Merge pull request #9226 from Kelebek1/regs_regression

[video_core] Fix a couple regs regressions

src/video_core/engines/maxwell_3d.cpp

@@ -249,6 +249,11 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume
         return;
     case MAXWELL3D_REG_INDEX(fragment_barrier):
         return rasterizer->FragmentBarrier();
+    case MAXWELL3D_REG_INDEX(invalidate_texture_data_cache):
+        rasterizer->InvalidateGPUCache();
+        return rasterizer->WaitForIdle();
+    case MAXWELL3D_REG_INDEX(tiled_cache_barrier):
+        return rasterizer->TiledCacheBarrier();
     }
 }
 

src/video_core/engines/maxwell_3d.h

@@ -707,7 +707,7 @@ public:
                 case Size::Size_A2_B10_G10_R10:
                     return "2_10_10_10";
                 case Size::Size_B10_G11_R11:
-                    return "10_11_12";
+                    return "10_11_11";
                 default:
                     ASSERT(false);
                     return {};
@@ -2639,7 +2639,7 @@ public:
                 L2CacheControl l2_cache_control;                                       ///< 0x0218
                 InvalidateShaderCache invalidate_shader_cache;                         ///< 0x021C
                 INSERT_PADDING_BYTES_NOINIT(0xA8);
-                SyncInfo sync_info; ///< 0x02C8
+                SyncInfo sync_info;                                                    ///< 0x02C8
                 INSERT_PADDING_BYTES_NOINIT(0x4);
                 u32 prim_circular_buffer_throttle;                                     ///< 0x02D0
                 u32 flush_invalidate_rop_mini_cache;                                   ///< 0x02D4
@@ -2731,7 +2731,11 @@ public:
                 s32 stencil_back_ref;                                                  ///< 0x0F54
                 u32 stencil_back_mask;                                                 ///< 0x0F58
                 u32 stencil_back_func_mask;                                            ///< 0x0F5C
-                INSERT_PADDING_BYTES_NOINIT(0x24);
+                INSERT_PADDING_BYTES_NOINIT(0x14);
+                u32 invalidate_texture_data_cache;                                     ///< 0x0F74 Assumed - Not in official docs.
+                INSERT_PADDING_BYTES_NOINIT(0x4);
+                u32 tiled_cache_barrier;                                               ///< 0x0F7C Assumed - Not in official docs.
+                INSERT_PADDING_BYTES_NOINIT(0x4);
                 VertexStreamSubstitute vertex_stream_substitute;                       ///< 0x0F84
                 u32 line_mode_clip_generated_edge_do_not_draw;                         ///< 0x0F8C
                 u32 color_mask_common;                                                 ///< 0x0F90
@@ -2791,7 +2795,8 @@ public:
                 FillViaTriangleMode fill_via_triangle_mode;                            ///< 0x113C
                 u32 blend_per_format_snorm8_unorm16_snorm16_enabled;                   ///< 0x1140
                 u32 flush_pending_writes_sm_gloal_store;                               ///< 0x1144
-                INSERT_PADDING_BYTES_NOINIT(0x18);
+                u32 conservative_raster_enable;                                        ///< 0x1148 Assumed - Not in official docs.
+                INSERT_PADDING_BYTES_NOINIT(0x14);
                 std::array<VertexAttribute, NumVertexAttributes> vertex_attrib_format; ///< 0x1160
                 std::array<MsaaSampleLocation, 4> multisample_sample_locations;        ///< 0x11E0
                 u32 offset_render_target_index_by_viewport_index;                      ///< 0x11F0
@@ -3287,6 +3292,8 @@ ASSERT_REG_POSITION(const_color_rendering, 0x0F40);
 ASSERT_REG_POSITION(stencil_back_ref, 0x0F54);
 ASSERT_REG_POSITION(stencil_back_mask, 0x0F58);
 ASSERT_REG_POSITION(stencil_back_func_mask, 0x0F5C);
+ASSERT_REG_POSITION(invalidate_texture_data_cache, 0x0F74);
+ASSERT_REG_POSITION(tiled_cache_barrier, 0x0F7C);
 ASSERT_REG_POSITION(vertex_stream_substitute, 0x0F84);
 ASSERT_REG_POSITION(line_mode_clip_generated_edge_do_not_draw, 0x0F8C);
 ASSERT_REG_POSITION(color_mask_common, 0x0F90);
@@ -3343,6 +3350,7 @@ ASSERT_REG_POSITION(post_ps_use_pre_ps_coverage, 0x1138);
 ASSERT_REG_POSITION(fill_via_triangle_mode, 0x113C);
 ASSERT_REG_POSITION(blend_per_format_snorm8_unorm16_snorm16_enabled, 0x1140);
 ASSERT_REG_POSITION(flush_pending_writes_sm_gloal_store, 0x1144);
+ASSERT_REG_POSITION(conservative_raster_enable, 0x1148);
 ASSERT_REG_POSITION(vertex_attrib_format, 0x1160);
 ASSERT_REG_POSITION(multisample_sample_locations, 0x11E0);
 ASSERT_REG_POSITION(offset_render_target_index_by_viewport_index, 0x11F0);

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -770,7 +770,7 @@ void RasterizerOpenGL::SyncStencilTestState() {
 
     if (regs.stencil_two_side_enable) {
         glStencilFuncSeparate(GL_BACK, MaxwellToGL::ComparisonOp(regs.stencil_back_op.func),
-                              regs.stencil_back_ref, regs.stencil_back_mask);
+                              regs.stencil_back_ref, regs.stencil_back_func_mask);
         glStencilOpSeparate(GL_BACK, MaxwellToGL::StencilOp(regs.stencil_back_op.fail),
                             MaxwellToGL::StencilOp(regs.stencil_back_op.zfail),
                             MaxwellToGL::StencilOp(regs.stencil_back_op.zpass));

src/video_core/renderer_vulkan/fixed_pipeline_state.cpp

@@ -90,6 +90,7 @@ void FixedPipelineState::Refresh(Tegra::Engines::Maxwell3D& maxwell3d,
     depth_format.Assign(static_cast<u32>(regs.zeta.format));
     y_negate.Assign(regs.window_origin.mode != Maxwell::WindowOrigin::Mode::UpperLeft ? 1 : 0);
     provoking_vertex_last.Assign(regs.provoking_vertex == Maxwell::ProvokingVertex::Last ? 1 : 0);
+    conservative_raster_enable.Assign(regs.conservative_raster_enable != 0 ? 1 : 0);
     smooth_lines.Assign(regs.line_anti_alias_enable != 0 ? 1 : 0);
 
     for (size_t i = 0; i < regs.rt.size(); ++i) {

src/video_core/renderer_vulkan/fixed_pipeline_state.h

@@ -193,6 +193,7 @@ struct FixedPipelineState {
         BitField<6, 5, u32> depth_format;
         BitField<11, 1, u32> y_negate;
         BitField<12, 1, u32> provoking_vertex_last;
+        BitField<13, 1, u32> conservative_raster_enable;
         BitField<14, 1, u32> smooth_lines;
     };
     std::array<u8, Maxwell::NumRenderTargets> color_formats;

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -680,6 +680,15 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
         .lineStippleFactor = 0,
         .lineStipplePattern = 0,
     };
+    VkPipelineRasterizationConservativeStateCreateInfoEXT conservative_raster{
+        .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_CONSERVATIVE_STATE_CREATE_INFO_EXT,
+        .pNext = nullptr,
+        .flags = 0,
+        .conservativeRasterizationMode = key.state.conservative_raster_enable != 0
+                                             ? VK_CONSERVATIVE_RASTERIZATION_MODE_OVERESTIMATE_EXT
+                                             : VK_CONSERVATIVE_RASTERIZATION_MODE_DISABLED_EXT,
+        .extraPrimitiveOverestimationSize = 0.0f,
+    };
     VkPipelineRasterizationProvokingVertexStateCreateInfoEXT provoking_vertex{
         .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_PROVOKING_VERTEX_STATE_CREATE_INFO_EXT,
         .pNext = nullptr,
@@ -690,6 +699,9 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
     if (IsLine(input_assembly_topology) && device.IsExtLineRasterizationSupported()) {
         line_state.pNext = std::exchange(rasterization_ci.pNext, &line_state);
     }
+    if (device.IsExtConservativeRasterizationSupported()) {
+        conservative_raster.pNext = std::exchange(rasterization_ci.pNext, &conservative_raster);
+    }
     if (device.IsExtProvokingVertexSupported()) {
         provoking_vertex.pNext = std::exchange(rasterization_ci.pNext, &provoking_vertex);
     }