF;
scissor.top = (pa_sc_window_scissor_tl >> 16) & 0x7FFF;
scissor.right = pa_sc_window_scissor_br & 0x7FFF;
scissor.bottom = (pa_sc_window_scissor_br >> 16) & 0x7FFF;
if (!(pa_sc_window_scissor_tl & (1u << 31))) {
// !WINDOW_OFFSET_DISABLE.
scissor.left = std::max(scissor.left + window_offset_x, LONG(0));
scissor.top = std::max(scissor.top + window_offset_y, LONG(0));
scissor.right = std::max(scissor.right + window_offset_x, LONG(0));
scissor.bottom = std::max(scissor.bottom + window_offset_y, LONG(0));
scissor.left = pa_sc_window_scissor_tl.tl_x;
scissor.top = pa_sc_window_scissor_tl.tl_y;
scissor.right = pa_sc_window_scissor_br.br_x;
scissor.bottom = pa_sc_window_scissor_br.br_y;
if (!pa_sc_window_scissor_tl.window_offset_disable) {
scissor.left =
std::max(scissor.left + pa_sc_window_offset.window_x_offset, LONG(0));
scissor.top =
std::max(scissor.top + pa_sc_window_offset.window_y_offset, LONG(0));
scissor.right =
std::max(scissor.right + pa_sc_window_offset.window_x_offset, LONG(0));
scissor.bottom =
std::max(scissor.bottom + pa_sc_window_offset.window_y_offset, LONG(0));
}
scissor.left *= pixel_size_x;
scissor.top *= pixel_size_y;
@ -1992,13 +1975,17 @@ void D3D12CommandProcessor::UpdateFixedFunctionState(bool primitive_two_faced) {
// Stencil reference value. Per-face reference not supported by Direct3D 12,
// choose the back face one only if drawing only back faces.
uint32_t stencil_ref;
if (primitive_two_faced && (regs[XE_GPU_REG_RB_DEPTHCONTROL].u32 & 0x80) &&
(regs[XE_GPU_REG_PA_SU_SC_MODE_CNTL].u32 & 0x3) == 1) {
stencil_ref = regs[XE_GPU_REG_RB_STENCILREFMASK_BF].u32 & 0xFF;
Register stencil_ref_mask_reg;
auto pa_su_sc_mode_cntl = regs.Get<reg::PA_SU_SC_MODE_CNTL>();
if (primitive_two_faced &&
regs.Get<reg::RB_DEPTHCONTROL>().backface_enable &&
pa_su_sc_mode_cntl.cull_front && !pa_su_sc_mode_cntl.cull_back) {
stencil_ref_mask_reg = XE_GPU_REG_RB_STENCILREFMASK_BF;
} else {
stencil_ref = regs[XE_GPU_REG_RB_STENCILREFMASK].u32 & 0xFF;
stencil_ref_mask_reg = XE_GPU_REG_RB_STENCILREFMASK;
}
uint32_t stencil_ref =
regs.Get<reg::RB_STENCILREFMASK>(stencil_ref_mask_reg).stencilref;
ff_stencil_ref_update_needed_ |= ff_stencil_ref_ != stencil_ref;
if (ff_stencil_ref_update_needed_) {
ff_stencil_ref_ = stencil_ref;
@ -2019,64 +2006,50 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
SCOPE_profile_cpu_f("gpu");
#endif // FINE_GRAINED_DRAW_SCOPES
uint32_t pa_cl_clip_cntl = regs[XE_GPU_REG_PA_CL_CLIP_CNTL].u32;
uint32_t pa_cl_vte_cntl = regs[XE_GPU_REG_PA_CL_VTE_CNTL].u32;
uint32_t pa_su_point_minmax = regs[XE_GPU_REG_PA_SU_POINT_MINMAX].u32;
uint32_t pa_su_point_size = regs[XE_GPU_REG_PA_SU_POINT_SIZE].u32;
uint32_t pa_su_sc_mode_cntl = regs[XE_GPU_REG_PA_SU_SC_MODE_CNTL].u32;
uint32_t pa_su_vtx_cntl = regs[XE_GPU_REG_PA_SU_VTX_CNTL].u32;
auto pa_cl_clip_cntl = regs.Get<reg::PA_CL_CLIP_CNTL>();
auto pa_cl_vte_cntl = regs.Get<reg::PA_CL_VTE_CNTL>();
auto pa_su_point_minmax = regs.Get<reg::PA_SU_POINT_MINMAX>();
auto pa_su_point_size = regs.Get<reg::PA_SU_POINT_SIZE>();
auto pa_su_sc_mode_cntl = regs.Get<reg::PA_SU_SC_MODE_CNTL>();
auto pa_su_vtx_cntl = regs.Get<reg::PA_SU_VTX_CNTL>();
float rb_alpha_ref = regs[XE_GPU_REG_RB_ALPHA_REF].f32;
uint32_t rb_colorcontrol = regs[XE_GPU_REG_RB_COLORCONTROL].u32;
uint32_t rb_depth_info = regs[XE_GPU_REG_RB_DEPTH_INFO].u32;
uint32_t rb_depthcontrol = regs[XE_GPU_REG_RB_DEPTHCONTROL].u32;
uint32_t rb_stencilrefmask = regs[XE_GPU_REG_RB_STENCILREFMASK].u32;
uint32_t rb_stencilrefmask_bf = regs[XE_GPU_REG_RB_STENCILREFMASK_BF].u32;
uint32_t rb_surface_info = regs[XE_GPU_REG_RB_SURFACE_INFO].u32;
uint32_t sq_context_misc = regs[XE_GPU_REG_SQ_CONTEXT_MISC].u32;
uint32_t sq_program_cntl = regs[XE_GPU_REG_SQ_PROGRAM_CNTL].u32;
auto rb_colorcontrol = regs.Get<reg::RB_COLORCONTROL>();
auto rb_depth_info = regs.Get<reg::RB_DEPTH_INFO>();
auto rb_depthcontrol = regs.Get<reg::RB_DEPTHCONTROL>();
auto rb_stencilrefmask = regs.Get<reg::RB_STENCILREFMASK>();
auto rb_stencilrefmask_bf =
regs.Get<reg::RB_STENCILREFMASK>(XE_GPU_REG_RB_STENCILREFMASK_BF);
auto rb_surface_info = regs.Get<reg::RB_SURFACE_INFO>();
auto sq_context_misc = regs.Get<reg::SQ_CONTEXT_MISC>();
auto sq_program_cntl = regs.Get<reg::SQ_PROGRAM_CNTL>();
int32_t vgt_indx_offset = int32_t(regs[XE_GPU_REG_VGT_INDX_OFFSET].u32);
// Get the color info register values for each render target, and also put
// some safety measures for the ROV path - disable fully aliased render
// targets. Also, for ROV, exclude components that don't exist in the format
// from the write mask.
uint32_t color_infos[4];
ColorRenderTargetFormat color_formats[4];
reg::RB_COLOR_INFO color_infos[4];
float rt_clamp[4][4];
uint32_t rt_keep_masks[4][2];
for (uint32_t i = 0; i < 4; ++i) {
uint32_t color_info;
switch (i) {
case 1:
color_info = regs[XE_GPU_REG_RB_COLOR1_INFO].u32;
break;
case 2:
color_info = regs[XE_GPU_REG_RB_COLOR2_INFO].u32;
break;
case 3:
color_info = regs[XE_GPU_REG_RB_COLOR3_INFO].u32;
break;
default:
color_info = regs[XE_GPU_REG_RB_COLOR_INFO].u32;
}
auto color_info = regs.Get<reg::RB_COLOR_INFO>(
reg::RB_COLOR_INFO::rt_register_indices[i]);
color_infos[i] = color_info;
color_formats[i] = ColorRenderTargetFormat((color_info >> 16) & 0xF);
if (IsROVUsedForEDRAM()) {
// Get the mask for keeping previous color's components unmodified,
// or two UINT32_MAX if no colors actually existing in the RT are written.
DxbcShaderTranslator::ROV_GetColorFormatSystemConstants(
color_formats[i], (color_mask >> (i * 4)) & 0b1111, rt_clamp[i][0],
rt_clamp[i][1], rt_clamp[i][2], rt_clamp[i][3], rt_keep_masks[i][0],
rt_keep_masks[i][1]);
color_info.color_format, (color_mask >> (i * 4)) & 0b1111,
rt_clamp[i][0], rt_clamp[i][1], rt_clamp[i][2], rt_clamp[i][3],
rt_keep_masks[i][0], rt_keep_masks[i][1]);
// Disable the render target if it has the same EDRAM base as another one
// (with a smaller index - assume it's more important).
if (rt_keep_masks[i][0] == UINT32_MAX &&
rt_keep_masks[i][1] == UINT32_MAX) {
uint32_t edram_base = color_info & 0xFFF;
for (uint32_t j = 0; j < i; ++j) {
if (edram_base == (color_infos[j] & 0xFFF) &&
if (color_info.color_base == color_infos[j].color_base &&
(rt_keep_masks[j][0] != UINT32_MAX ||
rt_keep_masks[j][1] != UINT32_MAX)) {
rt_keep_masks[i][0] = UINT32_MAX;
@ -2091,20 +2064,21 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
// Disable depth and stencil if it aliases a color render target (for
// instance, during the XBLA logo in Banjo-Kazooie, though depth writing is
// already disabled there).
if (IsROVUsedForEDRAM() && (rb_depthcontrol & (0x1 | 0x2))) {
uint32_t edram_base_depth = rb_depth_info & 0xFFF;
bool depth_stencil_enabled =
rb_depthcontrol.stencil_enable || rb_depthcontrol.z_enable;
if (IsROVUsedForEDRAM() && depth_stencil_enabled) {
for (uint32_t i = 0; i < 4; ++i) {
if (edram_base_depth == (color_infos[i] & 0xFFF) &&
if (rb_depth_info.depth_base == color_infos[i].color_base &&
(rt_keep_masks[i][0] != UINT32_MAX ||
rt_keep_masks[i][1] != UINT32_MAX)) {
rb_depthcontrol &= ~(uint32_t(0x1 | 0x2));
depth_stencil_enabled = false;
break;
}
}
}
// Get viewport Z scale - needed for flags and ROV output.
float viewport_scale_z = (pa_cl_vte_cntl & (1 << 4))
float viewport_scale_z = pa_cl_vte_cntl.vport_z_scale_ena
? regs[XE_GPU_REG_PA_CL_VPORT_ZSCALE].f32
: 1.0f;
@ -2126,18 +2100,18 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
// = false: multiply the Z coordinate by 1/W0.
// 10: VTX_W0_FMT = true: the incoming W0 is not 1/W0. Perform the reciprocal
// to get 1/W0.
if (pa_cl_vte_cntl & (1 << 8)) {
if (pa_cl_vte_cntl.vtx_xy_fmt) {
flags |= DxbcShaderTranslator::kSysFlag_XYDividedByW;
}
if (pa_cl_vte_cntl & (1 << 9)) {
if (pa_cl_vte_cntl.vtx_z_fmt) {
flags |= DxbcShaderTranslator::kSysFlag_ZDividedByW;
}
if (pa_cl_vte_cntl & (1 << 10)) {
if (pa_cl_vte_cntl.vtx_w0_fmt) {
flags |= DxbcShaderTranslator::kSysFlag_WNotReciprocal;
}
// User clip planes (UCP_ENA_#), when not CLIP_DISABLE.
if (!(pa_cl_clip_cntl & (1 << 16))) {
flags |= (pa_cl_clip_cntl & 0b111111)
if (!pa_cl_clip_cntl.clip_disable) {
flags |= (pa_cl_clip_cntl.value & 0b111111)
<< DxbcShaderTranslator::kSysFlag_UserClipPlane0_Shift;
}
// Reversed depth.
@ -2145,8 +2119,8 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
flags |= DxbcShaderTranslator::kSysFlag_ReverseZ;
}
// Alpha test.
if (rb_colorcontrol & 0x8) {
flags |= (rb_colorcontrol & 0x7)
if (rb_colorcontrol.alpha_test_enable) {
flags |= uint32_t(rb_colorcontrol.alpha_func)
<< DxbcShaderTranslator::kSysFlag_AlphaPassIfLess_Shift;
} else {
flags |= DxbcShaderTranslator::kSysFlag_AlphaPassIfLess |
@ -2154,25 +2128,25 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
DxbcShaderTranslator::kSysFlag_AlphaPassIfGreater;
}
// Alpha to coverage.
if (rb_colorcontrol & 0x10) {
if (rb_colorcontrol.alpha_to_mask_enable) {
flags |= DxbcShaderTranslator::kSysFlag_AlphaToCoverage;
}
// Gamma writing.
for (uint32_t i = 0; i < 4; ++i) {
if (color_formats[i] == ColorRenderTargetFormat::k_8_8_8_8_GAMMA) {
if (color_infos[i].color_format ==
ColorRenderTargetFormat::k_8_8_8_8_GAMMA) {
flags |= DxbcShaderTranslator::kSysFlag_Color0Gamma << i;
}
}
if (IsROVUsedForEDRAM() && (rb_depthcontrol & (0x1 | 0x2))) {
if (IsROVUsedForEDRAM() && depth_stencil_enabled) {
flags |= DxbcShaderTranslator::kSysFlag_ROVDepthStencil;
if (DepthRenderTargetFormat((rb_depth_info >> 16) & 0x1) ==
DepthRenderTargetFormat::kD24FS8) {
if (rb_depth_info.depth_format == DepthRenderTargetFormat::kD24FS8) {
flags |= DxbcShaderTranslator::kSysFlag_ROVDepthFloat24;
}
if (rb_depthcontrol & 0x2) {
flags |= ((rb_depthcontrol >> 4) & 0x7)
if (rb_depthcontrol.z_enable) {
flags |= uint32_t(rb_depthcontrol.zfunc)
<< DxbcShaderTranslator::kSysFlag_ROVDepthPassIfLess_Shift;
if (rb_depthcontrol & 0x4) {
if (rb_depthcontrol.z_write_enable) {
flags |= DxbcShaderTranslator::kSysFlag_ROVDepthWrite;
}
} else {
@ -2182,7 +2156,7 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
DxbcShaderTranslator::kSysFlag_ROVDepthPassIfEqual |
DxbcShaderTranslator::kSysFlag_ROVDepthPassIfGreater;
}
if (rb_depthcontrol & 0x1) {
if (rb_depthcontrol.stencil_enable) {
flags |= DxbcShaderTranslator::kSysFlag_ROVStencilTest;
}
if (early_z) {
@ -2223,9 +2197,9 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
index_endian_and_edge_factors;
// User clip planes (UCP_ENA_#), when not CLIP_DISABLE.
if (!(pa_cl_clip_cntl & (1 << 16))) {
if (!pa_cl_clip_cntl.clip_disable) {
for (uint32_t i = 0; i < 6; ++i) {
if (!(pa_cl_clip_cntl & (1 << i))) {
if (!(pa_cl_clip_cntl.value & (1 << i))) {
continue;
}
const float* ucp = &regs[XE_GPU_REG_PA_CL_UCP_0_X + i * 4].f32;
@ -2249,45 +2223,49 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
// different register (and if there's such register at all).
float viewport_scale_x = regs[XE_GPU_REG_PA_CL_VPORT_XSCALE].f32;
float viewport_scale_y = regs[XE_GPU_REG_PA_CL_VPORT_YSCALE].f32;
// When VPORT_Z_SCALE_ENA is disabled, Z/W is directly what is expected to be
// written to the depth buffer, and for some reason DX_CLIP_SPACE_DEF isn't
// set in this case in draws in games.
bool gl_clip_space_def =
!(pa_cl_clip_cntl & (1 << 19)) && (pa_cl_vte_cntl & (1 << 4));
!pa_cl_clip_cntl.dx_clip_space_def && pa_cl_vte_cntl.vport_z_scale_ena;
float ndc_scale_x, ndc_scale_y, ndc_scale_z;
if (primitive_two_faced && (pa_su_sc_mode_cntl & 0x3) == 0x3) {
if (primitive_two_faced && pa_su_sc_mode_cntl.cull_front &&
pa_su_sc_mode_cntl.cull_back) {
// Kill all primitives if both faces are culled, but the vertex shader still
// needs to do memexport (not NaN because of comparison for setting the
// dirty flag).
ndc_scale_x = ndc_scale_y = ndc_scale_z = 0;
} else {
if (pa_cl_vte_cntl & (1 << 0)) {
if (pa_cl_vte_cntl.vport_x_scale_ena) {
ndc_scale_x = viewport_scale_x >= 0.0f ? 1.0f : -1.0f;
} else {
ndc_scale_x = 1.0f / 1280.0f;
}
if (pa_cl_vte_cntl & (1 << 2)) {
if (pa_cl_vte_cntl.vport_y_scale_ena) {
ndc_scale_y = viewport_scale_y >= 0.0f ? -1.0f : 1.0f;
} else {
ndc_scale_y = -1.0f / 1280.0f;
}
ndc_scale_z = gl_clip_space_def ? 0.5f : 1.0f;
}
float ndc_offset_x = (pa_cl_vte_cntl & (1 << 1)) ? 0.0f : -1.0f;
float ndc_offset_y = (pa_cl_vte_cntl & (1 << 3)) ? 0.0f : 1.0f;
float ndc_offset_x = pa_cl_vte_cntl.vport_x_offset_ena ? 0.0f : -1.0f;
float ndc_offset_y = pa_cl_vte_cntl.vport_y_offset_ena ? 0.0f : 1.0f;
float ndc_offset_z = gl_clip_space_def ? 0.5f : 0.0f;
// Like in OpenGL - VPOS giving pixel centers.
// TODO(Triang3l): Check if ps_param_gen should give center positions in
// OpenGL mode on the Xbox 360.
float pixel_half_pixel_offset = 0.5f;
if (cvars::d3d12_half_pixel_offset && !(pa_su_vtx_cntl & (1 << 0))) {
if (cvars::d3d12_half_pixel_offset && !pa_su_vtx_cntl.pix_center) {
// Signs are hopefully correct here, tested in GTA IV on both clearing
// (without a viewport) and drawing things near the edges of the screen.
if (pa_cl_vte_cntl & (1 << 0)) {
if (pa_cl_vte_cntl.vport_x_scale_ena) {
if (viewport_scale_x != 0.0f) {
ndc_offset_x += 0.5f / viewport_scale_x;
}
} else {
ndc_offset_x += 1.0f / 2560.0f;
}
if (pa_cl_vte_cntl & (1 << 2)) {
if (pa_cl_vte_cntl.vport_y_scale_ena) {
if (viewport_scale_y != 0.0f) {
ndc_offset_y += 0.5f / viewport_scale_y;
}
@ -2313,10 +2291,10 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
system_constants_.pixel_half_pixel_offset = pixel_half_pixel_offset;
// Point size.
float point_size_x = float(pa_su_point_size >> 16) * 0.125f;
float point_size_y = float(pa_su_point_size & 0xFFFF) * 0.125f;
float point_size_min = float(pa_su_point_minmax & 0xFFFF) * 0.125f;
float point_size_max = float(pa_su_point_minmax >> 16) * 0.125f;
float point_size_x = float(pa_su_point_size.width) * 0.125f;
float point_size_y = float(pa_su_point_size.height) * 0.125f;
float point_size_min = float(pa_su_point_minmax.min_size) * 0.125f;
float point_size_max = float(pa_su_point_minmax.max_size) * 0.125f;
dirty |= system_constants_.point_size[0] != point_size_x;
dirty |= system_constants_.point_size[1] != point_size_y;
dirty |= system_constants_.point_size_min_max[0] != point_size_min;
@ -2326,13 +2304,13 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
system_constants_.point_size_min_max[0] = point_size_min;
system_constants_.point_size_min_max[1] = point_size_max;
float point_screen_to_ndc_x, point_screen_to_ndc_y;
if (pa_cl_vte_cntl & (1 << 0)) {
if (pa_cl_vte_cntl.vport_x_scale_ena) {
point_screen_to_ndc_x =
(viewport_scale_x != 0.0f) ? (0.5f / viewport_scale_x) : 0.0f;
} else {
point_screen_to_ndc_x = 1.0f / 2560.0f;
}
if (pa_cl_vte_cntl & (1 << 2)) {
if (pa_cl_vte_cntl.vport_y_scale_ena) {
point_screen_to_ndc_y =
(viewport_scale_y != 0.0f) ? (-0.5f / viewport_scale_y) : 0.0f;
} else {
@ -2345,15 +2323,16 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
// Pixel position register.
uint32_t pixel_pos_reg =
(sq_program_cntl & (1 << 18)) ? (sq_context_misc >> 8) & 0xFF : UINT_MAX;
sq_program_cntl.param_gen ? sq_context_misc.param_gen_pos : UINT_MAX;
dirty |= system_constants_.pixel_pos_reg != pixel_pos_reg;
system_constants_.pixel_pos_reg = pixel_pos_reg;
// Log2 of sample count, for scaling VPOS with SSAA (without ROV) and for
// EDRAM address calculation with MSAA (with ROV).
MsaaSamples msaa_samples = MsaaSamples((rb_surface_info >> 16) & 0x3);
uint32_t sample_count_log2_x = msaa_samples >= MsaaSamples::k4X ? 1 : 0;
uint32_t sample_count_log2_y = msaa_samples >= MsaaSamples::k2X ? 1 : 0;
uint32_t sample_count_log2_x =
rb_surface_info.msaa_samples >= MsaaSamples::k4X ? 1 : 0;
uint32_t sample_count_log2_y =
rb_surface_info.msaa_samples >= MsaaSamples::k2X ? 1 : 0;
dirty |= system_constants_.sample_count_log2[0] != sample_count_log2_x;
dirty |= system_constants_.sample_count_log2[1] != sample_count_log2_y;
system_constants_.sample_count_log2[0] = sample_count_log2_x;
@ -2365,43 +2344,22 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
// EDRAM pitch for ROV writing.
if (IsROVUsedForEDRAM()) {
uint32_t edram_pitch_tiles = ((std::min(rb_surface_info & 0x3FFFu, 2560u) *
(msaa_samples >= MsaaSamples::k4X ? 2 : 1)) +
79) /
80;
uint32_t edram_pitch_tiles =
((std::min(rb_surface_info.surface_pitch, 2560u) *
(rb_surface_info.msaa_samples >= MsaaSamples::k4X ? 2 : 1)) +
79) /
80;
dirty |= system_constants_.edram_pitch_tiles != edram_pitch_tiles;
system_constants_.edram_pitch_tiles = edram_pitch_tiles;
}
// Color exponent bias and output index mapping or ROV render target writing.
bool colorcontrol_blend_enable = (rb_colorcontrol & 0x20) == 0;
for (uint32_t i = 0; i < 4; ++i) {
uint32_t color_info = color_infos[i];
uint32_t blend_factors_ops;
if (colorcontrol_blend_enable) {
switch (i) {
case 1:
blend_factors_ops = regs[XE_GPU_REG_RB_BLENDCONTROL_1].u32;
break;
case 2:
blend_factors_ops = regs[XE_GPU_REG_RB_BLENDCONTROL_2].u32;
break;
case 3:
blend_factors_ops = regs[XE_GPU_REG_RB_BLENDCONTROL_3].u32;
break;
default:
blend_factors_ops = regs[XE_GPU_REG_RB_BLENDCONTROL_0].u32;
break;
}
blend_factors_ops &= 0x1FFF1FFF;
} else {
blend_factors_ops = 0x00010001;
}
reg::RB_COLOR_INFO color_info = color_infos[i];
// Exponent bias is in bits 20:25 of RB_COLOR_INFO.
int32_t color_exp_bias = int32_t(color_info << 6) >> 26;
ColorRenderTargetFormat color_format = color_formats[i];
if (color_format == ColorRenderTargetFormat::k_16_16 ||
color_format == ColorRenderTargetFormat::k_16_16_16_16) {
int32_t color_exp_bias = color_info.color_exp_bias;
if (color_info.color_format == ColorRenderTargetFormat::k_16_16 ||
color_info.color_format == ColorRenderTargetFormat::k_16_16_16_16) {
// On the Xbox 360, k_16_16_EDRAM and k_16_16_16_16_EDRAM internally have
// -32...32 range and expect shaders to give -32...32 values, but they're
// emulated using normalized RG16/RGBA16 when not using the ROV, so the
@ -2427,7 +2385,7 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
system_constants_.edram_rt_keep_mask[i][1] = rt_keep_masks[i][1];
if (rt_keep_masks[i][0] != UINT32_MAX ||
rt_keep_masks[i][1] != UINT32_MAX) {
uint32_t rt_base_dwords_scaled = (color_info & 0xFFF) * 1280;
uint32_t rt_base_dwords_scaled = color_info.color_base * 1280;
if (texture_cache_->IsResolutionScale2X()) {
rt_base_dwords_scaled <<= 2;
}
@ -2435,8 +2393,8 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
rt_base_dwords_scaled;
system_constants_.edram_rt_base_dwords_scaled[i] =
rt_base_dwords_scaled;
uint32_t format_flags =
DxbcShaderTranslator::ROV_AddColorFormatFlags(color_format);
uint32_t format_flags = DxbcShaderTranslator::ROV_AddColorFormatFlags(
color_info.color_format);
dirty |= system_constants_.edram_rt_format_flags[i] != format_flags;
system_constants_.edram_rt_format_flags[i] = format_flags;
// Can't do float comparisons here because NaNs would result in always
@ -2445,6 +2403,8 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
4 * sizeof(float)) != 0;
std::memcpy(system_constants_.edram_rt_clamp[i], rt_clamp[i],
4 * sizeof(float));
uint32_t blend_factors_ops =
regs[reg::RB_BLENDCONTROL::rt_register_indices[i]].u32 & 0x1FFF1FFF;
dirty |= system_constants_.edram_rt_blend_factors_ops[i] !=
blend_factors_ops;
system_constants_.edram_rt_blend_factors_ops[i] = blend_factors_ops;
@ -2465,7 +2425,7 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
resolution_square_scale;
system_constants_.edram_resolution_square_scale = resolution_square_scale;
uint32_t depth_base_dwords = (rb_depth_info & 0xFFF) * 1280;
uint32_t depth_base_dwords = rb_depth_info.depth_base * 1280;
dirty |= system_constants_.edram_depth_base_dwords != depth_base_dwords;
system_constants_.edram_depth_base_dwords = depth_base_dwords;
@ -2474,7 +2434,7 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
float depth_range_scale = std::abs(viewport_scale_z);
dirty |= system_constants_.edram_depth_range_scale != depth_range_scale;
system_constants_.edram_depth_range_scale = depth_range_scale;
float depth_range_offset = (pa_cl_vte_cntl & (1 << 5))
float depth_range_offset = pa_cl_vte_cntl.vport_z_offset_ena
? regs[XE_GPU_REG_PA_CL_VPORT_ZOFFSET].f32
: 0.0f;
if (viewport_scale_z < 0.0f) {
@ -2490,20 +2450,20 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
float poly_offset_front_scale = 0.0f, poly_offset_front_offset = 0.0f;
float poly_offset_back_scale = 0.0f, poly_offset_back_offset = 0.0f;
if (primitive_two_faced) {
if (pa_su_sc_mode_cntl & (1 << 11)) {
if (pa_su_sc_mode_cntl.poly_offset_front_enable) {
poly_offset_front_scale =
regs[XE_GPU_REG_PA_SU_POLY_OFFSET_FRONT_SCALE].f32;
poly_offset_front_offset =
regs[XE_GPU_REG_PA_SU_POLY_OFFSET_FRONT_OFFSET].f32;
}
if (pa_su_sc_mode_cntl & (1 << 12)) {
if (pa_su_sc_mode_cntl.poly_offset_back_enable) {
poly_offset_back_scale =
regs[XE_GPU_REG_PA_SU_POLY_OFFSET_BACK_SCALE].f32;
poly_offset_back_offset =
regs[XE_GPU_REG_PA_SU_POLY_OFFSET_BACK_OFFSET].f32;
}
} else {
if (pa_su_sc_mode_cntl & (1 << 13)) {
if (pa_su_sc_mode_cntl.poly_offset_para_enable) {
poly_offset_front_scale =
regs[XE_GPU_REG_PA_SU_POLY_OFFSET_FRONT_SCALE].f32;
poly_offset_front_offset =
@ -2533,39 +2493,43 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
poly_offset_back_offset;
system_constants_.edram_poly_offset_back_offset = poly_offset_back_offset;
if (rb_depthcontrol & 0x1) {
uint32_t stencil_value;
stencil_value = rb_stencilrefmask & 0xFF;
dirty |= system_constants_.edram_stencil_front_reference != stencil_value;
system_constants_.edram_stencil_front_reference = stencil_value;
stencil_value = (rb_stencilrefmask >> 8) & 0xFF;
dirty |= system_constants_.edram_stencil_front_read_mask != stencil_value;
system_constants_.edram_stencil_front_read_mask = stencil_value;
stencil_value = (rb_stencilrefmask >> 16) & 0xFF;
if (depth_stencil_enabled && rb_depthcontrol.stencil_enable) {
dirty |= system_constants_.edram_stencil_front_reference !=
rb_stencilrefmask.stencilref;
system_constants_.edram_stencil_front_reference =
rb_stencilrefmask.stencilref;
dirty |= system_constants_.edram_stencil_front_read_mask !=
rb_stencilrefmask.stencilmask;
system_constants_.edram_stencil_front_read_mask =
rb_stencilrefmask.stencilmask;
dirty |= system_constants_.edram_stencil_front_write_mask !=
rb_stencilrefmask.stencilwritemask;
system_constants_.edram_stencil_front_write_mask =
rb_stencilrefmask.stencilwritemask;
uint32_t stencil_func_ops =
(rb_depthcontrol.value >> 8) & ((1 << 12) - 1);
dirty |=
system_constants_.edram_stencil_front_write_mask != stencil_value;
system_constants_.edram_stencil_front_write_mask = stencil_value;
stencil_value = (rb_depthcontrol >> 8) & ((1 << 12) - 1);
dirty |= system_constants_.edram_stencil_front_func_ops != stencil_value;
system_constants_.edram_stencil_front_func_ops = stencil_value;
system_constants_.edram_stencil_front_func_ops != stencil_func_ops;
system_constants_.edram_stencil_front_func_ops = stencil_func_ops;
if (primitive_two_faced && (rb_depthcontrol & 0x80)) {
stencil_value = rb_stencilrefmask_bf & 0xFF;
dirty |=
system_constants_.edram_stencil_back_reference != stencil_value;
system_constants_.edram_stencil_back_reference = stencil_value;
stencil_value = (rb_stencilrefmask_bf >> 8) & 0xFF;
dirty |=
system_constants_.edram_stencil_back_read_mask != stencil_value;
system_constants_.edram_stencil_back_read_mask = stencil_value;
stencil_value = (rb_stencilrefmask_bf >> 16) & 0xFF;
dirty |=
system_constants_.edram_stencil_back_write_mask != stencil_value;
system_constants_.edram_stencil_back_write_mask = stencil_value;
stencil_value = (rb_depthcontrol >> 20) & ((1 << 12) - 1);
dirty |= system_constants_.edram_stencil_back_func_ops != stencil_value;
system_constants_.edram_stencil_back_func_ops = stencil_value;
if (primitive_two_faced && rb_depthcontrol.backface_enable) {
dirty |= system_constants_.edram_stencil_back_reference !=
rb_stencilrefmask_bf.stencilref;
system_constants_.edram_stencil_back_reference =
rb_stencilrefmask_bf.stencilref;
dirty |= system_constants_.edram_stencil_back_read_mask !=
rb_stencilrefmask_bf.stencilmask;
system_constants_.edram_stencil_back_read_mask =
rb_stencilrefmask_bf.stencilmask;
dirty |= system_constants_.edram_stencil_back_write_mask !=
rb_stencilrefmask_bf.stencilwritemask;
system_constants_.edram_stencil_back_write_mask =
rb_stencilrefmask_bf.stencilwritemask;
uint32_t stencil_func_ops_bf =
(rb_depthcontrol.value >> 20) & ((1 << 12) - 1);
dirty |= system_constants_.edram_stencil_back_func_ops !=
stencil_func_ops_bf;
system_constants_.edram_stencil_back_func_ops = stencil_func_ops_bf;
} else {
dirty |= std::memcmp(system_constants_.edram_stencil_back,
system_constants_.edram_stencil_front,

221
src/xenia/gpu/d3d12/pipeline_cache.cc
View File

@ -207,8 +207,17 @@ bool PipelineCache::EnsureShadersTranslated(D3D12Shader* vertex_shader,
assert_true(regs[XE_GPU_REG_SQ_PS_CONST].u32 == 0x000FF100 ||
regs[XE_GPU_REG_SQ_PS_CONST].u32 == 0x00000000);
xenos::xe_gpu_program_cntl_t sq_program_cntl;
sq_program_cntl.dword_0 = regs[XE_GPU_REG_SQ_PROGRAM_CNTL].u32;
auto sq_program_cntl = regs.Get<reg::SQ_PROGRAM_CNTL>();
// Normal vertex shaders only, for now.
assert_true(sq_program_cntl.vs_export_mode ==
xenos::VertexShaderExportMode::kPosition1Vector ||
sq_program_cntl.vs_export_mode ==
xenos::VertexShaderExportMode::kPosition2VectorsSprite ||
sq_program_cntl.vs_export_mode ==
xenos::VertexShaderExportMode::kMultipass);
assert_false(sq_program_cntl.gen_index_vtx);
if (!vertex_shader->is_translated() &&
!TranslateShader(vertex_shader, sq_program_cntl, tessellated,
primitive_type)) {
@ -294,8 +303,7 @@ bool PipelineCache::ConfigurePipeline(
}
bool PipelineCache::TranslateShader(D3D12Shader* shader,
xenos::xe_gpu_program_cntl_t cntl,
bool tessellated,
reg::SQ_PROGRAM_CNTL cntl, bool tessellated,
PrimitiveType primitive_type) {
// Perform translation.
// If this fails the shader will be marked as invalid and ignored later.
@ -355,7 +363,7 @@ bool PipelineCache::GetCurrentStateDescription(
const RenderTargetCache::PipelineRenderTarget render_targets[5],
PipelineDescription& description_out) {
auto& regs = *register_file_;
uint32_t pa_su_sc_mode_cntl = regs[XE_GPU_REG_PA_SU_SC_MODE_CNTL].u32;
auto pa_su_sc_mode_cntl = regs.Get<reg::PA_SU_SC_MODE_CNTL>();
bool primitive_two_faced = IsPrimitiveTwoFaced(tessellated, primitive_type);
// Initialize all unused fields to zero for comparison/hashing.
@ -373,7 +381,7 @@ bool PipelineCache::GetCurrentStateDescription(
description_out.pixel_shader = pixel_shader;
// Index buffer strip cut value.
if (pa_su_sc_mode_cntl & (1 << 21)) {
if (pa_su_sc_mode_cntl.multi_prim_ib_ena) {
// Not using 0xFFFF with 32-bit indices because in index buffers it will be
// 0xFFFF0000 anyway due to endianness.
description_out.strip_cut_index = index_format == IndexFormat::kInt32
@ -385,12 +393,12 @@ bool PipelineCache::GetCurrentStateDescription(
// Primitive topology type, tessellation mode and geometry shader.
if (tessellated) {
switch (TessellationMode(regs[XE_GPU_REG_VGT_HOS_CNTL].u32 & 0x3)) {
case TessellationMode::kContinuous:
switch (regs.Get<reg::VGT_HOS_CNTL>().tess_mode) {
case xenos::TessellationMode::kContinuous:
description_out.tessellation_mode =
PipelineTessellationMode::kContinuous;
break;
case TessellationMode::kAdaptive:
case xenos::TessellationMode::kAdaptive:
description_out.tessellation_mode =
cvars::d3d12_tessellation_adaptive
? PipelineTessellationMode::kAdaptive
@ -471,53 +479,60 @@ bool PipelineCache::GetCurrentStateDescription(
// Xenos fill mode 1).
// Here we also assume that only one side is culled - if two sides are culled,
// the D3D12 command processor will drop such draw early.
uint32_t cull_mode = primitive_two_faced ? (pa_su_sc_mode_cntl & 0x3) : 0;
bool cull_front, cull_back;
if (primitive_two_faced) {
cull_front = pa_su_sc_mode_cntl.cull_front != 0;
cull_back = pa_su_sc_mode_cntl.cull_back != 0;
} else {
cull_front = false;
cull_back = false;
}
float poly_offset = 0.0f, poly_offset_scale = 0.0f;
if (primitive_two_faced) {
description_out.front_counter_clockwise = (pa_su_sc_mode_cntl & 0x4) == 0;
if (cull_mode == 1) {
description_out.front_counter_clockwise = pa_su_sc_mode_cntl.face == 0;
if (cull_front) {
description_out.cull_mode = PipelineCullMode::kFront;
} else if (cull_mode == 2) {
} else if (cull_back) {
description_out.cull_mode = PipelineCullMode::kBack;
} else {
description_out.cull_mode = PipelineCullMode::kNone;
}
// With ROV, the depth bias is applied in the pixel shader because
// per-sample depth is needed for MSAA.
if (cull_mode != 1) {
if (!cull_front) {
// Front faces aren't culled.
uint32_t fill_mode = (pa_su_sc_mode_cntl >> 5) & 0x7;
if (fill_mode == 0 || fill_mode == 1) {
// Direct3D 12, unfortunately, doesn't support point fill mode.
if (pa_su_sc_mode_cntl.polymode_front_ptype !=
xenos::PolygonType::kTriangles) {
description_out.fill_mode_wireframe = 1;
}
if (!edram_rov_used_ && (pa_su_sc_mode_cntl & (1 << 11))) {
if (!edram_rov_used_ && pa_su_sc_mode_cntl.poly_offset_front_enable) {
poly_offset = regs[XE_GPU_REG_PA_SU_POLY_OFFSET_FRONT_OFFSET].f32;
poly_offset_scale = regs[XE_GPU_REG_PA_SU_POLY_OFFSET_FRONT_SCALE].f32;
}
}
if (cull_mode != 2) {
if (!cull_back) {
// Back faces aren't culled.
uint32_t fill_mode = (pa_su_sc_mode_cntl >> 8) & 0x7;
if (fill_mode == 0 || fill_mode == 1) {
if (pa_su_sc_mode_cntl.polymode_back_ptype !=
xenos::PolygonType::kTriangles) {
description_out.fill_mode_wireframe = 1;
}
// Prefer front depth bias because in general, front faces are the ones
// that are rendered (except for shadow volumes).
if (!edram_rov_used_ && (pa_su_sc_mode_cntl & (1 << 12)) &&
if (!edram_rov_used_ && pa_su_sc_mode_cntl.poly_offset_back_enable &&
poly_offset == 0.0f && poly_offset_scale == 0.0f) {
poly_offset = regs[XE_GPU_REG_PA_SU_POLY_OFFSET_BACK_OFFSET].f32;
poly_offset_scale = regs[XE_GPU_REG_PA_SU_POLY_OFFSET_BACK_SCALE].f32;
}
}
if (((pa_su_sc_mode_cntl >> 3) & 0x3) == 0) {
// Fill mode is disabled.
if (pa_su_sc_mode_cntl.poly_mode == xenos::PolygonModeEnable::kDisabled) {
description_out.fill_mode_wireframe = 0;
}
} else {
// Filled front faces only.
// Use front depth bias if POLY_OFFSET_PARA_ENABLED
// (POLY_OFFSET_FRONT_ENABLED is for two-sided primitives).
if (!edram_rov_used_ && (pa_su_sc_mode_cntl & (1 << 13))) {
if (!edram_rov_used_ && pa_su_sc_mode_cntl.poly_offset_para_enable) {
poly_offset = regs[XE_GPU_REG_PA_SU_POLY_OFFSET_FRONT_OFFSET].f32;
poly_offset_scale = regs[XE_GPU_REG_PA_SU_POLY_OFFSET_FRONT_SCALE].f32;
}
@ -535,8 +550,8 @@ bool PipelineCache::GetCurrentStateDescription(
// of Duty 4 (vehicledamage map explosion decals) and Red Dead Redemption
// (shadows - 2^17 is not enough, 2^18 hasn't been tested, but 2^19
// eliminates the acne).
if (((register_file_->values[XE_GPU_REG_RB_DEPTH_INFO].u32 >> 16) & 0x1) ==
uint32_t(DepthRenderTargetFormat::kD24FS8)) {
if (regs.Get<reg::RB_DEPTH_INFO>().depth_format ==
DepthRenderTargetFormat::kD24FS8) {
poly_offset *= float(1 << 19);
} else {
poly_offset *= float(1 << 23);
@ -556,58 +571,49 @@ bool PipelineCache::GetCurrentStateDescription(
primitive_type == PrimitiveType::kQuadPatch)) {
description_out.fill_mode_wireframe = 1;
}
// CLIP_DISABLE
description_out.depth_clip =
(regs[XE_GPU_REG_PA_CL_CLIP_CNTL].u32 & (1 << 16)) == 0;
// TODO(DrChat): This seem to differ. Need to examine this.
// https://github.com/decaf-emu/decaf-emu/blob/c017a9ff8128852fb9a5da19466778a171cea6e1/src/libdecaf/src/gpu/latte_registers_pa.h#L11
// ZCLIP_NEAR_DISABLE
// description_out.depth_clip = (PA_CL_CLIP_CNTL & (1 << 26)) == 0;
// RASTERIZER_DISABLE
// Disable rendering in command processor if PA_CL_CLIP_CNTL & (1 << 22)?
description_out.depth_clip = !regs.Get<reg::PA_CL_CLIP_CNTL>().clip_disable;
if (edram_rov_used_) {
description_out.rov_msaa =
((regs[XE_GPU_REG_RB_SURFACE_INFO].u32 >> 16) & 0x3) != 0;
}
if (!edram_rov_used_) {
uint32_t rb_colorcontrol = regs[XE_GPU_REG_RB_COLORCONTROL].u32;
regs.Get<reg::RB_SURFACE_INFO>().msaa_samples != MsaaSamples::k1X;
} else {
// Depth/stencil. No stencil, always passing depth test and no depth writing
// means depth disabled.
if (render_targets[4].format != DXGI_FORMAT_UNKNOWN) {
uint32_t rb_depthcontrol = regs[XE_GPU_REG_RB_DEPTHCONTROL].u32;
if (rb_depthcontrol & 0x2) {
description_out.depth_func = (rb_depthcontrol >> 4) & 0x7;
description_out.depth_write = (rb_depthcontrol & 0x4) != 0;
auto rb_depthcontrol = regs.Get<reg::RB_DEPTHCONTROL>();
if (rb_depthcontrol.z_enable) {
description_out.depth_func = rb_depthcontrol.zfunc;
description_out.depth_write = rb_depthcontrol.z_write_enable;
} else {
description_out.depth_func = 0b111;
description_out.depth_func = CompareFunction::kAlways;
}
if (rb_depthcontrol & 0x1) {
if (rb_depthcontrol.stencil_enable) {
description_out.stencil_enable = 1;
bool stencil_backface_enable =
primitive_two_faced && (rb_depthcontrol & 0x80);
uint32_t stencil_masks;
primitive_two_faced && rb_depthcontrol.backface_enable;
// Per-face masks not supported by Direct3D 12, choose the back face
// ones only if drawing only back faces.
if (stencil_backface_enable && cull_mode == 1) {
stencil_masks = regs[XE_GPU_REG_RB_STENCILREFMASK_BF].u32;
Register stencil_ref_mask_reg;
if (stencil_backface_enable && cull_front) {
stencil_ref_mask_reg = XE_GPU_REG_RB_STENCILREFMASK_BF;
} else {
stencil_masks = regs[XE_GPU_REG_RB_STENCILREFMASK].u32;
stencil_ref_mask_reg = XE_GPU_REG_RB_STENCILREFMASK;
}
description_out.stencil_read_mask = (stencil_masks >> 8) & 0xFF;
description_out.stencil_write_mask = (stencil_masks >> 16) & 0xFF;
description_out.stencil_front_fail_op = (rb_depthcontrol >> 11) & 0x7;
auto stencil_ref_mask =
regs.Get<reg::RB_STENCILREFMASK>(stencil_ref_mask_reg);
description_out.stencil_read_mask = stencil_ref_mask.stencilmask;
description_out.stencil_write_mask = stencil_ref_mask.stencilwritemask;
description_out.stencil_front_fail_op = rb_depthcontrol.stencilfail;
description_out.stencil_front_depth_fail_op =
(rb_depthcontrol >> 17) & 0x7;
description_out.stencil_front_pass_op = (rb_depthcontrol >> 14) & 0x7;
description_out.stencil_front_func = (rb_depthcontrol >> 8) & 0x7;
rb_depthcontrol.stencilzfail;
description_out.stencil_front_pass_op = rb_depthcontrol.stencilzpass;
description_out.stencil_front_func = rb_depthcontrol.stencilfunc;
if (stencil_backface_enable) {
description_out.stencil_back_fail_op = (rb_depthcontrol >> 23) & 0x7;
description_out.stencil_back_fail_op = rb_depthcontrol.stencilfail_bf;
description_out.stencil_back_depth_fail_op =
(rb_depthcontrol >> 29) & 0x7;
description_out.stencil_back_pass_op = (rb_depthcontrol >> 26) & 0x7;
description_out.stencil_back_func = (rb_depthcontrol >> 20) & 0x7;
rb_depthcontrol.stencilzfail_bf;
description_out.stencil_back_pass_op =
rb_depthcontrol.stencilzpass_bf;
description_out.stencil_back_func = rb_depthcontrol.stencilfunc_bf;
} else {
description_out.stencil_back_fail_op =
description_out.stencil_front_fail_op;
@ -620,13 +626,13 @@ bool PipelineCache::GetCurrentStateDescription(
}
}
// If not binding the DSV, ignore the format in the hash.
if (description_out.depth_func != 0b111 || description_out.depth_write ||
description_out.stencil_enable) {
description_out.depth_format = DepthRenderTargetFormat(
(regs[XE_GPU_REG_RB_DEPTH_INFO].u32 >> 16) & 1);
if (description_out.depth_func != CompareFunction::kAlways ||
description_out.depth_write || description_out.stencil_enable) {
description_out.depth_format =
regs.Get<reg::RB_DEPTH_INFO>().depth_format;
}
} else {
description_out.depth_func = 0b111;
description_out.depth_func = CompareFunction::kAlways;
}
if (early_z) {
description_out.force_early_z = 1;
@ -686,38 +692,25 @@ bool PipelineCache::GetCurrentStateDescription(
if (render_targets[i].format == DXGI_FORMAT_UNKNOWN) {
break;
}
uint32_t guest_rt_index = render_targets[i].guest_render_target;
uint32_t color_info, blendcontrol;
switch (guest_rt_index) {
case 1:
color_info = regs[XE_GPU_REG_RB_COLOR1_INFO].u32;
blendcontrol = regs[XE_GPU_REG_RB_BLENDCONTROL_1].u32;
break;
case 2:
color_info = regs[XE_GPU_REG_RB_COLOR2_INFO].u32;
blendcontrol = regs[XE_GPU_REG_RB_BLENDCONTROL_2].u32;
break;
case 3:
color_info = regs[XE_GPU_REG_RB_COLOR3_INFO].u32;
blendcontrol = regs[XE_GPU_REG_RB_BLENDCONTROL_3].u32;
break;
default:
color_info = regs[XE_GPU_REG_RB_COLOR_INFO].u32;
blendcontrol = regs[XE_GPU_REG_RB_BLENDCONTROL_0].u32;
break;
}
PipelineRenderTarget& rt = description_out.render_targets[i];
rt.used = 1;
rt.format = RenderTargetCache::GetBaseColorFormat(
ColorRenderTargetFormat((color_info >> 16) & 0xF));
uint32_t guest_rt_index = render_targets[i].guest_render_target;
auto color_info = regs.Get<reg::RB_COLOR_INFO>(
reg::RB_COLOR_INFO::rt_register_indices[guest_rt_index]);
rt.format =
RenderTargetCache::GetBaseColorFormat(color_info.color_format);
rt.write_mask = (color_mask >> (guest_rt_index * 4)) & 0xF;
if (!(rb_colorcontrol & 0x20) && rt.write_mask) {
rt.src_blend = kBlendFactorMap[blendcontrol & 0x1F];
rt.dest_blend = kBlendFactorMap[(blendcontrol >> 8) & 0x1F];
rt.blend_op = BlendOp((blendcontrol >> 5) & 0x7);
rt.src_blend_alpha = kBlendFactorAlphaMap[(blendcontrol >> 16) & 0x1F];
rt.dest_blend_alpha = kBlendFactorAlphaMap[(blendcontrol >> 24) & 0x1F];
rt.blend_op_alpha = BlendOp((blendcontrol >> 21) & 0x7);
if (rt.write_mask) {
auto blendcontrol = regs.Get<reg::RB_BLENDCONTROL>(
reg::RB_BLENDCONTROL::rt_register_indices[guest_rt_index]);
rt.src_blend = kBlendFactorMap[uint32_t(blendcontrol.color_srcblend)];
rt.dest_blend = kBlendFactorMap[uint32_t(blendcontrol.color_destblend)];
rt.blend_op = blendcontrol.color_comb_fcn;
rt.src_blend_alpha =
kBlendFactorAlphaMap[uint32_t(blendcontrol.alpha_srcblend)];
rt.dest_blend_alpha =
kBlendFactorAlphaMap[uint32_t(blendcontrol.alpha_destblend)];
rt.blend_op_alpha = blendcontrol.alpha_comb_fcn;
} else {
rt.src_blend = PipelineBlendFactor::kOne;
rt.dest_blend = PipelineBlendFactor::kZero;
@ -943,15 +936,17 @@ ID3D12PipelineState* PipelineCache::CreatePipelineState(
if (!edram_rov_used_) {
// Depth/stencil.
if (description.depth_func != 0b111 || description.depth_write) {
if (description.depth_func != CompareFunction::kAlways ||
description.depth_write) {
state_desc.DepthStencilState.DepthEnable = TRUE;
state_desc.DepthStencilState.DepthWriteMask =
description.depth_write ? D3D12_DEPTH_WRITE_MASK_ALL
: D3D12_DEPTH_WRITE_MASK_ZERO;
// Comparison functions are the same in Direct3D 12 but plus one (minus
// one, bit 0 for less, bit 1 for equal, bit 2 for greater).
state_desc.DepthStencilState.DepthFunc = D3D12_COMPARISON_FUNC(
uint32_t(D3D12_COMPARISON_FUNC_NEVER) + description.depth_func);
state_desc.DepthStencilState.DepthFunc =
D3D12_COMPARISON_FUNC(uint32_t(D3D12_COMPARISON_FUNC_NEVER) +
uint32_t(description.depth_func));
}
if (description.stencil_enable) {
state_desc.DepthStencilState.StencilEnable = TRUE;
@ -960,26 +955,30 @@ ID3D12PipelineState* PipelineCache::CreatePipelineState(
state_desc.DepthStencilState.StencilWriteMask =
description.stencil_write_mask;
// Stencil operations are the same in Direct3D 12 too but plus one.
state_desc.DepthStencilState.FrontFace.StencilFailOp = D3D12_STENCIL_OP(
uint32_t(D3D12_STENCIL_OP_KEEP) + description.stencil_front_fail_op);
state_desc.DepthStencilState.FrontFace.StencilFailOp =
D3D12_STENCIL_OP(uint32_t(D3D12_STENCIL_OP_KEEP) +
uint32_t(description.stencil_front_fail_op));
state_desc.DepthStencilState.FrontFace.StencilDepthFailOp =
D3D12_STENCIL_OP(uint32_t(D3D12_STENCIL_OP_KEEP) +
description.stencil_front_depth_fail_op);
state_desc.DepthStencilState.FrontFace.StencilPassOp = D3D12_STENCIL_OP(
uint32_t(D3D12_STENCIL_OP_KEEP) + description.stencil_front_pass_op);
uint32_t(description.stencil_front_depth_fail_op));
state_desc.DepthStencilState.FrontFace.StencilPassOp =
D3D12_STENCIL_OP(uint32_t(D3D12_STENCIL_OP_KEEP) +
uint32_t(description.stencil_front_pass_op));
state_desc.DepthStencilState.FrontFace.StencilFunc =
D3D12_COMPARISON_FUNC(uint32_t(D3D12_COMPARISON_FUNC_NEVER) +
description.stencil_front_func);
state_desc.DepthStencilState.BackFace.StencilFailOp = D3D12_STENCIL_OP(
uint32_t(D3D12_STENCIL_OP_KEEP) + description.stencil_back_fail_op);
uint32_t(description.stencil_front_func));
state_desc.DepthStencilState.BackFace.StencilFailOp =
D3D12_STENCIL_OP(uint32_t(D3D12_STENCIL_OP_KEEP) +
uint32_t(description.stencil_back_fail_op));
state_desc.DepthStencilState.BackFace.StencilDepthFailOp =
D3D12_STENCIL_OP(uint32_t(D3D12_STENCIL_OP_KEEP) +
description.stencil_back_depth_fail_op);
state_desc.DepthStencilState.BackFace.StencilPassOp = D3D12_STENCIL_OP(
uint32_t(D3D12_STENCIL_OP_KEEP) + description.stencil_back_pass_op);
uint32_t(description.stencil_back_depth_fail_op));
state_desc.DepthStencilState.BackFace.StencilPassOp =
D3D12_STENCIL_OP(uint32_t(D3D12_STENCIL_OP_KEEP) +
uint32_t(description.stencil_back_pass_op));
state_desc.DepthStencilState.BackFace.StencilFunc =
D3D12_COMPARISON_FUNC(uint32_t(D3D12_COMPARISON_FUNC_NEVER) +
description.stencil_back_func);
uint32_t(description.stencil_back_func));
}
if (state_desc.DepthStencilState.DepthEnable ||
state_desc.DepthStencilState.StencilEnable) {

22
src/xenia/gpu/d3d12/pipeline_cache.h
View File

@ -152,26 +152,26 @@ class PipelineCache {
uint32_t depth_clip : 1; // 15
uint32_t rov_msaa : 1; // 16
DepthRenderTargetFormat depth_format : 1; // 17
uint32_t depth_func : 3; // 20
CompareFunction depth_func : 3; // 20
uint32_t depth_write : 1; // 21
uint32_t stencil_enable : 1; // 22
uint32_t stencil_read_mask : 8; // 30
uint32_t force_early_z : 1; // 31
uint32_t stencil_write_mask : 8; // 8
uint32_t stencil_front_fail_op : 3; // 11
uint32_t stencil_front_depth_fail_op : 3; // 14
uint32_t stencil_front_pass_op : 3; // 17
uint32_t stencil_front_func : 3; // 20
uint32_t stencil_back_fail_op : 3; // 23
uint32_t stencil_back_depth_fail_op : 3; // 26
uint32_t stencil_back_pass_op : 3; // 29
uint32_t stencil_back_func : 3; // 32
uint32_t stencil_write_mask : 8; // 8
StencilOp stencil_front_fail_op : 3; // 11
StencilOp stencil_front_depth_fail_op : 3; // 14
StencilOp stencil_front_pass_op : 3; // 17
CompareFunction stencil_front_func : 3; // 20
StencilOp stencil_back_fail_op : 3; // 23
StencilOp stencil_back_depth_fail_op : 3; // 26
StencilOp stencil_back_pass_op : 3; // 29
CompareFunction stencil_back_func : 3; // 32
PipelineRenderTarget render_targets[4];
};
bool TranslateShader(D3D12Shader* shader, xenos::xe_gpu_program_cntl_t cntl,
bool TranslateShader(D3D12Shader* shader, reg::SQ_PROGRAM_CNTL cntl,
bool tessellated, PrimitiveType primitive_type);
bool GetCurrentStateDescription(

2
src/xenia/gpu/d3d12/primitive_converter.cc
View File

@ -192,7 +192,7 @@ PrimitiveConverter::ConversionResult PrimitiveConverter::ConvertPrimitives(
D3D12_GPU_VIRTUAL_ADDRESS& gpu_address_out, uint32_t& index_count_out) {
bool index_32bit = index_format == IndexFormat::kInt32;
auto& regs = *register_file_;
bool reset = (regs[XE_GPU_REG_PA_SU_SC_MODE_CNTL].u32 & (1 << 21)) != 0;
bool reset = regs.Get<reg::PA_SU_SC_MODE_CNTL>().multi_prim_ib_ena;
// Swap the reset index because we will be comparing unswapped values to it.
uint32_t reset_index = xenos::GpuSwap(
regs[XE_GPU_REG_VGT_MULTI_PRIM_IB_RESET_INDX].u32, index_endianness);

260
src/xenia/gpu/d3d12/render_target_cache.cc
View File

@ -541,16 +541,17 @@ bool RenderTargetCache::UpdateRenderTargets(const D3D12Shader* pixel_shader) {
bool rov_used = command_processor_->IsROVUsedForEDRAM();
uint32_t rb_surface_info = regs[XE_GPU_REG_RB_SURFACE_INFO].u32;
uint32_t surface_pitch = std::min(rb_surface_info & 0x3FFF, 2560u);
auto rb_surface_info = regs.Get<reg::RB_SURFACE_INFO>();
uint32_t surface_pitch = std::min(rb_surface_info.surface_pitch, 2560u);
if (surface_pitch == 0) {
// TODO(Triang3l): Do something if a memexport-only draw has 0 surface
// pitch (never seen in any game so far, not sure if even legal).
return false;
}
MsaaSamples msaa_samples = MsaaSamples((rb_surface_info >> 16) & 0x3);
uint32_t msaa_samples_x = msaa_samples >= MsaaSamples::k4X ? 2 : 1;
uint32_t msaa_samples_y = msaa_samples >= MsaaSamples::k2X ? 2 : 1;
uint32_t msaa_samples_x =
rb_surface_info.msaa_samples >= MsaaSamples::k4X ? 2 : 1;
uint32_t msaa_samples_y =
rb_surface_info.msaa_samples >= MsaaSamples::k2X ? 2 : 1;
// Extract color/depth info in an unified way.
bool enabled[5];
@ -558,26 +559,27 @@ bool RenderTargetCache::UpdateRenderTargets(const D3D12Shader* pixel_shader) {
uint32_t formats[5];
bool formats_are_64bpp[5];
uint32_t color_mask = command_processor_->GetCurrentColorMask(pixel_shader);
uint32_t rb_color_info[4] = {
regs[XE_GPU_REG_RB_COLOR_INFO].u32, regs[XE_GPU_REG_RB_COLOR1_INFO].u32,
regs[XE_GPU_REG_RB_COLOR2_INFO].u32, regs[XE_GPU_REG_RB_COLOR3_INFO].u32};
for (uint32_t i = 0; i < 4; ++i) {
enabled[i] = (color_mask & (0xF << (i * 4))) != 0;
edram_bases[i] = std::min(rb_color_info[i] & 0xFFF, 2048u);
formats[i] = uint32_t(GetBaseColorFormat(
ColorRenderTargetFormat((rb_color_info[i] >> 16) & 0xF)));
auto color_info = regs.Get<reg::RB_COLOR_INFO>(
reg::RB_COLOR_INFO::rt_register_indices[i]);
edram_bases[i] = std::min(color_info.color_base, 2048u);
formats[i] = uint32_t(GetBaseColorFormat(color_info.color_format));
formats_are_64bpp[i] =
IsColorFormat64bpp(ColorRenderTargetFormat(formats[i]));
}
uint32_t rb_depthcontrol = regs[XE_GPU_REG_RB_DEPTHCONTROL].u32;
uint32_t rb_depth_info = regs[XE_GPU_REG_RB_DEPTH_INFO].u32;
auto rb_depthcontrol = regs.Get<reg::RB_DEPTHCONTROL>();
auto rb_depth_info = regs.Get<reg::RB_DEPTH_INFO>();
// 0x1 = stencil test, 0x2 = depth test.
enabled[4] = (rb_depthcontrol & (0x1 | 0x2)) != 0;
edram_bases[4] = std::min(rb_depth_info & 0xFFF, 2048u);
formats[4] = (rb_depth_info >> 16) & 0x1;
enabled[4] = rb_depthcontrol.stencil_enable || rb_depthcontrol.z_enable;
edram_bases[4] = std::min(rb_depth_info.depth_base, 2048u);
formats[4] = uint32_t(rb_depth_info.depth_format);
formats_are_64bpp[4] = false;
// Don't mark depth regions as dirty if not writing the depth.
bool depth_readonly = (rb_depthcontrol & (0x1 | 0x4)) == 0;
// TODO(Triang3l): Make a common function for checking if stencil writing is
// really done?
bool depth_readonly =
!rb_depthcontrol.stencil_enable && !rb_depthcontrol.z_write_enable;
bool full_update = false;
@ -590,7 +592,7 @@ bool RenderTargetCache::UpdateRenderTargets(const D3D12Shader* pixel_shader) {
// in the beginning of the frame or after resolves by setting the current
// pitch to 0.
if (current_surface_pitch_ != surface_pitch ||
current_msaa_samples_ != msaa_samples) {
current_msaa_samples_ != rb_surface_info.msaa_samples) {
full_update = true;
}
@ -632,26 +634,22 @@ bool RenderTargetCache::UpdateRenderTargets(const D3D12Shader* pixel_shader) {
// Get EDRAM usage of the current draw so dirty regions can be calculated.
// See D3D12CommandProcessor::UpdateFixedFunctionState for more info.
int16_t window_offset_y =
(regs[XE_GPU_REG_PA_SC_WINDOW_OFFSET].u32 >> 16) & 0x7FFF;
if (window_offset_y & 0x4000) {
window_offset_y |= 0x8000;
}
uint32_t pa_cl_vte_cntl = regs[XE_GPU_REG_PA_CL_VTE_CNTL].u32;
float viewport_scale_y = (pa_cl_vte_cntl & (1 << 2))
int32_t window_offset_y =
regs.Get<reg::PA_SC_WINDOW_OFFSET>().window_y_offset;
auto pa_cl_vte_cntl = regs.Get<reg::PA_CL_VTE_CNTL>();
float viewport_scale_y = pa_cl_vte_cntl.vport_y_scale_ena
? regs[XE_GPU_REG_PA_CL_VPORT_YSCALE].f32
: 1280.0f;
float viewport_offset_y = (pa_cl_vte_cntl & (1 << 3))
float viewport_offset_y = pa_cl_vte_cntl.vport_y_offset_ena
? regs[XE_GPU_REG_PA_CL_VPORT_YOFFSET].f32
: std::abs(viewport_scale_y);
if (regs[XE_GPU_REG_PA_SU_SC_MODE_CNTL].u32 & (1 << 16)) {
if (regs.Get<reg::PA_SU_SC_MODE_CNTL>().vtx_window_offset_enable) {
viewport_offset_y += float(window_offset_y);
}
uint32_t viewport_bottom = uint32_t(std::max(
0.0f, std::ceil(viewport_offset_y + std::abs(viewport_scale_y))));
uint32_t scissor_bottom =
(regs[XE_GPU_REG_PA_SC_WINDOW_SCISSOR_BR].u32 >> 16) & 0x7FFF;
if (!(regs[XE_GPU_REG_PA_SC_WINDOW_SCISSOR_TL].u32 & (1u << 31))) {
uint32_t scissor_bottom = regs.Get<reg::PA_SC_WINDOW_SCISSOR_BR>().br_y;
if (!regs.Get<reg::PA_SC_WINDOW_SCISSOR_TL>().window_offset_disable) {
scissor_bottom = std::max(int32_t(scissor_bottom) + window_offset_y, 0);
}
uint32_t dirty_bottom =
@ -769,7 +767,7 @@ bool RenderTargetCache::UpdateRenderTargets(const D3D12Shader* pixel_shader) {
ClearBindings();
current_surface_pitch_ = surface_pitch;
current_msaa_samples_ = msaa_samples;
current_msaa_samples_ = rb_surface_info.msaa_samples;
if (!rov_used) {
current_edram_max_rows_ = edram_max_rows;
}
@ -801,8 +799,8 @@ bool RenderTargetCache::UpdateRenderTargets(const D3D12Shader* pixel_shader) {
#endif
}
XELOGGPU("RT Cache: %s update - pitch %u, samples %u, RTs to attach %u",
full_update ? "Full" : "Partial", surface_pitch, msaa_samples,
render_targets_to_attach);
full_update ? "Full" : "Partial", surface_pitch,
rb_surface_info.msaa_samples, render_targets_to_attach);
#if 0
auto device =
@ -891,7 +889,7 @@ bool RenderTargetCache::UpdateRenderTargets(const D3D12Shader* pixel_shader) {
if (!rov_used) {
// Sample positions when loading depth must match sample positions when
// drawing.
command_processor_->SetSamplePositions(msaa_samples);
command_processor_->SetSamplePositions(rb_surface_info.msaa_samples);
// Load the contents of the new render targets from the EDRAM buffer (will
// change the state of the render targets to copy destination).
@ -1007,18 +1005,14 @@ bool RenderTargetCache::Resolve(SharedMemory* shared_memory,
auto& regs = *register_file_;
// Get the render target properties.
uint32_t rb_surface_info = regs[XE_GPU_REG_RB_SURFACE_INFO].u32;
uint32_t surface_pitch = std::min(rb_surface_info & 0x3FFF, 2560u);
auto rb_surface_info = regs.Get<reg::RB_SURFACE_INFO>();
uint32_t surface_pitch = std::min(rb_surface_info.surface_pitch, 2560u);
if (surface_pitch == 0) {
return true;
}
MsaaSamples msaa_samples = MsaaSamples((rb_surface_info >> 16) & 0x3);
uint32_t rb_copy_control = regs[XE_GPU_REG_RB_COPY_CONTROL].u32;
// Depth info is always needed because color resolve may also clear depth.
uint32_t rb_depth_info = regs[XE_GPU_REG_RB_DEPTH_INFO].u32;
uint32_t depth_edram_base = rb_depth_info & 0xFFF;
uint32_t depth_format = (rb_depth_info >> 16) & 0x1;
uint32_t surface_index = rb_copy_control & 0x7;
auto rb_depth_info = regs.Get<reg::RB_DEPTH_INFO>();
uint32_t surface_index = regs.Get<reg::RB_COPY_CONTROL>().copy_src_select;
if (surface_index > 4) {
assert_always();
return false;
@ -1027,43 +1021,28 @@ bool RenderTargetCache::Resolve(SharedMemory* shared_memory,
uint32_t surface_edram_base;
uint32_t surface_format;
if (surface_is_depth) {
surface_edram_base = depth_edram_base;
surface_format = depth_format;
surface_edram_base = rb_depth_info.depth_base;
surface_format = uint32_t(rb_depth_info.depth_format);
} else {
uint32_t rb_color_info;
switch (surface_index) {
case 1:
rb_color_info = regs[XE_GPU_REG_RB_COLOR1_INFO].u32;
break;
case 2:
rb_color_info = regs[XE_GPU_REG_RB_COLOR2_INFO].u32;
break;
case 3:
rb_color_info = regs[XE_GPU_REG_RB_COLOR3_INFO].u32;
break;
default:
rb_color_info = regs[XE_GPU_REG_RB_COLOR_INFO].u32;
break;
}
surface_edram_base = rb_color_info & 0xFFF;
surface_format = uint32_t(GetBaseColorFormat(
ColorRenderTargetFormat((rb_color_info >> 16) & 0xF)));
auto color_info = regs.Get<reg::RB_COLOR_INFO>(
reg::RB_COLOR_INFO::rt_register_indices[surface_index]);
surface_edram_base = color_info.color_base;
surface_format = uint32_t(GetBaseColorFormat(color_info.color_format));
}
// Get the resolve region since both copying and clearing need it.
// HACK: Vertices to use are always in vf0.
auto fetch_group = reinterpret_cast<const xenos::xe_gpu_fetch_group_t*>(
&regs.values[XE_GPU_REG_SHADER_CONSTANT_FETCH_00_0]);
const auto& fetch = fetch_group->vertex_fetch_0;
const auto& fetch = regs.Get<xenos::xe_gpu_vertex_fetch_t>(
XE_GPU_REG_SHADER_CONSTANT_FETCH_00_0);
assert_true(fetch.type == 3);
assert_true(fetch.endian == 2);
assert_true(fetch.endian == Endian::k8in32);
assert_true(fetch.size == 6);
const uint8_t* src_vertex_address =
memory->TranslatePhysical(fetch.address << 2);
float vertices[6];
// Most vertices have a negative half pixel offset applied, which we reverse.
float vertex_offset =
(regs[XE_GPU_REG_PA_SU_VTX_CNTL].u32 & 0x1) ? 0.0f : 0.5f;
regs.Get<reg::PA_SU_VTX_CNTL>().pix_center ? 0.0f : 0.5f;
for (uint32_t i = 0; i < 6; ++i) {
vertices[i] =
xenos::GpuSwap(xe::load<float>(src_vertex_address + i * sizeof(float)),
@ -1097,39 +1076,34 @@ bool RenderTargetCache::Resolve(SharedMemory* shared_memory,
// vertices (-640,0)->(640,720), however, the destination texture pointer is
// adjusted properly to the right half of the texture, and the source render
// target has a pitch of 800).
auto pa_sc_window_offset = regs.Get<reg::PA_SC_WINDOW_OFFSET>();
D3D12_RECT rect;
rect.left = LONG(std::min(std::min(vertices[0], vertices[2]), vertices[4]));
rect.right = LONG(std::max(std::max(vertices[0], vertices[2]), vertices[4]));
rect.top = LONG(std::min(std::min(vertices[1], vertices[3]), vertices[5]));
rect.bottom = LONG(std::max(std::max(vertices[1], vertices[3]), vertices[5]));
if (regs.Get<reg::PA_SU_SC_MODE_CNTL>().vtx_window_offset_enable) {
rect.left += pa_sc_window_offset.window_x_offset;
rect.right += pa_sc_window_offset.window_x_offset;
rect.top += pa_sc_window_offset.window_y_offset;
rect.bottom += pa_sc_window_offset.window_y_offset;
}
D3D12_RECT scissor;
uint32_t window_scissor_tl = regs[XE_GPU_REG_PA_SC_WINDOW_SCISSOR_TL].u32;
uint32_t window_scissor_br = regs[XE_GPU_REG_PA_SC_WINDOW_SCISSOR_BR].u32;
scissor.left = LONG(window_scissor_tl & 0x7FFF);
scissor.right = LONG(window_scissor_br & 0x7FFF);
scissor.top = LONG((window_scissor_tl >> 16) & 0x7FFF);
scissor.bottom = LONG((window_scissor_br >> 16) & 0x7FFF);
if (regs[XE_GPU_REG_PA_SU_SC_MODE_CNTL].u32 & (1 << 16)) {
uint32_t pa_sc_window_offset = regs[XE_GPU_REG_PA_SC_WINDOW_OFFSET].u32;
int16_t window_offset_x = pa_sc_window_offset & 0x7FFF;
int16_t window_offset_y = (pa_sc_window_offset >> 16) & 0x7FFF;
if (window_offset_x & 0x4000) {
window_offset_x |= 0x8000;
}
if (window_offset_y & 0x4000) {
window_offset_y |= 0x8000;
}
rect.left += window_offset_x;
rect.right += window_offset_x;
rect.top += window_offset_y;
rect.bottom += window_offset_y;
if (!(window_scissor_tl & (1u << 31))) {
scissor.left = std::max(LONG(scissor.left + window_offset_x), LONG(0));
scissor.right = std::max(LONG(scissor.right + window_offset_x), LONG(0));
scissor.top = std::max(LONG(scissor.top + window_offset_y), LONG(0));
scissor.bottom =
std::max(LONG(scissor.bottom + window_offset_y), LONG(0));
}
auto pa_sc_window_scissor_tl = regs.Get<reg::PA_SC_WINDOW_SCISSOR_TL>();
auto pa_sc_window_scissor_br = regs.Get<reg::PA_SC_WINDOW_SCISSOR_BR>();
scissor.left = pa_sc_window_scissor_tl.tl_x;
scissor.right = pa_sc_window_scissor_br.br_x;
scissor.top = pa_sc_window_scissor_tl.tl_y;
scissor.bottom = pa_sc_window_scissor_br.br_y;
if (!pa_sc_window_scissor_tl.window_offset_disable) {
scissor.left = std::max(
LONG(scissor.left + pa_sc_window_offset.window_x_offset), LONG(0));
scissor.right = std::max(
LONG(scissor.right + pa_sc_window_offset.window_x_offset), LONG(0));
scissor.top = std::max(
LONG(scissor.top + pa_sc_window_offset.window_y_offset), LONG(0));
scissor.bottom = std::max(
LONG(scissor.bottom + pa_sc_window_offset.window_y_offset), LONG(0));
}
rect.left = std::max(rect.left, scissor.left);
rect.right = std::min(rect.right, scissor.right);
@ -1140,9 +1114,9 @@ bool RenderTargetCache::Resolve(SharedMemory* shared_memory,
"Resolve: (%d,%d)->(%d,%d) of RT %u (pitch %u, %u sample%s, format %u) "
"at %u",
rect.left, rect.top, rect.right, rect.bottom, surface_index,
surface_pitch, 1 << uint32_t(msaa_samples),
msaa_samples != MsaaSamples::k1X ? "s" : "", surface_format,
surface_edram_base);
surface_pitch, 1 << uint32_t(rb_surface_info.msaa_samples),
rb_surface_info.msaa_samples != MsaaSamples::k1X ? "s" : "",
surface_format, surface_edram_base);
if (rect.left >= rect.right || rect.top >= rect.bottom) {
// Nothing to copy.
@ -1157,18 +1131,20 @@ bool RenderTargetCache::Resolve(SharedMemory* shared_memory,
// GetEDRAMLayout in ResolveCopy and ResolveClear will perform the needed
// clamping to the source render target size.
bool result =
ResolveCopy(shared_memory, texture_cache, surface_edram_base,
surface_pitch, msaa_samples, surface_is_depth, surface_format,
rect, written_address_out, written_length_out);
bool result = ResolveCopy(shared_memory, texture_cache, surface_edram_base,
surface_pitch, rb_surface_info.msaa_samples,
surface_is_depth, surface_format, rect,
written_address_out, written_length_out);
// Clear the color RT if needed.
if (!surface_is_depth) {
result &= ResolveClear(surface_edram_base, surface_pitch, msaa_samples,
false, surface_format, rect);
result &=
ResolveClear(surface_edram_base, surface_pitch,
rb_surface_info.msaa_samples, false, surface_format, rect);
}
// Clear the depth RT if needed (may be cleared alongside color).
result &= ResolveClear(depth_edram_base, surface_pitch, msaa_samples, true,
depth_format, rect);
result &= ResolveClear(rb_depth_info.depth_base, surface_pitch,
rb_surface_info.msaa_samples, true,
uint32_t(rb_depth_info.depth_format), rect);
return result;
}
@ -1183,19 +1159,18 @@ bool RenderTargetCache::ResolveCopy(SharedMemory* shared_memory,
auto& regs = *register_file_;
uint32_t rb_copy_control = regs[XE_GPU_REG_RB_COPY_CONTROL].u32;
xenos::CopyCommand copy_command =
xenos::CopyCommand((rb_copy_control >> 20) & 0x3);
if (copy_command != xenos::CopyCommand::kRaw &&
copy_command != xenos::CopyCommand::kConvert) {
auto rb_copy_control = regs.Get<reg::RB_COPY_CONTROL>();
if (rb_copy_control.copy_command != xenos::CopyCommand::kRaw &&
rb_copy_control.copy_command != xenos::CopyCommand::kConvert) {
// TODO(Triang3l): Handle kConstantOne and kNull.
assert_always();
return false;
}
auto command_list = command_processor_->GetDeferredCommandList();
// Get format info.
uint32_t rb_copy_dest_info = regs[XE_GPU_REG_RB_COPY_DEST_INFO].u32;
auto rb_copy_dest_info = regs.Get<reg::RB_COPY_DEST_INFO>();
TextureFormat src_texture_format;
bool src_64bpp;
if (is_depth) {
@ -1222,14 +1197,15 @@ bool RenderTargetCache::ResolveCopy(SharedMemory* shared_memory,
// The destination format is specified as k_8_8_8_8 when resolving depth, but
// no format conversion is done for depth, so ignore it.
TextureFormat dest_format =
is_depth ? src_texture_format
: GetBaseFormat(TextureFormat((rb_copy_dest_info >> 7) & 0x3F));
is_depth
? src_texture_format
: GetBaseFormat(TextureFormat(rb_copy_dest_info.copy_dest_format));
const FormatInfo* dest_format_info = FormatInfo::Get(dest_format);
// Get the destination region and clamp the source region to it.
uint32_t rb_copy_dest_pitch = regs[XE_GPU_REG_RB_COPY_DEST_PITCH].u32;
uint32_t dest_pitch = rb_copy_dest_pitch & 0x3FFF;
uint32_t dest_height = (rb_copy_dest_pitch >> 16) & 0x3FFF;
auto rb_copy_dest_pitch = regs.Get<reg::RB_COPY_DEST_PITCH>();
uint32_t dest_pitch = rb_copy_dest_pitch.copy_dest_pitch;
uint32_t dest_height = rb_copy_dest_pitch.copy_dest_height;
if (dest_pitch == 0 || dest_height == 0) {
// Nothing to copy.
return true;
@ -1263,8 +1239,7 @@ bool RenderTargetCache::ResolveCopy(SharedMemory* shared_memory,
uint32_t dest_address = regs[XE_GPU_REG_RB_COPY_DEST_BASE].u32 & 0x1FFFFFFF;
// An example of a 3D resolve destination is the color grading LUT (used
// starting from the developer/publisher intro) in Dead Space 3.
bool dest_3d = (rb_copy_dest_info & (1 << 3)) != 0;
if (dest_3d) {
if (rb_copy_dest_info.copy_dest_array) {
dest_address += texture_util::GetTiledOffset3D(
int(rect.left & ~LONG(31)), int(rect.top & ~LONG(31)), 0, dest_pitch,
dest_height, xe::log2_floor(dest_format_info->bits_per_pixel >> 3));
@ -1279,21 +1254,20 @@ bool RenderTargetCache::ResolveCopy(SharedMemory* shared_memory,
// resolve to 8bpp or 16bpp textures at very odd locations.
return false;
}
uint32_t dest_z = dest_3d ? ((rb_copy_dest_info >> 4) & 0x7) : 0;
uint32_t dest_z =
rb_copy_dest_info.copy_dest_array ? rb_copy_dest_info.copy_dest_slice : 0;
// See what samples we need and what we should do with them.
xenos::CopySampleSelect sample_select =
xenos::CopySampleSelect((rb_copy_control >> 4) & 0x7);
xenos::CopySampleSelect sample_select = rb_copy_control.copy_sample_select;
if (is_depth && sample_select > xenos::CopySampleSelect::k3) {
assert_always();
return false;
}
Endian128 dest_endian = Endian128(rb_copy_dest_info & 0x7);
int32_t dest_exp_bias;
if (is_depth) {
dest_exp_bias = 0;
} else {
dest_exp_bias = int32_t((rb_copy_dest_info >> 16) << 26) >> 26;
dest_exp_bias = rb_copy_dest_info.copy_dest_exp_bias;
if (ColorRenderTargetFormat(src_format) ==
ColorRenderTargetFormat::k_16_16 ||
ColorRenderTargetFormat(src_format) ==
@ -1309,14 +1283,14 @@ bool RenderTargetCache::ResolveCopy(SharedMemory* shared_memory,
}
}
}
bool dest_swap = !is_depth && ((rb_copy_dest_info >> 24) & 0x1);
bool dest_swap = !is_depth && rb_copy_dest_info.copy_dest_swap;
XELOGGPU(
"Resolve: Copying samples %u to 0x%.8X (%ux%u, %cD), destination Z %u, "
"destination format %s, exponent bias %d, red and blue %sswapped",
uint32_t(sample_select), dest_address, dest_pitch, dest_height,
dest_3d ? '3' : '2', dest_z, dest_format_info->name, dest_exp_bias,
dest_swap ? "" : "not ");
rb_copy_dest_info.copy_dest_array ? '3' : '2', dest_z,
dest_format_info->name, dest_exp_bias, dest_swap ? "" : "not ");
// There are 2 paths for resolving in this function - they don't necessarily
// have to map directly to kRaw and kConvert CopyCommands.
@ -1344,7 +1318,7 @@ bool RenderTargetCache::ResolveCopy(SharedMemory* shared_memory,
resolution_scale_2x_ &&
cvars::d3d12_resolution_scale_resolve_edge_clamp &&
cvars::d3d12_half_pixel_offset &&
!(regs[XE_GPU_REG_PA_SU_VTX_CNTL].u32 & 0x1);
!regs.Get<reg::PA_SU_VTX_CNTL>().pix_center;
if (sample_select <= xenos::CopySampleSelect::k3 &&
src_texture_format == dest_format && dest_exp_bias == 0) {
// *************************************************************************
@ -1363,7 +1337,7 @@ bool RenderTargetCache::ResolveCopy(SharedMemory* shared_memory,
uint32_t dest_size;
uint32_t dest_modified_start = dest_address;
uint32_t dest_modified_length;
if (dest_3d) {
if (rb_copy_dest_info.copy_dest_array) {
// Depth granularity is 4 (though TiledAddress chaining is possible with 8
// granularity).
dest_size = texture_util::GetGuestMipSliceStorageSize(
@ -1442,8 +1416,10 @@ bool RenderTargetCache::ResolveCopy(SharedMemory* shared_memory,
assert_true(dest_pitch <= 8192);
root_constants.tile_sample_dest_info =
((dest_pitch + 31) >> 5) |
(dest_3d ? (((dest_height + 31) >> 5) << 9) : 0) |
(uint32_t(sample_select) << 18) | (uint32_t(dest_endian) << 20);
(rb_copy_dest_info.copy_dest_array ? (((dest_height + 31) >> 5) << 9)
: 0) |
(uint32_t(sample_select) << 18) |
(uint32_t(rb_copy_dest_info.copy_dest_endian) << 20);
if (dest_swap) {
root_constants.tile_sample_dest_info |= (1 << 23) | (src_format << 24);
}
@ -1797,10 +1773,12 @@ bool RenderTargetCache::ResolveCopy(SharedMemory* shared_memory,
copy_buffer_state = D3D12_RESOURCE_STATE_NON_PIXEL_SHADER_RESOURCE;
// dest_address already adjusted, so offsets are & 31.
texture_cache->TileResolvedTexture(
dest_format, dest_address, dest_pitch, dest_height, dest_3d,
uint32_t(rect.left) & 31, uint32_t(rect.top) & 31, dest_z, copy_width,
copy_height, dest_endian, copy_buffer, resolve_target->copy_buffer_size,
resolve_target->footprint, &written_address_out, &written_length_out);
dest_format, dest_address, dest_pitch, dest_height,
rb_copy_dest_info.copy_dest_array != 0, uint32_t(rect.left) & 31,
uint32_t(rect.top) & 31, dest_z, copy_width, copy_height,
rb_copy_dest_info.copy_dest_endian, copy_buffer,
resolve_target->copy_buffer_size, resolve_target->footprint,
&written_address_out, &written_length_out);
// Done with the copy buffer.
@ -1817,9 +1795,15 @@ bool RenderTargetCache::ResolveClear(uint32_t edram_base,
auto& regs = *register_file_;
// Check if clearing is enabled.
uint32_t rb_copy_control = regs[XE_GPU_REG_RB_COPY_CONTROL].u32;
if (!(rb_copy_control & (is_depth ? (1 << 9) : (1 << 8)))) {
return true;
auto rb_copy_control = regs.Get<reg::RB_COPY_CONTROL>();
if (is_depth) {
if (!rb_copy_control.depth_clear_enable) {
return true;
}
} else {
if (!rb_copy_control.color_clear_enable) {
return true;
}
}
XELOGGPU("Resolve: Clearing the %s render target",
@ -1886,7 +1870,7 @@ bool RenderTargetCache::ResolveClear(uint32_t edram_base,
} else if (is_64bpp) {
// TODO(Triang3l): Check which 32-bit portion is in which register.
root_constants.clear_color_high = regs[XE_GPU_REG_RB_COLOR_CLEAR].u32;
root_constants.clear_color_low = regs[XE_GPU_REG_RB_COLOR_CLEAR_LOW].u32;
root_constants.clear_color_low = regs[XE_GPU_REG_RB_COLOR_CLEAR_LO].u32;
command_processor_->SetComputePipeline(edram_clear_64bpp_pipeline_);
} else {
Register reg =

15
src/xenia/gpu/d3d12/shared_memory.cc
View File

@ -34,7 +34,6 @@ namespace d3d12 {
constexpr uint32_t SharedMemory::kBufferSizeLog2;
constexpr uint32_t SharedMemory::kBufferSize;
constexpr uint32_t SharedMemory::kAddressMask;
constexpr uint32_t SharedMemory::kHeapSizeLog2;
constexpr uint32_t SharedMemory::kHeapSize;
constexpr uint32_t SharedMemory::kWatchBucketSizeLog2;
@ -198,10 +197,9 @@ void SharedMemory::UnregisterGlobalWatch(GlobalWatchHandle handle) {
SharedMemory::WatchHandle SharedMemory::WatchMemoryRange(
uint32_t start, uint32_t length, WatchCallback callback,
void* callback_context, void* callback_data, uint64_t callback_argument) {
if (length == 0) {
if (length == 0 || start >= kBufferSize) {
return nullptr;
}
start &= kAddressMask;
length = std::min(length, kBufferSize - start);
uint32_t watch_page_first = start >> page_size_log2_;
uint32_t watch_page_last = (start + length - 1) >> page_size_log2_;
@ -278,9 +276,7 @@ bool SharedMemory::MakeTilesResident(uint32_t start, uint32_t length) {
// Some texture is empty, for example - safe to draw in this case.
return true;
}
start &= kAddressMask;
if ((kBufferSize - start) < length) {
// Exceeds the physical address space.
if (start > kBufferSize || (kBufferSize - start) < length) {
return false;
}
@ -343,9 +339,7 @@ bool SharedMemory::RequestRange(uint32_t start, uint32_t length) {
// Some texture is empty, for example - safe to draw in this case.
return true;
}
start &= kAddressMask;
if ((kBufferSize - start) < length) {
// Exceeds the physical address space.
if (start > kBufferSize || (kBufferSize - start) < length) {
return false;
}
uint32_t last = start + length - 1;
@ -433,8 +427,7 @@ void SharedMemory::FireWatches(uint32_t page_first, uint32_t page_last,
}
void SharedMemory::RangeWrittenByGPU(uint32_t start, uint32_t length) {
start &= kAddressMask;
if (length == 0) {
if (length == 0 || start >= kBufferSize) {
return;
}
length = std::min(length, kBufferSize - start);

1
src/xenia/gpu/d3d12/shared_memory.h
View File

@ -138,7 +138,6 @@ class SharedMemory {
// The 512 MB tiled buffer.
static constexpr uint32_t kBufferSizeLog2 = 29;
static constexpr uint32_t kBufferSize = 1 << kBufferSizeLog2;
static constexpr uint32_t kAddressMask = kBufferSize - 1;
ID3D12Resource* buffer_ = nullptr;
D3D12_GPU_VIRTUAL_ADDRESS buffer_gpu_address_ = 0;
D3D12_RESOURCE_STATES buffer_state_ = D3D12_RESOURCE_STATE_COPY_DEST;

987
src/xenia/gpu/d3d12/texture_cache.cc
View File

File diff suppressed because it is too large Load Diff

9
src/xenia/gpu/d3d12/texture_cache.h
View File

@ -266,13 +266,8 @@ class TextureCache {
DXGI_FORMAT dxgi_format_resolve_tile;
ResolveTileMode resolve_tile_mode;
// Whether the red component must be replicated in the SRV swizzle, for
// single-component formats. At least for DXT3A/DXT5A, this is according to
// http://fileadmin.cs.lth.se/cs/Personal/Michael_Doggett/talks/unc-xenos-doggett.pdf
// k_8 is also used with RGBA swizzle, but assumes replicated components, in
// Halo 3 sprites, thus it appears that all single-component formats should
// have RRRR swizzle.
bool replicate_component;
// Mapping of Xenos swizzle components to DXGI format components.
uint8_t swizzle[4];
};
union TextureKey {

281
src/xenia/gpu/dxbc_shader_translator.cc
View File

@ -416,9 +416,13 @@ void DxbcShaderTranslator::ConvertPWLGamma(
}
void DxbcShaderTranslator::StartVertexShader_LoadVertexIndex() {
if (register_count() < 1) {
return;
}
// Vertex index is in an input bound to SV_VertexID, byte swapped according to
// xe_vertex_index_endian_and_edge_factors system constant and written to GPR
// 0 (which is always present because register_count includes +1).
// xe_vertex_index_endian_and_edge_factors system constant and written to
// GPR 0.
// xe_vertex_index_endian_and_edge_factors & 0b11 is:
// - 00 for no swap.
@ -756,157 +760,161 @@ void DxbcShaderTranslator::StartVertexOrDomainShader() {
// Write the vertex index to GPR 0.
StartVertexShader_LoadVertexIndex();
} else if (IsDxbcDomainShader()) {
uint32_t temp_register_operand_length =
uses_register_dynamic_addressing() ? 3 : 2;
// Copy the domain location to r0.yz (for quad patches) or r0.xyz (for
// triangle patches), and also set the domain in STAT.
uint32_t domain_location_mask, domain_location_swizzle;
if (patch_primitive_type() == PrimitiveType::kTrianglePatch) {
domain_location_mask = 0b0111;
// ZYX swizzle with r1.y == 0, according to the water shader in
// Banjo-Kazooie: Nuts & Bolts.
domain_location_swizzle = 0b00000110;
stat_.tessellator_domain = D3D11_SB_TESSELLATOR_DOMAIN_TRI;
} else {
// TODO(Triang3l): Support line patches.
assert_true(patch_primitive_type() == PrimitiveType::kQuadPatch);
// According to the ground shader in Viva Pinata, though it's impossible
// (as of December 12th, 2018) to test there since it possibly requires
// memexport for ground control points (the memory region with them is
// filled with zeros).
domain_location_mask = 0b0110;
domain_location_swizzle = 0b00000100;
stat_.tessellator_domain = D3D11_SB_TESSELLATOR_DOMAIN_QUAD;
}
shader_code_.push_back(ENCODE_D3D10_SB_OPCODE_TYPE(D3D10_SB_OPCODE_MOV) |
ENCODE_D3D10_SB_TOKENIZED_INSTRUCTION_LENGTH(
2 + temp_register_operand_length));
if (uses_register_dynamic_addressing()) {
shader_code_.push_back(EncodeVectorMaskedOperand(
D3D10_SB_OPERAND_TYPE_INDEXABLE_TEMP, domain_location_mask, 2));
shader_code_.push_back(0);
} else {
shader_code_.push_back(EncodeVectorMaskedOperand(
D3D10_SB_OPERAND_TYPE_TEMP, domain_location_mask, 1));
}
shader_code_.push_back(0);
shader_code_.push_back(EncodeVectorSwizzledOperand(
D3D11_SB_OPERAND_TYPE_INPUT_DOMAIN_POINT, domain_location_swizzle, 0));
++stat_.instruction_count;
if (uses_register_dynamic_addressing()) {
++stat_.array_instruction_count;
} else {
++stat_.mov_instruction_count;
}
assert_true(register_count() >= 2);
if (register_count() != 0) {
uint32_t temp_register_operand_length =
uses_register_dynamic_addressing() ? 3 : 2;
// Copy the primitive index to r0.x (for quad patches) or r1.x (for
// triangle patches) as a float.
// When using indexable temps, copy through a r# because x# are apparently
// only accessible via mov.
// TODO(Triang3l): Investigate what should be written for primitives (or
// even control points) for non-adaptive tessellation modes (they may
// possibly have an index buffer).
// TODO(Triang3l): Support line patches.
uint32_t primitive_id_gpr_index =
patch_primitive_type() == PrimitiveType::kTrianglePatch ? 1 : 0;
if (register_count() > primitive_id_gpr_index) {
uint32_t primitive_id_temp = uses_register_dynamic_addressing()
? PushSystemTemp()
: primitive_id_gpr_index;
shader_code_.push_back(ENCODE_D3D10_SB_OPCODE_TYPE(D3D10_SB_OPCODE_UTOF) |
ENCODE_D3D10_SB_TOKENIZED_INSTRUCTION_LENGTH(4));
shader_code_.push_back(
EncodeVectorMaskedOperand(D3D10_SB_OPERAND_TYPE_TEMP, 0b0001, 1));
shader_code_.push_back(primitive_id_temp);
shader_code_.push_back(
EncodeScalarOperand(D3D10_SB_OPERAND_TYPE_INPUT_PRIMITIVEID, 0));
++stat_.instruction_count;
++stat_.conversion_instruction_count;
if (uses_register_dynamic_addressing()) {
shader_code_.push_back(
ENCODE_D3D10_SB_OPCODE_TYPE(D3D10_SB_OPCODE_MOV) |
ENCODE_D3D10_SB_TOKENIZED_INSTRUCTION_LENGTH(6));
shader_code_.push_back(EncodeVectorMaskedOperand(
D3D10_SB_OPERAND_TYPE_INDEXABLE_TEMP, 0b0001, 2));
shader_code_.push_back(0);
shader_code_.push_back(primitive_id_gpr_index);
shader_code_.push_back(
EncodeVectorSelectOperand(D3D10_SB_OPERAND_TYPE_TEMP, 0, 1));
shader_code_.push_back(primitive_id_temp);
++stat_.instruction_count;
++stat_.array_instruction_count;
// Release primitive_id_temp.
PopSystemTemp();
// Copy the domain location to r0.yz (for quad patches) or r0.xyz (for
// triangle patches), and also set the domain in STAT.
uint32_t domain_location_mask, domain_location_swizzle;
if (patch_primitive_type() == PrimitiveType::kTrianglePatch) {
domain_location_mask = 0b0111;
// ZYX swizzle with r1.y == 0, according to the water shader in
// Banjo-Kazooie: Nuts & Bolts.
domain_location_swizzle = 0b00000110;
stat_.tessellator_domain = D3D11_SB_TESSELLATOR_DOMAIN_TRI;
} else {
// TODO(Triang3l): Support line patches.
assert_true(patch_primitive_type() == PrimitiveType::kQuadPatch);
// According to the ground shader in Viva Pinata, though it's impossible
// (as of December 12th, 2018) to test there since it possibly requires
// memexport for ground control points (the memory region with them is
// filled with zeros).
domain_location_mask = 0b0110;
domain_location_swizzle = 0b00000100;
stat_.tessellator_domain = D3D11_SB_TESSELLATOR_DOMAIN_QUAD;
}
}
if (register_count() >= 2) {
// Write the swizzle of the barycentric/UV coordinates to r1.x (for quad
// patches) or r1.y (for triangle patches). It appears that the
// tessellator offloads the reordering of coordinates for edges to game
// shaders.
//
// In Banjo-Kazooie: Nuts & Bolts (triangle patches with per-edge
// factors), the shader multiplies the first control point's position by
// r0.z, the second CP's by r0.y, and the third CP's by r0.x. But before
// doing that it swizzles r0.xyz the following way depending on the value
// in r1.y:
// - ZXY for 1.0.
// - YZX for 2.0.
// - XZY for 4.0.
// - YXZ for 5.0.
// - ZYX for 6.0.
// Possibly, the logic here is that the value itself is the amount of
// rotation of the swizzle to the right, and 1 << 2 is set when the
// swizzle needs to be flipped before rotating.
//
// In Viva Pinata (quad patches with per-edge factors - not possible to
// test however as of December 12th, 2018), if we assume that r0.y is V
// and r0.z is U, the factors each control point value is multiplied by
// are the following:
// - (1-v)*(1-u), v*(1-u), (1-v)*u, v*u for 0.0 (base swizzle).
// - v*(1-u), (1-v)*(1-u), v*u, (1-v)*u for 1.0 (YXWZ).
// - v*u, (1-v)*u, v*(1-u), (1-v)*(1-u) for 2.0 (WZYX).
// - (1-v)*u, v*u, (1-v)*(1-u), v*(1-u) for 3.0 (ZWXY).
// According to the control point order at
// https://www.khronos.org/registry/OpenGL/extensions/AMD/AMD_vertex_shader_tessellator.txt
// the first is located at (0,0), the second at (0,1), the third at (1,0)
// and the fourth at (1,1). So, swizzle index 0 appears to be the correct
// one. But, this hasn't been tested yet.
//
// Direct3D 12 appears to be passing the coordinates in a consistent
// order, so we can just use ZYX for triangle patches.
//
// TODO(Triang3l): Support line patches.
uint32_t domain_location_swizzle_mask =
patch_primitive_type() == PrimitiveType::kTrianglePatch ? 0b0010
: 0b0001;
shader_code_.push_back(ENCODE_D3D10_SB_OPCODE_TYPE(D3D10_SB_OPCODE_MOV) |
ENCODE_D3D10_SB_TOKENIZED_INSTRUCTION_LENGTH(
3 + temp_register_operand_length));
2 + temp_register_operand_length));
if (uses_register_dynamic_addressing()) {
shader_code_.push_back(
EncodeVectorMaskedOperand(D3D10_SB_OPERAND_TYPE_INDEXABLE_TEMP,
domain_location_swizzle_mask, 2));
shader_code_.push_back(EncodeVectorMaskedOperand(
D3D10_SB_OPERAND_TYPE_INDEXABLE_TEMP, domain_location_mask, 2));
shader_code_.push_back(0);
} else {
shader_code_.push_back(EncodeVectorMaskedOperand(
D3D10_SB_OPERAND_TYPE_TEMP, domain_location_swizzle_mask, 1));
D3D10_SB_OPERAND_TYPE_TEMP, domain_location_mask, 1));
}
shader_code_.push_back(1);
shader_code_.push_back(
EncodeScalarOperand(D3D10_SB_OPERAND_TYPE_IMMEDIATE32, 0));
shader_code_.push_back(0);
shader_code_.push_back(
EncodeVectorSwizzledOperand(D3D11_SB_OPERAND_TYPE_INPUT_DOMAIN_POINT,
domain_location_swizzle, 0));
++stat_.instruction_count;
if (uses_register_dynamic_addressing()) {
++stat_.array_instruction_count;
} else {
++stat_.mov_instruction_count;
}
// Copy the primitive index to r0.x (for quad patches) or r1.x (for
// triangle patches) as a float.
// When using indexable temps, copy through a r# because x# are apparently
// only accessible via mov.
// TODO(Triang3l): Investigate what should be written for primitives (or
// even control points) for non-adaptive tessellation modes (they may
// possibly have an index buffer).
// TODO(Triang3l): Support line patches.
uint32_t primitive_id_gpr_index =
patch_primitive_type() == PrimitiveType::kTrianglePatch ? 1 : 0;
if (register_count() > primitive_id_gpr_index) {
uint32_t primitive_id_temp = uses_register_dynamic_addressing()
? PushSystemTemp()
: primitive_id_gpr_index;
shader_code_.push_back(
ENCODE_D3D10_SB_OPCODE_TYPE(D3D10_SB_OPCODE_UTOF) |
ENCODE_D3D10_SB_TOKENIZED_INSTRUCTION_LENGTH(4));
shader_code_.push_back(
EncodeVectorMaskedOperand(D3D10_SB_OPERAND_TYPE_TEMP, 0b0001, 1));
shader_code_.push_back(primitive_id_temp);
shader_code_.push_back(
EncodeScalarOperand(D3D10_SB_OPERAND_TYPE_INPUT_PRIMITIVEID, 0));
++stat_.instruction_count;
++stat_.conversion_instruction_count;
if (uses_register_dynamic_addressing()) {
shader_code_.push_back(
ENCODE_D3D10_SB_OPCODE_TYPE(D3D10_SB_OPCODE_MOV) |
ENCODE_D3D10_SB_TOKENIZED_INSTRUCTION_LENGTH(6));
shader_code_.push_back(EncodeVectorMaskedOperand(
D3D10_SB_OPERAND_TYPE_INDEXABLE_TEMP, 0b0001, 2));
shader_code_.push_back(0);
shader_code_.push_back(primitive_id_gpr_index);
shader_code_.push_back(
EncodeVectorSelectOperand(D3D10_SB_OPERAND_TYPE_TEMP, 0, 1));
shader_code_.push_back(primitive_id_temp);
++stat_.instruction_count;
++stat_.array_instruction_count;
// Release primitive_id_temp.
PopSystemTemp();
}
}
if (register_count() >= 2) {
// Write the swizzle of the barycentric/UV coordinates to r1.x (for quad
// patches) or r1.y (for triangle patches). It appears that the
// tessellator offloads the reordering of coordinates for edges to game
// shaders.
//
// In Banjo-Kazooie: Nuts & Bolts (triangle patches with per-edge
// factors), the shader multiplies the first control point's position by
// r0.z, the second CP's by r0.y, and the third CP's by r0.x. But before
// doing that it swizzles r0.xyz the following way depending on the
// value in r1.y:
// - ZXY for 1.0.
// - YZX for 2.0.
// - XZY for 4.0.
// - YXZ for 5.0.
// - ZYX for 6.0.
// Possibly, the logic here is that the value itself is the amount of
// rotation of the swizzle to the right, and 1 << 2 is set when the
// swizzle needs to be flipped before rotating.
//
// In Viva Pinata (quad patches with per-edge factors - not possible to
// test however as of December 12th, 2018), if we assume that r0.y is V
// and r0.z is U, the factors each control point value is multiplied by
// are the following:
// - (1-v)*(1-u), v*(1-u), (1-v)*u, v*u for 0.0 (base swizzle).
// - v*(1-u), (1-v)*(1-u), v*u, (1-v)*u for 1.0 (YXWZ).
// - v*u, (1-v)*u, v*(1-u), (1-v)*(1-u) for 2.0 (WZYX).
// - (1-v)*u, v*u, (1-v)*(1-u), v*(1-u) for 3.0 (ZWXY).
// According to the control point order at
// https://www.khronos.org/registry/OpenGL/extensions/AMD/AMD_vertex_shader_tessellator.txt
// the first is located at (0,0), the second at (0,1), the third at
// (1,0) and the fourth at (1,1). So, swizzle index 0 appears to be the
// correct one. But, this hasn't been tested yet.
//
// Direct3D 12 appears to be passing the coordinates in a consistent
// order, so we can just use ZYX for triangle patches.
//
// TODO(Triang3l): Support line patches.
uint32_t domain_location_swizzle_mask =
patch_primitive_type() == PrimitiveType::kTrianglePatch ? 0b0010
: 0b0001;
shader_code_.push_back(
ENCODE_D3D10_SB_OPCODE_TYPE(D3D10_SB_OPCODE_MOV) |
ENCODE_D3D10_SB_TOKENIZED_INSTRUCTION_LENGTH(
3 + temp_register_operand_length));
if (uses_register_dynamic_addressing()) {
shader_code_.push_back(
EncodeVectorMaskedOperand(D3D10_SB_OPERAND_TYPE_INDEXABLE_TEMP,
domain_location_swizzle_mask, 2));
shader_code_.push_back(0);
} else {
shader_code_.push_back(EncodeVectorMaskedOperand(
D3D10_SB_OPERAND_TYPE_TEMP, domain_location_swizzle_mask, 1));
}
shader_code_.push_back(1);
shader_code_.push_back(
EncodeScalarOperand(D3D10_SB_OPERAND_TYPE_IMMEDIATE32, 0));
shader_code_.push_back(0);
++stat_.instruction_count;
if (uses_register_dynamic_addressing()) {
++stat_.array_instruction_count;
} else {
++stat_.mov_instruction_count;
}
}
}
}
}
@ -4796,6 +4804,7 @@ void DxbcShaderTranslator::WriteShaderCode() {
// General-purpose registers if using dynamic indexing (x0).
if (!is_depth_only_pixel_shader_ && uses_register_dynamic_addressing()) {
assert_true(register_count() != 0);
shader_object_.push_back(
ENCODE_D3D10_SB_OPCODE_TYPE(D3D10_SB_OPCODE_DCL_INDEXABLE_TEMP) |
ENCODE_D3D10_SB_TOKENIZED_INSTRUCTION_LENGTH(4));

3
src/xenia/gpu/dxbc_shader_translator.h
View File

@ -503,6 +503,9 @@ class DxbcShaderTranslator : public ShaderTranslator {
kVSOutPosition,
kVSOutClipDistance0123,
kVSOutClipDistance45,
// TODO(Triang3l): Use SV_CullDistance instead for
// PA_CL_CLIP_CNTL::UCP_CULL_ONLY_ENA, but can't have more than 8 clip and
// cull distances in total.
kPSInInterpolators = 0,
kPSInPointParameters = kPSInInterpolators + kInterpolatorCount,

22
src/xenia/gpu/register_file.h
View File

@ -13,15 +13,11 @@
#include <cstdint>
#include <cstdlib>
#include "xenia/gpu/registers.h"
namespace xe {
namespace gpu {
enum Register {
#define XE_GPU_REGISTER(index, type, name) XE_GPU_REG_##name = index,
#include "xenia/gpu/register_table.inc"
#undef XE_GPU_REGISTER
};
struct RegisterInfo {
enum class Type {
kDword,
@ -44,8 +40,20 @@ class RegisterFile {
};
RegisterValue values[kRegisterCount];
RegisterValue& operator[](int reg) { return values[reg]; }
RegisterValue& operator[](uint32_t reg) { return values[reg]; }
RegisterValue& operator[](Register reg) { return values[reg]; }
template <typename T>
T& Get(uint32_t reg) {
return *reinterpret_cast<T*>(&values[reg]);
}
template <typename T>
T& Get(Register reg) {
return *reinterpret_cast<T*>(&values[reg]);
}
template <typename T>
T& Get() {
return *reinterpret_cast<T*>(&values[T::register_index]);
}
};
} // namespace gpu

10
src/xenia/gpu/register_table.inc
View File

@ -134,7 +134,7 @@ XE_GPU_REGISTER(0x2184, kDword, SQ_WRAPPING_1)
XE_GPU_REGISTER(0x21F9, kDword, VGT_EVENT_INITIATOR)
XE_GPU_REGISTER(0x2200, kDword, RB_DEPTHCONTROL)
XE_GPU_REGISTER(0x2201, kDword, RB_BLENDCONTROL_0)
XE_GPU_REGISTER(0x2201, kDword, RB_BLENDCONTROL0)
XE_GPU_REGISTER(0x2202, kDword, RB_COLORCONTROL)
XE_GPU_REGISTER(0x2203, kDword, RB_HIZCONTROL)
XE_GPU_REGISTER(0x2204, kDword, PA_CL_CLIP_CNTL)
@ -142,9 +142,9 @@ XE_GPU_REGISTER(0x2205, kDword, PA_SU_SC_MODE_CNTL)
XE_GPU_REGISTER(0x2206, kDword, PA_CL_VTE_CNTL)
XE_GPU_REGISTER(0x2207, kDword, VGT_CURRENT_BIN_ID_MIN)
XE_GPU_REGISTER(0x2208, kDword, RB_MODECONTROL)
XE_GPU_REGISTER(0x2209, kDword, RB_BLENDCONTROL_1)
XE_GPU_REGISTER(0x220A, kDword, RB_BLENDCONTROL_2)
XE_GPU_REGISTER(0x220B, kDword, RB_BLENDCONTROL_3)
XE_GPU_REGISTER(0x2209, kDword, RB_BLENDCONTROL1)
XE_GPU_REGISTER(0x220A, kDword, RB_BLENDCONTROL2)
XE_GPU_REGISTER(0x220B, kDword, RB_BLENDCONTROL3)
XE_GPU_REGISTER(0x2280, kDword, PA_SU_POINT_SIZE)
XE_GPU_REGISTER(0x2281, kDword, PA_SU_POINT_MINMAX)
@ -199,7 +199,7 @@ XE_GPU_REGISTER(0x231B, kDword, RB_COPY_DEST_INFO)
XE_GPU_REGISTER(0x231C, kDword, RB_HIZ_CLEAR)
XE_GPU_REGISTER(0x231D, kDword, RB_DEPTH_CLEAR)
XE_GPU_REGISTER(0x231E, kDword, RB_COLOR_CLEAR)
XE_GPU_REGISTER(0x231F, kDword, RB_COLOR_CLEAR_LOW)
XE_GPU_REGISTER(0x231F, kDword, RB_COLOR_CLEAR_LO)
XE_GPU_REGISTER(0x2320, kDword, RB_COPY_FUNC)
XE_GPU_REGISTER(0x2321, kDword, RB_COPY_REF)
XE_GPU_REGISTER(0x2322, kDword, RB_COPY_MASK)

64
src/xenia/gpu/registers.cc Normal file
View File

@ -0,0 +1,64 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2019 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#include "xenia/gpu/registers.h"
namespace xe {
namespace gpu {
namespace reg {
constexpr Register COHER_STATUS_HOST::register_index;
constexpr Register WAIT_UNTIL::register_index;
constexpr Register SQ_PROGRAM_CNTL::register_index;
constexpr Register SQ_CONTEXT_MISC::register_index;
constexpr Register VGT_OUTPUT_PATH_CNTL::register_index;
constexpr Register VGT_HOS_CNTL::register_index;
constexpr Register PA_SU_POINT_MINMAX::register_index;
constexpr Register PA_SU_POINT_SIZE::register_index;
constexpr Register PA_SU_SC_MODE_CNTL::register_index;
constexpr Register PA_SU_VTX_CNTL::register_index;
constexpr Register PA_SC_MPASS_PS_CNTL::register_index;
constexpr Register PA_SC_VIZ_QUERY::register_index;
constexpr Register PA_CL_CLIP_CNTL::register_index;
constexpr Register PA_CL_VTE_CNTL::register_index;
constexpr Register PA_SC_WINDOW_OFFSET::register_index;
constexpr Register PA_SC_WINDOW_SCISSOR_TL::register_index;
constexpr Register PA_SC_WINDOW_SCISSOR_BR::register_index;
constexpr Register RB_MODECONTROL::register_index;
constexpr Register RB_SURFACE_INFO::register_index;
constexpr Register RB_COLORCONTROL::register_index;
constexpr Register RB_COLOR_INFO::register_index;
const Register RB_COLOR_INFO::rt_register_indices[4] = {
XE_GPU_REG_RB_COLOR_INFO,
XE_GPU_REG_RB_COLOR1_INFO,
XE_GPU_REG_RB_COLOR2_INFO,
XE_GPU_REG_RB_COLOR3_INFO,
};
constexpr Register RB_COLOR_MASK::register_index;
constexpr Register RB_BLENDCONTROL::register_index;
const Register RB_BLENDCONTROL::rt_register_indices[4] = {
XE_GPU_REG_RB_BLENDCONTROL0,
XE_GPU_REG_RB_BLENDCONTROL1,
XE_GPU_REG_RB_BLENDCONTROL2,
XE_GPU_REG_RB_BLENDCONTROL3,
};
constexpr Register RB_DEPTHCONTROL::register_index;
constexpr Register RB_STENCILREFMASK::register_index;
constexpr Register RB_DEPTH_INFO::register_index;
constexpr Register RB_COPY_CONTROL::register_index;
constexpr Register RB_COPY_DEST_INFO::register_index;
constexpr Register RB_COPY_DEST_PITCH::register_index;
} // namespace reg
} // namespace gpu
} // namespace xe

544
src/xenia/gpu/registers.h
View File

@ -13,65 +13,155 @@
#include <cstdint>
#include <cstdlib>
#include "xenia/base/bit_field.h"
#include "xenia/gpu/xenos.h"
// Most registers can be found from:
// https://github.com/UDOOboard/Kernel_Unico/blob/master/drivers/mxc/amd-gpu/include/reg/yamato/14/yamato_registers.h
// Some registers were added on Adreno specifically and are not referenced in
// game .pdb files and never set by games.
namespace xe {
namespace gpu {
enum Register {
#define XE_GPU_REGISTER(index, type, name) XE_GPU_REG_##name = index,
#include "xenia/gpu/register_table.inc"
#undef XE_GPU_REGISTER
};
namespace reg {
/**************************************************
/*******************************************************************************
___ ___ _ _ _____ ___ ___ _
/ __/ _ \| \| |_ _| _ \/ _ \| |
| (_| (_) | .` | | | | / (_) | |__
\___\___/|_|\_| |_| |_|_\\___/|____|
***************************************************/
*******************************************************************************/
union COHER_STATUS_HOST {
xe::bf<uint32_t, 0, 8> matching_contexts;
xe::bf<uint32_t, 8, 1> rb_copy_dest_base_ena;
xe::bf<uint32_t, 9, 1> dest_base_0_ena;
xe::bf<uint32_t, 10, 1> dest_base_1_ena;
xe::bf<uint32_t, 11, 1> dest_base_2_ena;
xe::bf<uint32_t, 12, 1> dest_base_3_ena;
xe::bf<uint32_t, 13, 1> dest_base_4_ena;
xe::bf<uint32_t, 14, 1> dest_base_5_ena;
xe::bf<uint32_t, 15, 1> dest_base_6_ena;
xe::bf<uint32_t, 16, 1> dest_base_7_ena;
xe::bf<uint32_t, 24, 1> vc_action_ena;
xe::bf<uint32_t, 25, 1> tc_action_ena;
xe::bf<uint32_t, 26, 1> pglb_action_ena;
xe::bf<uint32_t, 31, 1> status;
struct {
uint32_t matching_contexts : 8; // +0
uint32_t rb_copy_dest_base_ena : 1; // +8
uint32_t dest_base_0_ena : 1; // +9
uint32_t dest_base_1_ena : 1; // +10
uint32_t dest_base_2_ena : 1; // +11
uint32_t dest_base_3_ena : 1; // +12
uint32_t dest_base_4_ena : 1; // +13
uint32_t dest_base_5_ena : 1; // +14
uint32_t dest_base_6_ena : 1; // +15
uint32_t dest_base_7_ena : 1; // +16
uint32_t : 7; // +17
uint32_t vc_action_ena : 1; // +24
uint32_t tc_action_ena : 1; // +25
uint32_t pglb_action_ena : 1; // +26
uint32_t : 4; // +27
uint32_t status : 1; // +31
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_COHER_STATUS_HOST;
};
union WAIT_UNTIL {
xe::bf<uint32_t, 1, 1> wait_re_vsync;
xe::bf<uint32_t, 2, 1> wait_fe_vsync;
xe::bf<uint32_t, 3, 1> wait_vsync;
xe::bf<uint32_t, 4, 1> wait_dsply_id0;
xe::bf<uint32_t, 5, 1> wait_dsply_id1;
xe::bf<uint32_t, 6, 1> wait_dsply_id2;
xe::bf<uint32_t, 10, 1> wait_cmdfifo;
xe::bf<uint32_t, 14, 1> wait_2d_idle;
xe::bf<uint32_t, 15, 1> wait_3d_idle;
xe::bf<uint32_t, 16, 1> wait_2d_idleclean;
xe::bf<uint32_t, 17, 1> wait_3d_idleclean;
xe::bf<uint32_t, 20, 4> cmdfifo_entries;
struct {
uint32_t : 1; // +0
uint32_t wait_re_vsync : 1; // +1
uint32_t wait_fe_vsync : 1; // +2
uint32_t wait_vsync : 1; // +3
uint32_t wait_dsply_id0 : 1; // +4
uint32_t wait_dsply_id1 : 1; // +5
uint32_t wait_dsply_id2 : 1; // +6
uint32_t : 3; // +7
uint32_t wait_cmdfifo : 1; // +10
uint32_t : 3; // +11
uint32_t wait_2d_idle : 1; // +14
uint32_t wait_3d_idle : 1; // +15
uint32_t wait_2d_idleclean : 1; // +16
uint32_t wait_3d_idleclean : 1; // +17
uint32_t : 2; // +18
uint32_t cmdfifo_entries : 4; // +20
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_WAIT_UNTIL;
};
/**************************************************
/*******************************************************************************
___ ___ ___ _ _ ___ _ _ ___ ___ ___
/ __| __/ _ \| | | | __| \| |/ __| __| _ \
\__ \ _| (_) | |_| | _|| .` | (__| _|| /
|___/___\__\_\\___/|___|_|\_|\___|___|_|_\
*******************************************************************************/
union SQ_PROGRAM_CNTL {
struct {
// Note from a2xx.xml:
// Only 0x3F worth of valid register values for VS_NUM_REG and PS_NUM_REG,
// but high bit is set to indicate "0 registers used".
uint32_t vs_num_reg : 8; // +0
uint32_t ps_num_reg : 8; // +8
uint32_t vs_resource : 1; // +16
uint32_t ps_resource : 1; // +17
uint32_t param_gen : 1; // +18
uint32_t gen_index_pix : 1; // +19
uint32_t vs_export_count : 4; // +20
xenos::VertexShaderExportMode vs_export_mode : 3; // +24
uint32_t ps_export_mode : 4; // +27
uint32_t gen_index_vtx : 1; // +31
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_SQ_PROGRAM_CNTL;
};
union SQ_CONTEXT_MISC {
struct {
uint32_t inst_pred_optimize : 1; // +0
uint32_t sc_output_screen_xy : 1; // +1
xenos::SampleControl sc_sample_cntl : 2; // +2
uint32_t : 4; // +4
uint32_t param_gen_pos : 8; // +8
uint32_t perfcounter_ref : 1; // +16
uint32_t yeild_optimize : 1; // +17 sic
uint32_t tx_cache_sel : 1; // +18
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_SQ_CONTEXT_MISC;
};
/*******************************************************************************
__ _____ ___ _____ _____ __
\ \ / / __| _ \_ _| __\ \/ /
\ V /| _|| / | | | _| > <
\_/ |___|_|_\ |_| |___/_/\_\
___ ___ ___ _ _ ___ ___ ___ _ _ _ ___
/ __| _ \/ _ \| | | | _ \ __| _ \ /_\ | \| | \
| (_ | / (_) | |_| | _/ _|| / / _ \| .` | |) |
\___|_|_\\___/ \___/|_| |___|_|_\ /_/ \_\_|\_|___/
_____ ___ ___ ___ ___ _ _ _ _____ ___ ___
|_ _| __/ __/ __| __| | | | /_\_ _/ _ \| _ \
| | | _|\__ \__ \ _|| |__| |__ / _ \| || (_) | /
|_| |___|___/___/___|____|____/_/ \_\_| \___/|_|_\
*******************************************************************************/
union VGT_OUTPUT_PATH_CNTL {
struct {
xenos::VGTOutputPath path_select : 2; // +0
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_VGT_OUTPUT_PATH_CNTL;
};
union VGT_HOS_CNTL {
struct {
xenos::TessellationMode tess_mode : 2; // +0
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_VGT_HOS_CNTL;
};
/*******************************************************************************
___ ___ ___ __ __ ___ _____ _____ _____
| _ \ _ \_ _| \/ |_ _|_ _|_ _\ \ / / __|
| _/ /| || |\/| || | | | | | \ V /| _|
@ -82,206 +172,348 @@ union WAIT_UNTIL {
/ _ \\__ \__ \ _|| |\/| | _ \ |__| _|| /
/_/ \_\___/___/___|_| |_|___/____|___|_|_\
***************************************************/
*******************************************************************************/
union PA_SU_POINT_MINMAX {
struct {
// Radius, 12.4 fixed point.
uint32_t min_size : 16; // +0
uint32_t max_size : 16; // +16
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_PA_SU_POINT_MINMAX;
};
union PA_SU_POINT_SIZE {
struct {
// 1/2 width or height, 12.4 fixed point.
uint32_t height : 16; // +0
uint32_t width : 16; // +16
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_PA_SU_POINT_SIZE;
};
// Setup Unit / Scanline Converter mode cntl
union PA_SU_SC_MODE_CNTL {
xe::bf<uint32_t, 0, 1> cull_front;
xe::bf<uint32_t, 1, 1> cull_back;
xe::bf<uint32_t, 2, 1> face;
xe::bf<uint32_t, 3, 2> poly_mode;
xe::bf<uint32_t, 5, 3> polymode_front_ptype;
xe::bf<uint32_t, 8, 3> polymode_back_ptype;
xe::bf<uint32_t, 11, 1> poly_offset_front_enable;
xe::bf<uint32_t, 12, 1> poly_offset_back_enable;
xe::bf<uint32_t, 13, 1> poly_offset_para_enable;
xe::bf<uint32_t, 15, 1> msaa_enable;
xe::bf<uint32_t, 16, 1> vtx_window_offset_enable;
xe::bf<uint32_t, 18, 1> line_stipple_enable;
xe::bf<uint32_t, 19, 1> provoking_vtx_last;
xe::bf<uint32_t, 20, 1> persp_corr_dis;
xe::bf<uint32_t, 21, 1> multi_prim_ib_ena;
xe::bf<uint32_t, 23, 1> quad_order_enable;
xe::bf<uint32_t, 25, 1> wait_rb_idle_all_tri;
xe::bf<uint32_t, 26, 1> wait_rb_idle_first_tri_new_state;
struct {
uint32_t cull_front : 1; // +0
uint32_t cull_back : 1; // +1
// 0 - front is CCW, 1 - front is CW.
uint32_t face : 1; // +2
xenos::PolygonModeEnable poly_mode : 2; // +3
xenos::PolygonType polymode_front_ptype : 3; // +5
xenos::PolygonType polymode_back_ptype : 3; // +8
uint32_t poly_offset_front_enable : 1; // +11
uint32_t poly_offset_back_enable : 1; // +12
uint32_t poly_offset_para_enable : 1; // +13
uint32_t : 1; // +14
uint32_t msaa_enable : 1; // +15
uint32_t vtx_window_offset_enable : 1; // +16
// LINE_STIPPLE_ENABLE was added on Adreno.
uint32_t : 2; // +17
uint32_t provoking_vtx_last : 1; // +19
uint32_t persp_corr_dis : 1; // +20
uint32_t multi_prim_ib_ena : 1; // +21
uint32_t : 1; // +22
uint32_t quad_order_enable : 1; // +23
// WAIT_RB_IDLE_ALL_TRI and WAIT_RB_IDLE_FIRST_TRI_NEW_STATE were added on
// Adreno.
// TODO(Triang3l): Find SC_ONE_QUAD_PER_CLOCK offset.
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_PA_SU_SC_MODE_CNTL;
};
// Setup Unit Vertex Control
union PA_SU_VTX_CNTL {
xe::bf<uint32_t, 0, 1> pix_center; // 1 = half pixel offset
xe::bf<uint32_t, 1, 2> round_mode;
xe::bf<uint32_t, 3, 3> quant_mode;
struct {
uint32_t pix_center : 1; // +0 1 = half pixel offset (OpenGL).
uint32_t round_mode : 2; // +1
uint32_t quant_mode : 3; // +3
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_PA_SU_VTX_CNTL;
};
union PA_SC_MPASS_PS_CNTL {
xe::bf<uint32_t, 0, 20> mpass_pix_vec_per_pass;
xe::bf<uint32_t, 31, 1> mpass_ps_ena;
struct {
uint32_t mpass_pix_vec_per_pass : 20; // +0
uint32_t : 11; // +20
uint32_t mpass_ps_ena : 1; // +31
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_PA_SC_MPASS_PS_CNTL;
};
// Scanline converter viz query
union PA_SC_VIZ_QUERY {
xe::bf<uint32_t, 0, 1> viz_query_ena;
xe::bf<uint32_t, 1, 6> viz_query_id;
xe::bf<uint32_t, 7, 1> kill_pix_post_early_z;
struct {
uint32_t viz_query_ena : 1; // +0
uint32_t viz_query_id : 6; // +1
uint32_t kill_pix_post_early_z : 1; // +7
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_PA_SC_VIZ_QUERY;
};
// Clipper clip control
// TODO(DrChat): This seem to differ. Need to examine this.
// https://github.com/decaf-emu/decaf-emu/blob/c017a9ff8128852fb9a5da19466778a171cea6e1/src/libdecaf/src/gpu/latte_registers_pa.h#L11
union PA_CL_CLIP_CNTL {
xe::bf<uint32_t, 0, 1> ucp_ena_0;
xe::bf<uint32_t, 1, 1> ucp_ena_1;
xe::bf<uint32_t, 2, 1> ucp_ena_2;
xe::bf<uint32_t, 3, 1> ucp_ena_3;
xe::bf<uint32_t, 4, 1> ucp_ena_4;
xe::bf<uint32_t, 5, 1> ucp_ena_5;
xe::bf<uint32_t, 14, 2> ps_ucp_mode;
xe::bf<uint32_t, 16, 1> clip_disable;
xe::bf<uint32_t, 17, 1> ucp_cull_only_ena;
xe::bf<uint32_t, 18, 1> boundary_edge_flag_ena;
xe::bf<uint32_t, 19, 1> dx_clip_space_def;
xe::bf<uint32_t, 20, 1> dis_clip_err_detect;
xe::bf<uint32_t, 21, 1> vtx_kill_or;
xe::bf<uint32_t, 22, 1> xy_nan_retain;
xe::bf<uint32_t, 23, 1> z_nan_retain;
xe::bf<uint32_t, 24, 1> w_nan_retain;
struct {
uint32_t ucp_ena_0 : 1; // +0
uint32_t ucp_ena_1 : 1; // +1
uint32_t ucp_ena_2 : 1; // +2
uint32_t ucp_ena_3 : 1; // +3
uint32_t ucp_ena_4 : 1; // +4
uint32_t ucp_ena_5 : 1; // +5
uint32_t : 8; // +6
uint32_t ps_ucp_mode : 2; // +14
uint32_t clip_disable : 1; // +16
uint32_t ucp_cull_only_ena : 1; // +17
uint32_t boundary_edge_flag_ena : 1; // +18
uint32_t dx_clip_space_def : 1; // +19
uint32_t dis_clip_err_detect : 1; // +20
uint32_t vtx_kill_or : 1; // +21
uint32_t xy_nan_retain : 1; // +22
uint32_t z_nan_retain : 1; // +23
uint32_t w_nan_retain : 1; // +24
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_PA_CL_CLIP_CNTL;
};
// Viewport transform engine control
union PA_CL_VTE_CNTL {
xe::bf<uint32_t, 0, 1> vport_x_scale_ena;
xe::bf<uint32_t, 1, 1> vport_x_offset_ena;
xe::bf<uint32_t, 2, 1> vport_y_scale_ena;
xe::bf<uint32_t, 3, 1> vport_y_offset_ena;
xe::bf<uint32_t, 4, 1> vport_z_scale_ena;
xe::bf<uint32_t, 5, 1> vport_z_offset_ena;
xe::bf<uint32_t, 8, 1> vtx_xy_fmt;
xe::bf<uint32_t, 9, 1> vtx_z_fmt;
xe::bf<uint32_t, 10, 1> vtx_w0_fmt;
xe::bf<uint32_t, 11, 1> perfcounter_ref;
struct {
uint32_t vport_x_scale_ena : 1; // +0
uint32_t vport_x_offset_ena : 1; // +1
uint32_t vport_y_scale_ena : 1; // +2
uint32_t vport_y_offset_ena : 1; // +3
uint32_t vport_z_scale_ena : 1; // +4
uint32_t vport_z_offset_ena : 1; // +5
uint32_t : 2; // +6
uint32_t vtx_xy_fmt : 1; // +8
uint32_t vtx_z_fmt : 1; // +9
uint32_t vtx_w0_fmt : 1; // +10
uint32_t perfcounter_ref : 1; // +11
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_PA_CL_VTE_CNTL;
};
union PA_SC_WINDOW_OFFSET {
xe::bf<int32_t, 0, 15> window_x_offset;
xe::bf<int32_t, 16, 15> window_y_offset;
struct {
int32_t window_x_offset : 15; // +0
uint32_t : 1; // +15
int32_t window_y_offset : 15; // +16
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_PA_SC_WINDOW_OFFSET;
};
union PA_SC_WINDOW_SCISSOR_TL {
xe::bf<uint32_t, 0, 14> tl_x;
xe::bf<uint32_t, 16, 14> tl_y;
xe::bf<uint32_t, 31, 1> window_offset_disable;
struct {
uint32_t tl_x : 14; // +0
uint32_t : 2; // +14
uint32_t tl_y : 14; // +16
uint32_t : 1; // +30
uint32_t window_offset_disable : 1; // +31
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_PA_SC_WINDOW_SCISSOR_TL;
};
union PA_SC_WINDOW_SCISSOR_BR {
xe::bf<uint32_t, 0, 14> br_x;
xe::bf<uint32_t, 16, 14> br_y;
struct {
uint32_t br_x : 14; // +0
uint32_t : 2; // +14
uint32_t br_y : 14; // +16
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_PA_SC_WINDOW_SCISSOR_BR;
};
/**************************************************
/*******************************************************************************
___ ___
| _ \ _ )
| / _ \
|_|_\___/
***************************************************/
*******************************************************************************/
union RB_MODECONTROL {
xe::bf<xenos::ModeControl, 0, 3> edram_mode;
struct {
xenos::ModeControl edram_mode : 3; // +0
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_RB_MODECONTROL;
};
union RB_SURFACE_INFO {
xe::bf<uint32_t, 0, 14> surface_pitch;
xe::bf<MsaaSamples, 16, 2> msaa_samples;
xe::bf<uint32_t, 18, 14> hiz_pitch;
struct {
uint32_t surface_pitch : 14; // +0
uint32_t : 2; // +14
MsaaSamples msaa_samples : 2; // +16
uint32_t hiz_pitch : 14; // +18
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_RB_SURFACE_INFO;
};
union RB_COLORCONTROL {
xe::bf<uint32_t, 0, 3> alpha_func;
xe::bf<uint32_t, 3, 1> alpha_test_enable;
xe::bf<uint32_t, 4, 1> alpha_to_mask_enable;
xe::bf<uint32_t, 24, 2> alpha_to_mask_offset0;
xe::bf<uint32_t, 26, 2> alpha_to_mask_offset1;
xe::bf<uint32_t, 28, 2> alpha_to_mask_offset2;
xe::bf<uint32_t, 30, 2> alpha_to_mask_offset3;
struct {
CompareFunction alpha_func : 3; // +0
uint32_t alpha_test_enable : 1; // +3
uint32_t alpha_to_mask_enable : 1; // +4
// Everything in between was added on Adreno.
uint32_t : 19; // +5
uint32_t alpha_to_mask_offset0 : 2; // +24
uint32_t alpha_to_mask_offset1 : 2; // +26
uint32_t alpha_to_mask_offset2 : 2; // +28
uint32_t alpha_to_mask_offset3 : 2; // +30
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_RB_COLORCONTROL;
};
union RB_COLOR_INFO {
xe::bf<uint32_t, 0, 12> color_base;
xe::bf<ColorRenderTargetFormat, 16, 4> color_format;
xe::bf<uint32_t, 20, 6> color_exp_bias;
struct {
uint32_t color_base : 12; // +0
uint32_t : 4; // +12
ColorRenderTargetFormat color_format : 4; // +16
int32_t color_exp_bias : 6; // +20
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_RB_COLOR_INFO;
// RB_COLOR[1-3]_INFO also use this format.
static const Register rt_register_indices[4];
};
union RB_COLOR_MASK {
struct {
uint32_t write_red0 : 1; // +0
uint32_t write_green0 : 1; // +1
uint32_t write_blue0 : 1; // +2
uint32_t write_alpha0 : 1; // +3
uint32_t write_red1 : 1; // +4
uint32_t write_green1 : 1; // +5
uint32_t write_blue1 : 1; // +6
uint32_t write_alpha1 : 1; // +7
uint32_t write_red2 : 1; // +8
uint32_t write_green2 : 1; // +9
uint32_t write_blue2 : 1; // +10
uint32_t write_alpha2 : 1; // +11
uint32_t write_red3 : 1; // +12
uint32_t write_green3 : 1; // +13
uint32_t write_blue3 : 1; // +14
uint32_t write_alpha3 : 1; // +15
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_RB_COLOR_MASK;
};
union RB_BLENDCONTROL {
struct {
BlendFactor color_srcblend : 5; // +0
BlendOp color_comb_fcn : 3; // +5
BlendFactor color_destblend : 5; // +8
uint32_t : 3; // +13
BlendFactor alpha_srcblend : 5; // +16
BlendOp alpha_comb_fcn : 3; // +21
BlendFactor alpha_destblend : 5; // +24
// BLEND_FORCE_ENABLE and BLEND_FORCE were added on Adreno.
};
uint32_t value;
// RB_BLENDCONTROL[0-3] use this format.
static constexpr Register register_index = XE_GPU_REG_RB_BLENDCONTROL0;
static const Register rt_register_indices[4];
};
union RB_DEPTHCONTROL {
struct {
uint32_t stencil_enable : 1; // +0
uint32_t z_enable : 1; // +1
uint32_t z_write_enable : 1; // +2
// EARLY_Z_ENABLE was added on Adreno.
uint32_t : 1; // +3
CompareFunction zfunc : 3; // +4
uint32_t backface_enable : 1; // +7
CompareFunction stencilfunc : 3; // +8
StencilOp stencilfail : 3; // +11
StencilOp stencilzpass : 3; // +14
StencilOp stencilzfail : 3; // +17
CompareFunction stencilfunc_bf : 3; // +20
StencilOp stencilfail_bf : 3; // +23
StencilOp stencilzpass_bf : 3; // +26
StencilOp stencilzfail_bf : 3; // +29
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_RB_DEPTHCONTROL;
};
union RB_STENCILREFMASK {
struct {
uint32_t stencilref : 8; // +0
uint32_t stencilmask : 8; // +8
uint32_t stencilwritemask : 8; // +16
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_RB_STENCILREFMASK;
// RB_STENCILREFMASK_BF also uses this format.
};
union RB_DEPTH_INFO {
xe::bf<uint32_t, 0, 12> depth_base;
xe::bf<DepthRenderTargetFormat, 16, 1> depth_format;
struct {
uint32_t depth_base : 12; // +0
uint32_t : 4; // +12
DepthRenderTargetFormat depth_format : 1; // +16
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_RB_DEPTH_INFO;
};
union RB_COPY_CONTROL {
xe::bf<uint32_t, 0, 3> copy_src_select;
xe::bf<xenos::CopySampleSelect, 4, 3> copy_sample_select;
xe::bf<uint32_t, 8, 1> color_clear_enable;
xe::bf<uint32_t, 9, 1> depth_clear_enable;
xe::bf<xenos::CopyCommand, 20, 2> copy_command;
struct {
uint32_t copy_src_select : 3; // +0 Depth is 4.
uint32_t : 1; // +3
xenos::CopySampleSelect copy_sample_select : 3; // +4
uint32_t : 1; // +7
uint32_t color_clear_enable : 1; // +8
uint32_t depth_clear_enable : 1; // +9
uint32_t : 10; // +10
xenos::CopyCommand copy_command : 2; // +20
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_RB_COPY_CONTROL;
};
union RB_COPY_DEST_INFO {
xe::bf<Endian128, 0, 3> copy_dest_endian;
xe::bf<uint32_t, 3, 1> copy_dest_array;
xe::bf<uint32_t, 4, 3> copy_dest_slice;
xe::bf<ColorFormat, 7, 6> copy_dest_format;
xe::bf<uint32_t, 13, 3> copy_dest_number;
xe::bf<uint32_t, 16, 6> copy_dest_exp_bias;
xe::bf<uint32_t, 24, 1> copy_dest_swap;
struct {
Endian128 copy_dest_endian : 3; // +0
uint32_t copy_dest_array : 1; // +3
uint32_t copy_dest_slice : 3; // +4
ColorFormat copy_dest_format : 6; // +7
uint32_t copy_dest_number : 3; // +13
int32_t copy_dest_exp_bias : 6; // +16
uint32_t : 2; // +22
uint32_t copy_dest_swap : 1; // +24
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_RB_COPY_DEST_INFO;
};
union RB_COPY_DEST_PITCH {
xe::bf<uint32_t, 0, 14> copy_dest_pitch;
xe::bf<uint32_t, 16, 14> copy_dest_height;
struct {
uint32_t copy_dest_pitch : 14; // +0
uint32_t : 2; // +14
uint32_t copy_dest_height : 14; // +16
};
uint32_t value;
static constexpr Register register_index = XE_GPU_REG_RB_COPY_DEST_PITCH;
};
} // namespace reg
} // namespace gpu
} // namespace xe

16
src/xenia/gpu/sampler_info.cc
View File

@ -24,25 +24,25 @@ bool SamplerInfo::Prepare(const xenos::xe_gpu_texture_fetch_t& fetch,
out_info->min_filter =
fetch_instr.attributes.min_filter == TextureFilter::kUseFetchConst
? static_cast<TextureFilter>(fetch.min_filter)
? fetch.min_filter
: fetch_instr.attributes.min_filter;
out_info->mag_filter =
fetch_instr.attributes.mag_filter == TextureFilter::kUseFetchConst
? static_cast<TextureFilter>(fetch.mag_filter)
? fetch.mag_filter
: fetch_instr.attributes.mag_filter;
out_info->mip_filter =
fetch_instr.attributes.mip_filter == TextureFilter::kUseFetchConst
? static_cast<TextureFilter>(fetch.mip_filter)
? fetch.mip_filter
: fetch_instr.attributes.mip_filter;
out_info->clamp_u = static_cast<ClampMode>(fetch.clamp_x);
out_info->clamp_v = static_cast<ClampMode>(fetch.clamp_y);
out_info->clamp_w = static_cast<ClampMode>(fetch.clamp_z);
out_info->clamp_u = fetch.clamp_x;
out_info->clamp_v = fetch.clamp_y;
out_info->clamp_w = fetch.clamp_z;
out_info->aniso_filter =
fetch_instr.attributes.aniso_filter == AnisoFilter::kUseFetchConst
? static_cast<AnisoFilter>(fetch.aniso_filter)
? fetch.aniso_filter
: fetch_instr.attributes.aniso_filter;
out_info->border_color = static_cast<BorderColor>(fetch.border_color);
out_info->border_color = fetch.border_color;
out_info->lod_bias = (fetch.lod_bias) / 32.f;
out_info->mip_min_level = fetch.mip_min_level;
out_info->mip_max_level = fetch.mip_max_level;

7
src/xenia/gpu/shader_translator.cc
View File

@ -108,10 +108,11 @@ bool ShaderTranslator::GatherAllBindingInformation(Shader* shader) {
}
bool ShaderTranslator::Translate(Shader* shader, PrimitiveType patch_type,
xenos::xe_gpu_program_cntl_t cntl) {
reg::SQ_PROGRAM_CNTL cntl) {
Reset();
register_count_ = shader->type() == ShaderType::kVertex ? cntl.vs_regs + 1
: cntl.ps_regs + 1;
uint32_t cntl_num_reg =
shader->type() == ShaderType::kVertex ? cntl.vs_num_reg : cntl.ps_num_reg;
register_count_ = (cntl_num_reg & 0x80) ? 0 : (cntl_num_reg + 1);
return TranslateInternal(shader, patch_type);
}

5
src/xenia/gpu/shader_translator.h
View File

@ -17,6 +17,7 @@
#include "xenia/base/math.h"
#include "xenia/base/string_buffer.h"
#include "xenia/gpu/registers.h"
#include "xenia/gpu/shader.h"
#include "xenia/gpu/ucode.h"
#include "xenia/gpu/xenos.h"
@ -33,7 +34,7 @@ class ShaderTranslator {
bool GatherAllBindingInformation(Shader* shader);
bool Translate(Shader* shader, PrimitiveType patch_type,
xenos::xe_gpu_program_cntl_t cntl);
reg::SQ_PROGRAM_CNTL cntl);
bool Translate(Shader* shader, PrimitiveType patch_type);
protected:
@ -232,7 +233,7 @@ class ShaderTranslator {
PrimitiveType patch_primitive_type_;
const uint32_t* ucode_dwords_;
size_t ucode_dword_count_;
xenos::xe_gpu_program_cntl_t program_cntl_;
reg::SQ_PROGRAM_CNTL program_cntl_;
uint32_t register_count_;
// Accumulated translation errors.

1
src/xenia/gpu/spirv_shader_translator.cc
View File

@ -93,6 +93,7 @@ void SpirvShaderTranslator::StartTranslation() {
b.makeFunctionEntry(spv::NoPrecision, b.makeVoidType(), "translated_main",
{}, {}, &function_block);
assert_not_zero(register_count());
registers_type_ = b.makeArrayType(vec4_float_type_,
b.makeUintConstant(register_count()), 0);
registers_ptr_ = b.createVariable(spv::StorageClass::StorageClassFunction,

2
src/xenia/gpu/texture_conversion.cc
View File

@ -40,7 +40,7 @@ void CopySwapBlock(Endian endian, void* output, const void* input,
xe::copy_and_swap_16_in_32_unaligned(output, input, length);
break;
default:
case Endian::kUnspecified:
case Endian::kNone:
std::memcpy(output, input, length);
break;
}

4
src/xenia/gpu/texture_info.cc
View File

@ -33,8 +33,8 @@ bool TextureInfo::Prepare(const xe_gpu_texture_fetch_t& fetch,
auto& info = *out_info;
info.format = static_cast<TextureFormat>(fetch.format);
info.endianness = static_cast<Endian>(fetch.endianness);
info.format = fetch.format;
info.endianness = fetch.endianness;
info.dimension = static_cast<Dimension>(fetch.dimension);
info.width = info.height = info.depth = 0;

71
src/xenia/gpu/texture_info.h
View File

@ -19,77 +19,6 @@
namespace xe {
namespace gpu {
// a2xx_sq_surfaceformat +
// https://github.com/indirivacua/RAGE-Console-Texture-Editor/blob/master/Console.Xbox360.Graphics.pas
enum class TextureFormat : uint32_t {
k_1_REVERSE = 0,
k_1 = 1,
k_8 = 2,
k_1_5_5_5 = 3,
k_5_6_5 = 4,
k_6_5_5 = 5,
k_8_8_8_8 = 6,
k_2_10_10_10 = 7,
k_8_A = 8,
k_8_B = 9,
k_8_8 = 10,
k_Cr_Y1_Cb_Y0_REP = 11,
k_Y1_Cr_Y0_Cb_REP = 12,
k_16_16_EDRAM = 13,
k_8_8_8_8_A = 14,
k_4_4_4_4 = 15,
k_10_11_11 = 16,
k_11_11_10 = 17,
k_DXT1 = 18,
k_DXT2_3 = 19,
k_DXT4_5 = 20,
k_16_16_16_16_EDRAM = 21,
k_24_8 = 22,
k_24_8_FLOAT = 23,
k_16 = 24,
k_16_16 = 25,
k_16_16_16_16 = 26,
k_16_EXPAND = 27,
k_16_16_EXPAND = 28,
k_16_16_16_16_EXPAND = 29,
k_16_FLOAT = 30,
k_16_16_FLOAT = 31,
k_16_16_16_16_FLOAT = 32,
k_32 = 33,
k_32_32 = 34,
k_32_32_32_32 = 35,
k_32_FLOAT = 36,
k_32_32_FLOAT = 37,
k_32_32_32_32_FLOAT = 38,
k_32_AS_8 = 39,
k_32_AS_8_8 = 40,
k_16_MPEG = 41,
k_16_16_MPEG = 42,
k_8_INTERLACED = 43,
k_32_AS_8_INTERLACED = 44,
k_32_AS_8_8_INTERLACED = 45,
k_16_INTERLACED = 46,
k_16_MPEG_INTERLACED = 47,
k_16_16_MPEG_INTERLACED = 48,
k_DXN = 49,
k_8_8_8_8_AS_16_16_16_16 = 50,
k_DXT1_AS_16_16_16_16 = 51,
k_DXT2_3_AS_16_16_16_16 = 52,
k_DXT4_5_AS_16_16_16_16 = 53,
k_2_10_10_10_AS_16_16_16_16 = 54,
k_10_11_11_AS_16_16_16_16 = 55,
k_11_11_10_AS_16_16_16_16 = 56,
k_32_32_32_FLOAT = 57,
k_DXT3A = 58,
k_DXT5A = 59,
k_CTX1 = 60,
k_DXT3A_AS_1_1_1_1 = 61,
k_8_8_8_8_GAMMA_EDRAM = 62,
k_2_10_10_10_FLOAT_EDRAM = 63,
kUnknown = 0xFFFFFFFFu,
};
inline TextureFormat GetBaseFormat(TextureFormat texture_format) {
// These formats are used for resampling textures / gamma control.
switch (texture_format) {

12
src/xenia/gpu/trace_viewer.cc
View File

@ -824,7 +824,7 @@ void TraceViewer::DrawVertexFetcher(Shader* shader,
#define LOADEL(type, wo) \
GpuSwap(xe::load<type>(vstart + \
(attrib.fetch_instr.attributes.offset + wo) * 4), \
Endian(fetch->endian))
fetch->endian)
switch (attrib.fetch_instr.attributes.data_format) {
case VertexFormat::k_32:
ImGui::Text("%.8X", LOADEL(uint32_t, 0));
@ -1334,10 +1334,10 @@ void TraceViewer::DrawStateUI() {
regs[XE_GPU_REG_RB_COLOR3_INFO].u32,
};
uint32_t rb_blendcontrol[4] = {
regs[XE_GPU_REG_RB_BLENDCONTROL_0].u32,
regs[XE_GPU_REG_RB_BLENDCONTROL_1].u32,
regs[XE_GPU_REG_RB_BLENDCONTROL_2].u32,
regs[XE_GPU_REG_RB_BLENDCONTROL_3].u32,
regs[XE_GPU_REG_RB_BLENDCONTROL0].u32,
regs[XE_GPU_REG_RB_BLENDCONTROL1].u32,
regs[XE_GPU_REG_RB_BLENDCONTROL2].u32,
regs[XE_GPU_REG_RB_BLENDCONTROL3].u32,
};
ImGui::Columns(2);
for (int i = 0; i < xe::countof(color_info); ++i) {
@ -1713,7 +1713,7 @@ void TraceViewer::DrawStateUI() {
fetch = &group->vertex_fetch_2;
break;
}
assert_true(fetch->endian == 2);
assert_true(fetch->endian == Endian::k8in32);
char tree_root_id[32];
sprintf(tree_root_id, "#vertices_root_%d",
vertex_binding.fetch_constant);

236
src/xenia/gpu/ucode.h
View File

@ -146,12 +146,8 @@ enum class AllocType : uint32_t {
// Instruction data for ControlFlowOpcode::kExec and kExecEnd.
struct ControlFlowExecInstruction {
ControlFlowOpcode opcode() const {
return static_cast<ControlFlowOpcode>(opcode_);
}
AddressingMode addressing_mode() const {
return static_cast<AddressingMode>(address_mode_);
}
ControlFlowOpcode opcode() const { return opcode_; }
AddressingMode addressing_mode() const { return address_mode_; }
// Address of the instructions to execute.
uint32_t address() const { return address_; }
// Number of instructions being executed.
@ -176,19 +172,15 @@ struct ControlFlowExecInstruction {
uint32_t : 7;
uint32_t clean_ : 1;
uint32_t : 1;
uint32_t address_mode_ : 1;
uint32_t opcode_ : 4;
AddressingMode address_mode_ : 1;
ControlFlowOpcode opcode_ : 4;
};
static_assert_size(ControlFlowExecInstruction, 8);
// Instruction data for ControlFlowOpcode::kCondExec and kCondExecEnd.
struct ControlFlowCondExecInstruction {
ControlFlowOpcode opcode() const {
return static_cast<ControlFlowOpcode>(opcode_);
}
AddressingMode addressing_mode() const {
return static_cast<AddressingMode>(address_mode_);
}
ControlFlowOpcode opcode() const { return opcode_; }
AddressingMode addressing_mode() const { return address_mode_; }
// Address of the instructions to execute.
uint32_t address() const { return address_; }
// Number of instructions being executed.
@ -214,20 +206,16 @@ struct ControlFlowCondExecInstruction {
uint32_t vc_lo_ : 2;
uint32_t bool_address_ : 8;
uint32_t condition_ : 1;
uint32_t address_mode_ : 1;
uint32_t opcode_ : 4;
AddressingMode address_mode_ : 1;
ControlFlowOpcode opcode_ : 4;
};
static_assert_size(ControlFlowCondExecInstruction, 8);
// Instruction data for ControlFlowOpcode::kCondExecPred, kCondExecPredEnd,
// kCondExecPredClean, kCondExecPredCleanEnd.
struct ControlFlowCondExecPredInstruction {
ControlFlowOpcode opcode() const {
return static_cast<ControlFlowOpcode>(opcode_);
}
AddressingMode addressing_mode() const {
return static_cast<AddressingMode>(address_mode_);
}
ControlFlowOpcode opcode() const { return opcode_; }
AddressingMode addressing_mode() const { return address_mode_; }
// Address of the instructions to execute.
uint32_t address() const { return address_; }
// Number of instructions being executed.
@ -254,19 +242,15 @@ struct ControlFlowCondExecPredInstruction {
uint32_t : 7;
uint32_t clean_ : 1;
uint32_t condition_ : 1;
uint32_t address_mode_ : 1;
uint32_t opcode_ : 4;
AddressingMode address_mode_ : 1;
ControlFlowOpcode opcode_ : 4;
};
static_assert_size(ControlFlowCondExecPredInstruction, 8);
// Instruction data for ControlFlowOpcode::kLoopStart.
struct ControlFlowLoopStartInstruction {
ControlFlowOpcode opcode() const {
return static_cast<ControlFlowOpcode>(opcode_);
}
AddressingMode addressing_mode() const {
return static_cast<AddressingMode>(address_mode_);
}
ControlFlowOpcode opcode() const { return opcode_; }
AddressingMode addressing_mode() const { return address_mode_; }
// Target address to jump to when skipping the loop.
uint32_t address() const { return address_; }
// Whether to reuse the current aL instead of reset it to loop start.
@ -285,19 +269,15 @@ struct ControlFlowLoopStartInstruction {
// Word 1: (16 bits)
uint32_t : 11;
uint32_t address_mode_ : 1;
uint32_t opcode_ : 4;
AddressingMode address_mode_ : 1;
ControlFlowOpcode opcode_ : 4;
};
static_assert_size(ControlFlowLoopStartInstruction, 8);
// Instruction data for ControlFlowOpcode::kLoopEnd.
struct ControlFlowLoopEndInstruction {
ControlFlowOpcode opcode() const {
return static_cast<ControlFlowOpcode>(opcode_);
}
AddressingMode addressing_mode() const {
return static_cast<AddressingMode>(address_mode_);
}
ControlFlowOpcode opcode() const { return opcode_; }
AddressingMode addressing_mode() const { return address_mode_; }
// Target address of the start of the loop body.
uint32_t address() const { return address_; }
// Integer constant register that holds the loop parameters.
@ -319,19 +299,15 @@ struct ControlFlowLoopEndInstruction {
// Word 1: (16 bits)
uint32_t : 10;
uint32_t condition_ : 1;
uint32_t address_mode_ : 1;
uint32_t opcode_ : 4;
AddressingMode address_mode_ : 1;
ControlFlowOpcode opcode_ : 4;
};
static_assert_size(ControlFlowLoopEndInstruction, 8);
// Instruction data for ControlFlowOpcode::kCondCall.
struct ControlFlowCondCallInstruction {
ControlFlowOpcode opcode() const {
return static_cast<ControlFlowOpcode>(opcode_);
}
AddressingMode addressing_mode() const {
return static_cast<AddressingMode>(address_mode_);
}
ControlFlowOpcode opcode() const { return opcode_; }
AddressingMode addressing_mode() const { return address_mode_; }
// Target address.
uint32_t address() const { return address_; }
// Unconditional call - ignores condition/predication.
@ -354,19 +330,15 @@ struct ControlFlowCondCallInstruction {
uint32_t : 2;
uint32_t bool_address_ : 8;
uint32_t condition_ : 1;
uint32_t address_mode_ : 1;
uint32_t opcode_ : 4;
AddressingMode address_mode_ : 1;
ControlFlowOpcode opcode_ : 4;
};
static_assert_size(ControlFlowCondCallInstruction, 8);
// Instruction data for ControlFlowOpcode::kReturn.
struct ControlFlowReturnInstruction {
ControlFlowOpcode opcode() const {
return static_cast<ControlFlowOpcode>(opcode_);
}
AddressingMode addressing_mode() const {
return static_cast<AddressingMode>(address_mode_);
}
ControlFlowOpcode opcode() const { return opcode_; }
AddressingMode addressing_mode() const { return address_mode_; }
private:
// Word 0: (32 bits)
@ -381,12 +353,8 @@ static_assert_size(ControlFlowReturnInstruction, 8);
// Instruction data for ControlFlowOpcode::kCondJmp.
struct ControlFlowCondJmpInstruction {
ControlFlowOpcode opcode() const {
return static_cast<ControlFlowOpcode>(opcode_);
}
AddressingMode addressing_mode() const {
return static_cast<AddressingMode>(address_mode_);
}
ControlFlowOpcode opcode() const { return opcode_; }
AddressingMode addressing_mode() const { return address_mode_; }
// Target address.
uint32_t address() const { return address_; }
// Unconditional jump - ignores condition/predication.
@ -410,20 +378,18 @@ struct ControlFlowCondJmpInstruction {
uint32_t direction_ : 1;
uint32_t bool_address_ : 8;
uint32_t condition_ : 1;
uint32_t address_mode_ : 1;
uint32_t opcode_ : 4;
AddressingMode address_mode_ : 1;
ControlFlowOpcode opcode_ : 4;
};
static_assert_size(ControlFlowCondJmpInstruction, 8);
// Instruction data for ControlFlowOpcode::kAlloc.
struct ControlFlowAllocInstruction {
ControlFlowOpcode opcode() const {
return static_cast<ControlFlowOpcode>(opcode_);
}
ControlFlowOpcode opcode() const { return opcode_; }
// The total number of the given type allocated by this instruction.
uint32_t size() const { return size_; }
// Unconditional jump - ignores condition/predication.
AllocType alloc_type() const { return static_cast<AllocType>(alloc_type_); }
AllocType alloc_type() const { return alloc_type_; }
private:
// Word 0: (32 bits)
@ -433,16 +399,14 @@ struct ControlFlowAllocInstruction {
// Word 1: (16 bits)
uint32_t : 8;
uint32_t is_unserialized_ : 1;
uint32_t alloc_type_ : 2;
AllocType alloc_type_ : 2;
uint32_t : 1;
uint32_t opcode_ : 4;
ControlFlowOpcode opcode_ : 4;
};
static_assert_size(ControlFlowAllocInstruction, 8);
XEPACKEDUNION(ControlFlowInstruction, {
ControlFlowOpcode opcode() const {
return static_cast<ControlFlowOpcode>(opcode_value);
}
ControlFlowOpcode opcode() const { return opcode_value; }
ControlFlowExecInstruction exec; // kExec*
ControlFlowCondExecInstruction cond_exec; // kCondExec*
@ -457,7 +421,7 @@ XEPACKEDUNION(ControlFlowInstruction, {
XEPACKEDSTRUCTANONYMOUS({
uint32_t unused_0 : 32;
uint32_t unused_1 : 12;
uint32_t opcode_value : 4;
ControlFlowOpcode opcode_value : 4;
});
XEPACKEDSTRUCTANONYMOUS({
uint32_t dword_0;
@ -478,7 +442,7 @@ inline void UnpackControlFlowInstructions(const uint32_t* dwords,
out_b->dword_1 = dword_2 >> 16;
}
enum class FetchOpcode {
enum class FetchOpcode : uint32_t {
kVertexFetch = 0,
kTextureFetch = 1,
kGetTextureBorderColorFrac = 16,
@ -492,9 +456,7 @@ enum class FetchOpcode {
};
struct VertexFetchInstruction {
FetchOpcode opcode() const {
return static_cast<FetchOpcode>(data_.opcode_value);
}
FetchOpcode opcode() const { return data_.opcode_value; }
// Whether the jump is predicated (or conditional).
bool is_predicated() const { return data_.is_predicated; }
@ -538,13 +500,9 @@ struct VertexFetchInstruction {
uint32_t prefetch_count() const { return data_.prefetch_count; }
bool is_mini_fetch() const { return data_.is_mini_fetch == 1; }
VertexFormat data_format() const {
return static_cast<VertexFormat>(data_.format);
}
VertexFormat data_format() const { return data_.format; }
// [-32, 31]
int exp_adjust() const {
return ((static_cast<int>(data_.exp_adjust) << 26) >> 26);
}
int exp_adjust() const { return data_.exp_adjust; }
bool is_signed() const { return data_.fomat_comp_all == 1; }
bool is_normalized() const { return data_.num_format_all == 0; }
bool is_index_rounded() const { return data_.is_index_rounded == 1; }
@ -562,7 +520,7 @@ struct VertexFetchInstruction {
private:
XEPACKEDSTRUCT(Data, {
XEPACKEDSTRUCTANONYMOUS({
uint32_t opcode_value : 5;
FetchOpcode opcode_value : 5;
uint32_t src_reg : 6;
uint32_t src_reg_am : 1;
uint32_t dst_reg : 6;
@ -579,9 +537,9 @@ struct VertexFetchInstruction {
uint32_t num_format_all : 1;
uint32_t signed_rf_mode_all : 1;
uint32_t is_index_rounded : 1;
uint32_t format : 6;
VertexFormat format : 6;
uint32_t reserved2 : 2;
uint32_t exp_adjust : 6;
int32_t exp_adjust : 6;
uint32_t is_mini_fetch : 1;
uint32_t is_predicated : 1;
});
@ -595,9 +553,7 @@ struct VertexFetchInstruction {
};
struct TextureFetchInstruction {
FetchOpcode opcode() const {
return static_cast<FetchOpcode>(data_.opcode_value);
}
FetchOpcode opcode() const { return data_.opcode_value; }
// Whether the jump is predicated (or conditional).
bool is_predicated() const { return data_.is_predicated; }
@ -613,59 +569,49 @@ struct TextureFetchInstruction {
uint32_t src_swizzle() const { return data_.src_swiz; }
bool is_src_relative() const { return data_.src_reg_am; }
TextureDimension dimension() const {
return static_cast<TextureDimension>(data_.dimension);
}
TextureDimension dimension() const { return data_.dimension; }
bool fetch_valid_only() const { return data_.fetch_valid_only == 1; }
bool unnormalized_coordinates() const { return data_.tx_coord_denorm == 1; }
bool has_mag_filter() const { return data_.mag_filter != 0x3; }
TextureFilter mag_filter() const {
return static_cast<TextureFilter>(data_.mag_filter);
bool has_mag_filter() const {
return data_.mag_filter != TextureFilter::kUseFetchConst;
}
bool has_min_filter() const { return data_.min_filter != 0x3; }
TextureFilter min_filter() const {
return static_cast<TextureFilter>(data_.min_filter);
TextureFilter mag_filter() const { return data_.mag_filter; }
bool has_min_filter() const {
return data_.min_filter != TextureFilter::kUseFetchConst;
}
bool has_mip_filter() const { return data_.mip_filter != 0x3; }
TextureFilter mip_filter() const {
return static_cast<TextureFilter>(data_.mip_filter);
TextureFilter min_filter() const { return data_.min_filter; }
bool has_mip_filter() const {
return data_.mip_filter != TextureFilter::kUseFetchConst;
}
bool has_aniso_filter() const { return data_.aniso_filter != 0x7; }
AnisoFilter aniso_filter() const {
return static_cast<AnisoFilter>(data_.aniso_filter);
TextureFilter mip_filter() const { return data_.mip_filter; }
bool has_aniso_filter() const {
return data_.aniso_filter != AnisoFilter::kUseFetchConst;
}
bool has_vol_mag_filter() const { return data_.vol_mag_filter != 0x3; }
TextureFilter vol_mag_filter() const {
return static_cast<TextureFilter>(data_.vol_mag_filter);
AnisoFilter aniso_filter() const { return data_.aniso_filter; }
bool has_vol_mag_filter() const {
return data_.vol_mag_filter != TextureFilter::kUseFetchConst;
}
bool has_vol_min_filter() const { return data_.vol_min_filter != 0x3; }
TextureFilter vol_min_filter() const {
return static_cast<TextureFilter>(data_.vol_min_filter);
TextureFilter vol_mag_filter() const { return data_.vol_mag_filter; }
bool has_vol_min_filter() const {
return data_.vol_min_filter != TextureFilter::kUseFetchConst;
}
TextureFilter vol_min_filter() const { return data_.vol_min_filter; }
bool use_computed_lod() const { return data_.use_comp_lod == 1; }
bool use_register_lod() const { return data_.use_reg_lod == 1; }
bool use_register_gradients() const { return data_.use_reg_gradients == 1; }
SampleLocation sample_location() const {
return static_cast<SampleLocation>(data_.sample_location);
}
SampleLocation sample_location() const { return data_.sample_location; }
float lod_bias() const {
// http://web.archive.org/web/20090514012026/http://msdn.microsoft.com:80/en-us/library/bb313957.aspx
return ((static_cast<int>(data_.lod_bias) << 25) >> 25) / 16.0f;
}
float offset_x() const {
return ((static_cast<int>(data_.offset_x) << 27) >> 27) / 2.0f;
}
float offset_y() const {
return ((static_cast<int>(data_.offset_y) << 27) >> 27) / 2.0f;
}
float offset_z() const {
return ((static_cast<int>(data_.offset_z) << 27) >> 27) / 2.0f;
return data_.lod_bias * (1.0f / 16.0f);
}
float offset_x() const { return data_.offset_x * 0.5f; }
float offset_y() const { return data_.offset_y * 0.5f; }
float offset_z() const { return data_.offset_z * 0.5f; }
private:
XEPACKEDSTRUCT(Data, {
XEPACKEDSTRUCTANONYMOUS({
uint32_t opcode_value : 5;
FetchOpcode opcode_value : 5;
uint32_t src_reg : 6;
uint32_t src_reg_am : 1;
uint32_t dst_reg : 6;
@ -676,14 +622,14 @@ struct TextureFetchInstruction {
uint32_t src_swiz : 6; // xyz
});
XEPACKEDSTRUCTANONYMOUS({
uint32_t dst_swiz : 12; // xyzw
uint32_t mag_filter : 2; // instr_tex_filter_t
uint32_t min_filter : 2; // instr_tex_filter_t
uint32_t mip_filter : 2; // instr_tex_filter_t
uint32_t aniso_filter : 3; // instr_aniso_filter_t
uint32_t arbitrary_filter : 3; // instr_arbitrary_filter_t
uint32_t vol_mag_filter : 2; // instr_tex_filter_t
uint32_t vol_min_filter : 2; // instr_tex_filter_t
uint32_t dst_swiz : 12; // xyzw
TextureFilter mag_filter : 2;
TextureFilter min_filter : 2;
TextureFilter mip_filter : 2;
AnisoFilter aniso_filter : 3;
xenos::ArbitraryFilter arbitrary_filter : 3;
TextureFilter vol_mag_filter : 2;
TextureFilter vol_min_filter : 2;
uint32_t use_comp_lod : 1;
uint32_t use_reg_lod : 1;
uint32_t unk : 1;
@ -691,13 +637,13 @@ struct TextureFetchInstruction {
});
XEPACKEDSTRUCTANONYMOUS({
uint32_t use_reg_gradients : 1;
uint32_t sample_location : 1;
uint32_t lod_bias : 7;
SampleLocation sample_location : 1;
int32_t lod_bias : 7;
uint32_t unused : 5;
uint32_t dimension : 2;
uint32_t offset_x : 5;
uint32_t offset_y : 5;
uint32_t offset_z : 5;
TextureDimension dimension : 2;
int32_t offset_x : 5;
int32_t offset_y : 5;
int32_t offset_z : 5;
uint32_t pred_condition : 1;
});
});
@ -722,7 +668,7 @@ static_assert_size(TextureFetchInstruction, 12);
// when write masks are disabled or the instruction that would write them
// fails its predication check.
enum class AluScalarOpcode {
enum class AluScalarOpcode : uint32_t {
// Floating-Point Add
// adds dest, src0.ab
// dest.xyzw = src0.a + src0.b;
@ -1049,7 +995,7 @@ enum class AluScalarOpcode {
kRetainPrev = 50,
};
enum class AluVectorOpcode {
enum class AluVectorOpcode : uint32_t {
// Per-Component Floating-Point Add
// add dest, src0, src1
// dest.x = src0.x + src1.x;
@ -1373,9 +1319,7 @@ struct AluInstruction {
return vector_write_mask() || is_export() ||
AluVectorOpcodeHasSideEffects(vector_opcode());
}
AluVectorOpcode vector_opcode() const {
return static_cast<AluVectorOpcode>(data_.vector_opc);
}
AluVectorOpcode vector_opcode() const { return data_.vector_opc; }
uint32_t vector_write_mask() const { return data_.vector_write_mask; }
uint32_t vector_dest() const { return data_.vector_dest; }
bool is_vector_dest_relative() const { return data_.vector_dest_rel == 1; }
@ -1385,9 +1329,7 @@ struct AluInstruction {
return scalar_opcode() != AluScalarOpcode::kRetainPrev ||
(!is_export() && scalar_write_mask() != 0);
}
AluScalarOpcode scalar_opcode() const {
return static_cast<AluScalarOpcode>(data_.scalar_opc);
}
AluScalarOpcode scalar_opcode() const { return data_.scalar_opc; }
uint32_t scalar_write_mask() const { return data_.scalar_write_mask; }
uint32_t scalar_dest() const { return data_.scalar_dest; }
bool is_scalar_dest_relative() const { return data_.scalar_dest_rel == 1; }
@ -1459,7 +1401,7 @@ struct AluInstruction {
uint32_t scalar_write_mask : 4;
uint32_t vector_clamp : 1;
uint32_t scalar_clamp : 1;
uint32_t scalar_opc : 6; // instr_scalar_opc_t
AluScalarOpcode scalar_opc : 6;
});
XEPACKEDSTRUCTANONYMOUS({
uint32_t src3_swiz : 8;
@ -1478,7 +1420,7 @@ struct AluInstruction {
uint32_t src3_reg : 8;
uint32_t src2_reg : 8;
uint32_t src1_reg : 8;
uint32_t vector_opc : 5; // instr_vector_opc_t
AluVectorOpcode vector_opc : 5;
uint32_t src3_sel : 1;
uint32_t src2_sel : 1;
uint32_t src1_sel : 1;

5
src/xenia/gpu/vulkan/buffer_cache.cc
View File

@ -654,9 +654,8 @@ VkDescriptorSet BufferCache::PrepareVertexSet(
// trace_writer_.WriteMemoryRead(physical_address, source_length);
// Upload (or get a cached copy of) the buffer.
auto buffer_ref =
UploadVertexBuffer(command_buffer, physical_address, source_length,
static_cast<Endian>(fetch->endian), fence);
auto buffer_ref = UploadVertexBuffer(command_buffer, physical_address,
source_length, fetch->endian, fence);
if (buffer_ref.second == VK_WHOLE_SIZE) {
// Failed to upload buffer.
XELOGW("Failed to upload vertex buffer!");

63
src/xenia/gpu/vulkan/pipeline_cache.cc
View File

@ -364,7 +364,7 @@ VkPipeline PipelineCache::GetPipeline(const RenderState* render_state,
}
bool PipelineCache::TranslateShader(VulkanShader* shader,
xenos::xe_gpu_program_cntl_t cntl) {
reg::SQ_PROGRAM_CNTL cntl) {
// Perform translation.
// If this fails the shader will be marked as invalid and ignored later.
if (!shader_translator_->Translate(shader, PrimitiveType::kNone, cntl)) {
@ -808,44 +808,33 @@ bool PipelineCache::SetDynamicState(VkCommandBuffer command_buffer,
}
bool push_constants_dirty = full_update || viewport_state_dirty;
push_constants_dirty |=
SetShadowRegister(&regs.sq_program_cntl, XE_GPU_REG_SQ_PROGRAM_CNTL);
push_constants_dirty |= SetShadowRegister(&regs.sq_program_cntl.value,
XE_GPU_REG_SQ_PROGRAM_CNTL);
push_constants_dirty |=
SetShadowRegister(&regs.sq_context_misc, XE_GPU_REG_SQ_CONTEXT_MISC);
push_constants_dirty |=
SetShadowRegister(&regs.rb_colorcontrol, XE_GPU_REG_RB_COLORCONTROL);
push_constants_dirty |=
SetShadowRegister(&regs.rb_color_info, XE_GPU_REG_RB_COLOR_INFO);
SetShadowRegister(&regs.rb_color_info.value, XE_GPU_REG_RB_COLOR_INFO);
push_constants_dirty |=
SetShadowRegister(&regs.rb_color1_info, XE_GPU_REG_RB_COLOR1_INFO);
SetShadowRegister(&regs.rb_color1_info.value, XE_GPU_REG_RB_COLOR1_INFO);
push_constants_dirty |=
SetShadowRegister(&regs.rb_color2_info, XE_GPU_REG_RB_COLOR2_INFO);
SetShadowRegister(&regs.rb_color2_info.value, XE_GPU_REG_RB_COLOR2_INFO);
push_constants_dirty |=
SetShadowRegister(&regs.rb_color3_info, XE_GPU_REG_RB_COLOR3_INFO);
SetShadowRegister(&regs.rb_color3_info.value, XE_GPU_REG_RB_COLOR3_INFO);
push_constants_dirty |=
SetShadowRegister(&regs.rb_alpha_ref, XE_GPU_REG_RB_ALPHA_REF);
push_constants_dirty |=
SetShadowRegister(&regs.pa_su_point_size, XE_GPU_REG_PA_SU_POINT_SIZE);
if (push_constants_dirty) {
xenos::xe_gpu_program_cntl_t program_cntl;
program_cntl.dword_0 = regs.sq_program_cntl;
// Normal vertex shaders only, for now.
// TODO(benvanik): transform feedback/memexport.
// https://github.com/freedreno/freedreno/blob/master/includes/a2xx.xml.h
// Draw calls skipped if they have unsupported export modes.
// 0 = positionOnly
// 1 = unused
// 2 = sprite
// 3 = edge
// 4 = kill
// 5 = spriteKill
// 6 = edgeKill
// 7 = multipass
assert_true(program_cntl.vs_export_mode == 0 ||
program_cntl.vs_export_mode == 2 ||
program_cntl.vs_export_mode == 7);
assert_false(program_cntl.gen_index_vtx);
assert_true(regs.sq_program_cntl.vs_export_mode ==
xenos::VertexShaderExportMode::kPosition1Vector ||
regs.sq_program_cntl.vs_export_mode ==
xenos::VertexShaderExportMode::kPosition2VectorsSprite ||
regs.sq_program_cntl.vs_export_mode ==
xenos::VertexShaderExportMode::kMultipass);
assert_false(regs.sq_program_cntl.gen_index_vtx);
SpirvPushConstants push_constants = {};
@ -909,7 +898,8 @@ bool PipelineCache::SetDynamicState(VkCommandBuffer command_buffer,
// Whether to populate a register in the pixel shader with frag coord.
int ps_param_gen = (regs.sq_context_misc >> 8) & 0xFF;
push_constants.ps_param_gen = program_cntl.param_gen ? ps_param_gen : -1;
push_constants.ps_param_gen =
regs.sq_program_cntl.param_gen ? ps_param_gen : -1;
vkCmdPushConstants(command_buffer, pipeline_layout_,
VK_SHADER_STAGE_VERTEX_BIT |
@ -1061,7 +1051,8 @@ PipelineCache::UpdateStatus PipelineCache::UpdateShaderStages(
bool dirty = false;
dirty |= SetShadowRegister(&regs.pa_su_sc_mode_cntl,
XE_GPU_REG_PA_SU_SC_MODE_CNTL);
dirty |= SetShadowRegister(&regs.sq_program_cntl, XE_GPU_REG_SQ_PROGRAM_CNTL);
dirty |= SetShadowRegister(&regs.sq_program_cntl.value,
XE_GPU_REG_SQ_PROGRAM_CNTL);
dirty |= regs.vertex_shader != vertex_shader;
dirty |= regs.pixel_shader != pixel_shader;
dirty |= regs.primitive_type != primitive_type;
@ -1073,17 +1064,14 @@ PipelineCache::UpdateStatus PipelineCache::UpdateShaderStages(
return UpdateStatus::kCompatible;
}
xenos::xe_gpu_program_cntl_t sq_program_cntl;
sq_program_cntl.dword_0 = regs.sq_program_cntl;
if (!vertex_shader->is_translated() &&
!TranslateShader(vertex_shader, sq_program_cntl)) {
!TranslateShader(vertex_shader, regs.sq_program_cntl)) {
XELOGE("Failed to translate the vertex shader!");
return UpdateStatus::kError;
}
if (pixel_shader && !pixel_shader->is_translated() &&
!TranslateShader(pixel_shader, sq_program_cntl)) {
!TranslateShader(pixel_shader, regs.sq_program_cntl)) {
XELOGE("Failed to translate the pixel shader!");
return UpdateStatus::kError;
}
@ -1513,16 +1501,15 @@ PipelineCache::UpdateStatus PipelineCache::UpdateColorBlendState() {
auto& state_info = update_color_blend_state_info_;
bool dirty = false;
dirty |= SetShadowRegister(&regs.rb_colorcontrol, XE_GPU_REG_RB_COLORCONTROL);
dirty |= SetShadowRegister(&regs.rb_color_mask, XE_GPU_REG_RB_COLOR_MASK);
dirty |=
SetShadowRegister(&regs.rb_blendcontrol[0], XE_GPU_REG_RB_BLENDCONTROL_0);
SetShadowRegister(&regs.rb_blendcontrol[0], XE_GPU_REG_RB_BLENDCONTROL0);
dirty |=
SetShadowRegister(&regs.rb_blendcontrol[1], XE_GPU_REG_RB_BLENDCONTROL_1);
SetShadowRegister(&regs.rb_blendcontrol[1], XE_GPU_REG_RB_BLENDCONTROL1);
dirty |=
SetShadowRegister(&regs.rb_blendcontrol[2], XE_GPU_REG_RB_BLENDCONTROL_2);
SetShadowRegister(&regs.rb_blendcontrol[2], XE_GPU_REG_RB_BLENDCONTROL2);
dirty |=
SetShadowRegister(&regs.rb_blendcontrol[3], XE_GPU_REG_RB_BLENDCONTROL_3);
SetShadowRegister(&regs.rb_blendcontrol[3], XE_GPU_REG_RB_BLENDCONTROL3);
dirty |= SetShadowRegister(&regs.rb_modecontrol, XE_GPU_REG_RB_MODECONTROL);
XXH64_update(&hash_state_, &regs, sizeof(regs));
if (!dirty) {
@ -1568,7 +1555,7 @@ PipelineCache::UpdateStatus PipelineCache::UpdateColorBlendState() {
for (int i = 0; i < 4; ++i) {
uint32_t blend_control = regs.rb_blendcontrol[i];
auto& attachment_state = attachment_states[i];
attachment_state.blendEnable = !(regs.rb_colorcontrol & 0x20);
attachment_state.blendEnable = (blend_control & 0x1FFF1FFF) != 0x00010001;
// A2XX_RB_BLEND_CONTROL_COLOR_SRCBLEND
attachment_state.srcColorBlendFactor =
kBlendFactorMap[(blend_control & 0x0000001F) >> 0];

15
src/xenia/gpu/vulkan/pipeline_cache.h
View File

@ -79,7 +79,7 @@ class PipelineCache {
// state.
VkPipeline GetPipeline(const RenderState* render_state, uint64_t hash_key);
bool TranslateShader(VulkanShader* shader, xenos::xe_gpu_program_cntl_t cntl);
bool TranslateShader(VulkanShader* shader, reg::SQ_PROGRAM_CNTL cntl);
void DumpShaderDisasmAMD(VkPipeline pipeline);
void DumpShaderDisasmNV(const VkGraphicsPipelineCreateInfo& info);
@ -170,7 +170,7 @@ class PipelineCache {
struct UpdateShaderStagesRegisters {
PrimitiveType primitive_type;
uint32_t pa_su_sc_mode_cntl;
uint32_t sq_program_cntl;
reg::SQ_PROGRAM_CNTL sq_program_cntl;
VulkanShader* vertex_shader;
VulkanShader* pixel_shader;
@ -256,7 +256,6 @@ class PipelineCache {
VkPipelineDepthStencilStateCreateInfo update_depth_stencil_state_info_;
struct UpdateColorBlendStateRegisters {
uint32_t rb_colorcontrol;
uint32_t rb_color_mask;
uint32_t rb_blendcontrol[4];
uint32_t rb_modecontrol;
@ -290,13 +289,13 @@ class PipelineCache {
float rb_blend_rgba[4];
uint32_t rb_stencilrefmask;
uint32_t sq_program_cntl;
reg::SQ_PROGRAM_CNTL sq_program_cntl;
uint32_t sq_context_misc;
uint32_t rb_colorcontrol;
uint32_t rb_color_info;
uint32_t rb_color1_info;
uint32_t rb_color2_info;
uint32_t rb_color3_info;
reg::RB_COLOR_INFO rb_color_info;
reg::RB_COLOR_INFO rb_color1_info;
reg::RB_COLOR_INFO rb_color2_info;
reg::RB_COLOR_INFO rb_color3_info;
float rb_alpha_ref;
uint32_t pa_su_point_size;

16
src/xenia/gpu/vulkan/vulkan_command_processor.cc
View File

@ -962,7 +962,7 @@ bool VulkanCommandProcessor::IssueCopy() {
break;
}
assert_true(fetch->type == 3);
assert_true(fetch->endian == 2);
assert_true(fetch->endian == Endian::k8in32);
assert_true(fetch->size == 6);
const uint8_t* vertex_addr = memory_->TranslatePhysical(fetch->address << 2);
trace_writer_.WriteMemoryRead(fetch->address << 2, fetch->size * 4);
@ -974,7 +974,7 @@ bool VulkanCommandProcessor::IssueCopy() {
float dest_points[6];
for (int i = 0; i < 6; i++) {
dest_points[i] =
GpuSwap(xe::load<float>(vertex_addr + i * 4), Endian(fetch->endian)) +
GpuSwap(xe::load<float>(vertex_addr + i * 4), fetch->endian) +
vtx_offset;
}
@ -1000,10 +1000,10 @@ bool VulkanCommandProcessor::IssueCopy() {
if (is_color_source) {
// Source from a color target.
reg::RB_COLOR_INFO color_info[4] = {
regs[XE_GPU_REG_RB_COLOR_INFO].u32,
regs[XE_GPU_REG_RB_COLOR1_INFO].u32,
regs[XE_GPU_REG_RB_COLOR2_INFO].u32,
regs[XE_GPU_REG_RB_COLOR3_INFO].u32,
regs.Get<reg::RB_COLOR_INFO>(),
regs.Get<reg::RB_COLOR_INFO>(XE_GPU_REG_RB_COLOR1_INFO),
regs.Get<reg::RB_COLOR_INFO>(XE_GPU_REG_RB_COLOR2_INFO),
regs.Get<reg::RB_COLOR_INFO>(XE_GPU_REG_RB_COLOR3_INFO),
};
color_edram_base = color_info[copy_src_select].color_base;
color_format = color_info[copy_src_select].color_format;
@ -1023,7 +1023,7 @@ bool VulkanCommandProcessor::IssueCopy() {
Endian resolve_endian = Endian::k8in32;
if (copy_regs->copy_dest_info.copy_dest_endian <= Endian128::k16in32) {
resolve_endian =
static_cast<Endian>(copy_regs->copy_dest_info.copy_dest_endian.value());
static_cast<Endian>(copy_regs->copy_dest_info.copy_dest_endian);
}
// Demand a resolve texture from the texture cache.
@ -1289,7 +1289,7 @@ bool VulkanCommandProcessor::IssueCopy() {
// Perform any requested clears.
uint32_t copy_depth_clear = regs[XE_GPU_REG_RB_DEPTH_CLEAR].u32;
uint32_t copy_color_clear = regs[XE_GPU_REG_RB_COLOR_CLEAR].u32;
uint32_t copy_color_clear_low = regs[XE_GPU_REG_RB_COLOR_CLEAR_LOW].u32;
uint32_t copy_color_clear_low = regs[XE_GPU_REG_RB_COLOR_CLEAR_LO].u32;
assert_true(copy_color_clear == copy_color_clear_low);
if (color_clear_enabled) {

265
src/xenia/gpu/xenos.h
View File

@ -80,12 +80,6 @@ inline bool IsPrimitiveTwoFaced(bool tessellated, PrimitiveType type) {
return false;
}
enum class TessellationMode : uint32_t {
kDiscrete = 0,
kContinuous = 1,
kAdaptive = 2,
};
enum class Dimension : uint32_t {
k1D = 0,
k2D = 1,
@ -167,14 +161,14 @@ enum class SampleLocation : uint32_t {
};
enum class Endian : uint32_t {
kUnspecified = 0,
kNone = 0,
k8in16 = 1,
k8in32 = 2,
k16in32 = 3,
};
enum class Endian128 : uint32_t {
kUnspecified = 0,
kNone = 0,
k8in16 = 1,
k8in32 = 2,
k16in32 = 3,
@ -231,6 +225,77 @@ enum class DepthRenderTargetFormat : uint32_t {
kD24FS8 = 1,
};
// a2xx_sq_surfaceformat +
// https://github.com/indirivacua/RAGE-Console-Texture-Editor/blob/master/Console.Xbox360.Graphics.pas
enum class TextureFormat : uint32_t {
k_1_REVERSE = 0,
k_1 = 1,
k_8 = 2,
k_1_5_5_5 = 3,
k_5_6_5 = 4,
k_6_5_5 = 5,
k_8_8_8_8 = 6,
k_2_10_10_10 = 7,
k_8_A = 8,
k_8_B = 9,
k_8_8 = 10,
k_Cr_Y1_Cb_Y0_REP = 11,
k_Y1_Cr_Y0_Cb_REP = 12,
k_16_16_EDRAM = 13,
k_8_8_8_8_A = 14,
k_4_4_4_4 = 15,
k_10_11_11 = 16,
k_11_11_10 = 17,
k_DXT1 = 18,
k_DXT2_3 = 19,
k_DXT4_5 = 20,
k_16_16_16_16_EDRAM = 21,
k_24_8 = 22,
k_24_8_FLOAT = 23,
k_16 = 24,
k_16_16 = 25,
k_16_16_16_16 = 26,
k_16_EXPAND = 27,
k_16_16_EXPAND = 28,
k_16_16_16_16_EXPAND = 29,
k_16_FLOAT = 30,
k_16_16_FLOAT = 31,
k_16_16_16_16_FLOAT = 32,
k_32 = 33,
k_32_32 = 34,
k_32_32_32_32 = 35,
k_32_FLOAT = 36,
k_32_32_FLOAT = 37,
k_32_32_32_32_FLOAT = 38,
k_32_AS_8 = 39,
k_32_AS_8_8 = 40,
k_16_MPEG = 41,
k_16_16_MPEG = 42,
k_8_INTERLACED = 43,
k_32_AS_8_INTERLACED = 44,
k_32_AS_8_8_INTERLACED = 45,
k_16_INTERLACED = 46,
k_16_MPEG_INTERLACED = 47,
k_16_16_MPEG_INTERLACED = 48,
k_DXN = 49,
k_8_8_8_8_AS_16_16_16_16 = 50,
k_DXT1_AS_16_16_16_16 = 51,
k_DXT2_3_AS_16_16_16_16 = 52,
k_DXT4_5_AS_16_16_16_16 = 53,
k_2_10_10_10_AS_16_16_16_16 = 54,
k_10_11_11_AS_16_16_16_16 = 55,
k_11_11_10_AS_16_16_16_16 = 56,
k_32_32_32_FLOAT = 57,
k_DXT3A = 58,
k_DXT5A = 59,
k_CTX1 = 60,
k_DXT3A_AS_1_1_1_1 = 61,
k_8_8_8_8_GAMMA_EDRAM = 62,
k_2_10_10_10_FLOAT_EDRAM = 63,
kUnknown = 0xFFFFFFFFu,
};
// Subset of a2xx_sq_surfaceformat - formats that RTs can be resolved to.
enum class ColorFormat : uint32_t {
k_8 = 2,
@ -334,6 +399,28 @@ inline int GetVertexFormatSizeInWords(VertexFormat format) {
}
}
enum class CompareFunction : uint32_t {
kNever = 0b000,
kLess = 0b001,
kEqual = 0b010,
kLessEqual = 0b011,
kGreater = 0b100,
kNotEqual = 0b101,
kGreaterEqual = 0b110,
kAlways = 0b111,
};
enum class StencilOp : uint32_t {
kKeep = 0,
kZero = 1,
kReplace = 2,
kIncrementClamp = 3,
kDecrementClamp = 4,
kInvert = 5,
kIncrementWrap = 6,
kDecrementWrap = 7,
};
// adreno_rb_blend_factor
enum class BlendFactor : uint32_t {
kZero = 0,
@ -351,11 +438,7 @@ enum class BlendFactor : uint32_t {
kConstantAlpha = 14,
kOneMinusConstantAlpha = 15,
kSrcAlphaSaturate = 16,
// SRC1 likely not used on the Xbox 360 - only available in Direct3D 9Ex.
kSrc1Color = 20,
kOneMinusSrc1Color = 21,
kSrc1Alpha = 22,
kOneMinusSrc1Alpha = 23,
// SRC1 added on Adreno.
};
enum class BlendOp : uint32_t {
@ -375,6 +458,57 @@ typedef enum {
XE_GPU_INVALIDATE_MASK_ALL = 0x7FFF,
} XE_GPU_INVALIDATE_MASK;
// instr_arbitrary_filter_t
enum class ArbitraryFilter : uint32_t {
k2x4Sym = 0,
k2x4Asym = 1,
k4x2Sym = 2,
k4x2Asym = 3,
k4x4Sym = 4,
k4x4Asym = 5,
kUseFetchConst = 7,
};
// a2xx_sq_ps_vtx_mode
enum class VertexShaderExportMode : uint32_t {
kPosition1Vector = 0,
kPosition2VectorsSprite = 2,
kPosition2VectorsEdge = 3,
kPosition2VectorsKill = 4,
kPosition2VectorsSpriteKill = 5,
kPosition2VectorsEdgeKill = 6,
kMultipass = 7,
};
enum class SampleControl : uint32_t {
kCentroidsOnly = 0,
kCentersOnly = 1,
kCentroidsAndCenters = 2,
};
enum class VGTOutputPath : uint32_t {
kVertexReuse = 0,
kTessellationEnable = 1,
kPassthru = 2,
};
enum class TessellationMode : uint32_t {
kDiscrete = 0,
kContinuous = 1,
kAdaptive = 2,
};
enum class PolygonModeEnable : uint32_t {
kDisabled = 0, // Render triangles.
kDualMode = 1, // Send 2 sets of 3 polygons with the specified polygon type.
};
enum class PolygonType : uint32_t {
kPoints = 0,
kLines = 1,
kTriangles = 2,
};
enum class ModeControl : uint32_t {
kIgnore = 0,
kColorDepth = 4,
@ -426,7 +560,7 @@ typedef enum {
inline uint16_t GpuSwap(uint16_t value, Endian endianness) {
switch (endianness) {
case Endian::kUnspecified:
case Endian::kNone:
// No swap.
return value;
case Endian::k8in16:
@ -441,7 +575,7 @@ inline uint16_t GpuSwap(uint16_t value, Endian endianness) {
inline uint32_t GpuSwap(uint32_t value, Endian endianness) {
switch (endianness) {
default:
case Endian::kUnspecified:
case Endian::kNone:
// No swap.
return value;
case Endian::k8in16:
@ -471,35 +605,15 @@ inline uint32_t GpuToCpu(uint32_t p) { return p; }
inline uint32_t CpuToGpu(uint32_t p) { return p & 0x1FFFFFFF; }
// XE_GPU_REG_SQ_PROGRAM_CNTL
typedef union {
XEPACKEDSTRUCTANONYMOUS({
uint32_t vs_regs : 6;
uint32_t unk_0 : 2;
uint32_t ps_regs : 6;
uint32_t unk_1 : 2;
uint32_t vs_resource : 1;
uint32_t ps_resource : 1;
uint32_t param_gen : 1;
uint32_t gen_index_pix : 1;
uint32_t vs_export_count : 4;
uint32_t vs_export_mode : 3;
uint32_t ps_export_depth : 1;
uint32_t ps_export_count : 3;
uint32_t gen_index_vtx : 1;
});
XEPACKEDSTRUCTANONYMOUS({ uint32_t dword_0; });
} xe_gpu_program_cntl_t;
// XE_GPU_REG_SHADER_CONSTANT_FETCH_*
XEPACKEDUNION(xe_gpu_vertex_fetch_t, {
XEPACKEDSTRUCTANONYMOUS({
uint32_t type : 2; // +0
uint32_t address : 30; // +2
uint32_t address : 30; // +2 address in dwords
uint32_t endian : 2; // +0
uint32_t size : 24; // +2 size in words
uint32_t unk1 : 6; // +26
Endian endian : 2; // +0
uint32_t size : 24; // +2 size in words
uint32_t unk1 : 6; // +26
});
XEPACKEDSTRUCTANONYMOUS({
uint32_t dword_0;
@ -510,34 +624,36 @@ XEPACKEDUNION(xe_gpu_vertex_fetch_t, {
// XE_GPU_REG_SHADER_CONSTANT_FETCH_*
XEPACKEDUNION(xe_gpu_texture_fetch_t, {
XEPACKEDSTRUCTANONYMOUS({
uint32_t type : 2; // +0 dword_0
uint32_t sign_x : 2; // +2
uint32_t sign_y : 2; // +4
uint32_t sign_z : 2; // +6
uint32_t sign_w : 2; // +8
uint32_t clamp_x : 3; // +10
uint32_t clamp_y : 3; // +13
uint32_t clamp_z : 3; // +16
uint32_t unused_0 : 3; // +19
uint32_t pitch : 9; // +22 byte_pitch >> 5
uint32_t tiled : 1; // +31
uint32_t type : 2; // +0 dword_0
TextureSign sign_x : 2; // +2
TextureSign sign_y : 2; // +4
TextureSign sign_z : 2; // +6
TextureSign sign_w : 2; // +8
ClampMode clamp_x : 3; // +10
ClampMode clamp_y : 3; // +13
ClampMode clamp_z : 3; // +16
uint32_t signed_rf_mode_all : 1; // +19
// TODO(Triang3l): 1 or 2 dim_tbd bits?
uint32_t unk_0 : 2; // +20
uint32_t pitch : 9; // +22 byte_pitch >> 5
uint32_t tiled : 1; // +31
uint32_t format : 6; // +0 dword_1
uint32_t endianness : 2; // +6
uint32_t request_size : 2; // +8
uint32_t stacked : 1; // +10
uint32_t clamp_policy : 1; // +11 d3d/opengl
uint32_t base_address : 20; // +12
TextureFormat format : 6; // +0 dword_1
Endian endianness : 2; // +6
uint32_t request_size : 2; // +8
uint32_t stacked : 1; // +10
uint32_t nearest_clamp_policy : 1; // +11 d3d/opengl
uint32_t base_address : 20; // +12 base address >> 12
union { // dword_2
struct {
uint32_t width : 24;
uint32_t unused : 8;
uint32_t : 8;
} size_1d;
struct {
uint32_t width : 13;
uint32_t height : 13;
uint32_t unused : 6;
uint32_t : 6;
} size_2d;
struct {
uint32_t width : 13;
@ -551,15 +667,16 @@ XEPACKEDUNION(xe_gpu_texture_fetch_t, {
} size_3d;
};
uint32_t num_format : 1; // +0 dword_3 frac/int
uint32_t swizzle : 12; // +1 xyzw, 3b each (XE_GPU_SWIZZLE)
int32_t exp_adjust : 6; // +13
uint32_t mag_filter : 2; // +19
uint32_t min_filter : 2; // +21
uint32_t mip_filter : 2; // +23
uint32_t aniso_filter : 3; // +25
uint32_t unused_3 : 3; // +28
uint32_t border_size : 1; // +31
uint32_t num_format : 1; // +0 dword_3 frac/int
// xyzw, 3b each (XE_GPU_SWIZZLE)
uint32_t swizzle : 12; // +1
int32_t exp_adjust : 6; // +13
TextureFilter mag_filter : 2; // +19
TextureFilter min_filter : 2; // +21
TextureFilter mip_filter : 2; // +23
AnisoFilter aniso_filter : 3; // +25
xenos::ArbitraryFilter arbitrary_filter : 3; // +28
uint32_t border_size : 1; // +31
uint32_t vol_mag_filter : 1; // +0 dword_4
uint32_t vol_min_filter : 1; // +1
@ -571,13 +688,13 @@ XEPACKEDUNION(xe_gpu_texture_fetch_t, {
int32_t grad_exp_adjust_h : 5; // +22
int32_t grad_exp_adjust_v : 5; // +27
uint32_t border_color : 2; // +0 dword_5
uint32_t force_bcw_max : 1; // +2
uint32_t tri_clamp : 2; // +3
int32_t aniso_bias : 4; // +5
uint32_t dimension : 2; // +9
uint32_t packed_mips : 1; // +11
uint32_t mip_address : 20; // +12
BorderColor border_color : 2; // +0 dword_5
uint32_t force_bc_w_to_max : 1; // +2
uint32_t tri_clamp : 2; // +3
int32_t aniso_bias : 4; // +5
uint32_t dimension : 2; // +9
uint32_t packed_mips : 1; // +11
uint32_t mip_address : 20; // +12 mip address >> 12
});
XEPACKEDSTRUCTANONYMOUS({
uint32_t dword_0;