Merge branch 'master' into vulkan

src/xenia/gpu/d3d12/d3d12_command_processor.cc

@@ -1822,13 +1822,13 @@ bool D3D12CommandProcessor::IssueDraw(xenos::PrimitiveType primitive_type,
                                       uint32_t index_count,
                                       IndexBufferInfo* index_buffer_info,
                                       bool major_mode_explicit) {
-  auto device = GetD3D12Context().GetD3D12Provider().GetDevice();
-  auto& regs = *register_file_;
-
 #if XE_UI_D3D12_FINE_GRAINED_DRAW_SCOPES
   SCOPE_profile_cpu_f("gpu");
 #endif  // XE_UI_D3D12_FINE_GRAINED_DRAW_SCOPES
 
+  ID3D12Device* device = GetD3D12Context().GetD3D12Provider().GetDevice();
+  const RegisterFile& regs = *register_file_;
+
   xenos::ModeControl edram_mode = regs.Get<reg::RB_MODECONTROL>().edram_mode;
   if (edram_mode == xenos::ModeControl::kCopy) {
     // Special copy handling.
@@ -3241,9 +3241,10 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
         poly_offset_back_offset = poly_offset_front_offset;
       }
     }
-    // "slope computed in subpixels ([...] 1/16)" - R5xx Acceleration.
-    poly_offset_front_scale *= (1.0f / 16.0f) * resolution_scale;
-    poly_offset_back_scale *= (1.0f / 16.0f) * resolution_scale;
+    float poly_offset_scale_factor =
+        xenos::kPolygonOffsetScaleSubpixelUnit * resolution_scale;
+    poly_offset_front_scale *= poly_offset_scale_factor;
+    poly_offset_back_scale *= poly_offset_scale_factor;
     dirty |= system_constants_.edram_poly_offset_front_scale !=
              poly_offset_front_scale;
     system_constants_.edram_poly_offset_front_scale = poly_offset_front_scale;

src/xenia/gpu/d3d12/pipeline_cache.cc

@@ -36,6 +36,7 @@
 #include "xenia/gpu/d3d12/d3d12_render_target_cache.h"
 #include "xenia/gpu/draw_util.h"
 #include "xenia/gpu/gpu_flags.h"
+#include "xenia/gpu/xenos.h"
 #include "xenia/ui/d3d12/d3d12_util.h"
 
 DEFINE_bool(d3d12_dxbc_disasm, false,
@@ -1340,7 +1341,6 @@ bool PipelineCache::GetCurrentStateDescription(
         break;
       case xenos::PrimitiveType::kLineList:
       case xenos::PrimitiveType::kLineStrip:
-      case xenos::PrimitiveType::kLineLoop:
       // Quads are emulated as line lists with adjacency.
       case xenos::PrimitiveType::kQuadList:
       case xenos::PrimitiveType::k2DLineStrip:
@@ -1457,7 +1457,7 @@ bool PipelineCache::GetCurrentStateDescription(
     }
   }
   if (!edram_rov_used) {
-    float poly_offset_host_scale = draw_util::GetD3D10PolygonOffsetScale(
+    float poly_offset_host_scale = draw_util::GetD3D10PolygonOffsetFactor(
         regs.Get<reg::RB_DEPTH_INFO>().depth_format, true);
     // Using ceil here just in case a game wants the offset but passes a value
     // that is too small - it's better to apply more offset than to make depth
@@ -1468,7 +1468,7 @@ bool PipelineCache::GetCurrentStateDescription(
         (poly_offset < 0.0f ? -1 : 1);
     // "slope computed in subpixels ([...] 1/16)" - R5xx Acceleration.
     description_out.depth_bias_slope_scaled =
-        poly_offset_scale * (1.0f / 16.0f);
+        poly_offset_scale * xenos::kPolygonOffsetScaleSubpixelUnit;
   }
   if (tessellated && cvars::d3d12_tessellation_wireframe) {
     description_out.fill_mode_wireframe = 1;

src/xenia/gpu/draw_util.h

@@ -106,7 +106,7 @@ inline reg::RB_DEPTHCONTROL GetDepthControlForCurrentEdramMode(
   return regs.Get<reg::RB_DEPTHCONTROL>();
 }
 
-constexpr float GetD3D10PolygonOffsetScale(
+constexpr float GetD3D10PolygonOffsetFactor(
     xenos::DepthRenderTargetFormat depth_format, bool float24_as_0_to_0_5) {
   if (depth_format == xenos::DepthRenderTargetFormat::kD24S8) {
     return float(1 << 24);

src/xenia/gpu/primitive_processor.cc

@@ -458,6 +458,15 @@ bool PrimitiveProcessor::Process(ProcessingResult& result_out) {
       }
     } else {
       if (regs.Get<reg::PA_SU_SC_MODE_CNTL>().multi_prim_ib_ena) {
+        switch (guest_primitive_type) {
+          case xenos::PrimitiveType::kLineStrip:
+          case xenos::PrimitiveType::kTriangleFan:
+          case xenos::PrimitiveType::kTriangleStrip:
+          case xenos::PrimitiveType::kLineLoop:
+          case xenos::PrimitiveType::kQuadStrip:
+          case xenos::PrimitiveType::kPolygon:
+          case xenos::PrimitiveType::k2DLineStrip:
+          case xenos::PrimitiveType::k2DTriStrip:
             guest_primitive_reset_index_guest_endian = xenos::GpuSwap(
                 regs.Get<reg::VGT_MULTI_PRIM_IB_RESET_INDX>().reset_indx,
                 guest_index_endian);
@@ -468,6 +477,15 @@ bool PrimitiveProcessor::Process(ProcessingResult& result_out) {
                 guest_index_format == xenos::IndexFormat::kInt16
                     ? guest_primitive_reset_index_guest_endian <= UINT16_MAX
                     : true;
+            break;
+          default:
+            // Vulkan explicitly disallows primitive restart index for "list"
+            // topologies. In Direct3D 12, it's valid for non-strips, but has
+            // implementation-defined behavior. Make backend usage simpler by
+            // explicitly filtering lists out, and hope the guest never uses
+            // primitive reset for lists.
+            break;
+        }
       }
     }
 

src/xenia/gpu/primitive_processor.h

@@ -138,7 +138,8 @@ class PrimitiveProcessor {
     xenos::IndexFormat host_index_format;
     xenos::Endian host_index_endian;
     // The reset index, if enabled, is always 0xFFFF for host_index_format
-    // kInt16 and 0xFFFFFFFF for kInt32.
+    // kInt16 and 0xFFFFFFFF for kInt32. Never enabled for "list" primitive
+    // types, thus safe for direct usage on Vulkan.
     bool host_primitive_reset_enabled;
     // Backend-specific handle for the index buffer valid for the current draw,
     // only valid for index_buffer_type kHostConverted and kHostBuiltin.

src/xenia/gpu/xenos.h

@@ -347,6 +347,16 @@ constexpr float UNorm24To32(uint32_t n24) {
   return float(n24 + (n24 >> 23)) * (1.0f / float(1 << 24));
 }
 
+// Scale for conversion of slope scales from PA_SU_POLY_OFFSET_FRONT/BACK_SCALE
+// units to those used when the slope is computed from the difference between
+// adjacent pixels, for conversion from the guest to common host APIs or to
+// calculation using max(|ddx(z)|, |ddy(z)|).
+// "slope computed in subpixels (1/12 or 1/16)" - R5xx Acceleration.
+// But the correct scale for conversion of the slope scale from subpixels to
+// pixels is likely 1/16 according to:
+// https://github.com/mesa3d/mesa/blob/54ad9b444c8e73da498211870e785239ad3ff1aa/src/gallium/drivers/radeonsi/si_state.c#L946
+constexpr float kPolygonOffsetScaleSubpixelUnit = 1.0f / 16.0f;
+
 constexpr uint32_t kColorRenderTargetFormatBits = 4;
 constexpr uint32_t kDepthRenderTargetFormatBits = 1;
 constexpr uint32_t kRenderTargetFormatBits =