summaryrefslogtreecommitdiffstats
path: root/src/video_core
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core')
-rw-r--r--src/video_core/engines/maxwell_3d.cpp5
-rw-r--r--src/video_core/engines/maxwell_3d.h16
-rw-r--r--src/video_core/engines/maxwell_dma.cpp37
-rw-r--r--src/video_core/renderer_opengl/gl_rasterizer.cpp2
-rw-r--r--src/video_core/renderer_opengl/gl_shader_cache.cpp3
-rw-r--r--src/video_core/renderer_vulkan/fixed_pipeline_state.cpp1
-rw-r--r--src/video_core/renderer_vulkan/fixed_pipeline_state.h1
-rw-r--r--src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp12
-rw-r--r--src/video_core/renderer_vulkan/vk_pipeline_cache.cpp1
9 files changed, 55 insertions, 23 deletions
diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp
index 4a2f2c1fd..d502d181c 100644
--- a/src/video_core/engines/maxwell_3d.cpp
+++ b/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();
}
}
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index 910ab213a..34b085388 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/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);
diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp
index 27be4af87..1bf6ca2dd 100644
--- a/src/video_core/engines/maxwell_dma.cpp
+++ b/src/video_core/engines/maxwell_dma.cpp
@@ -102,26 +102,29 @@ void MaxwellDMA::Launch() {
const bool is_src_pitch = IsPitchKind(static_cast<PTEKind>(src_kind));
const bool is_dst_pitch = IsPitchKind(static_cast<PTEKind>(dst_kind));
if (!is_src_pitch && is_dst_pitch) {
- std::vector<u8> tmp_buffer(regs.line_length_in);
- std::vector<u8> dst_buffer(regs.line_length_in);
- memory_manager.ReadBlockUnsafe(regs.offset_in, tmp_buffer.data(),
- regs.line_length_in);
- for (u32 offset = 0; offset < regs.line_length_in; ++offset) {
- dst_buffer[offset] =
- tmp_buffer[convert_linear_2_blocklinear_addr(regs.offset_in + offset) -
- regs.offset_in];
+ UNIMPLEMENTED_IF(regs.line_length_in % 16 != 0);
+ UNIMPLEMENTED_IF(regs.offset_in % 16 != 0);
+ UNIMPLEMENTED_IF(regs.offset_out % 16 != 0);
+ std::vector<u8> tmp_buffer(16);
+ for (u32 offset = 0; offset < regs.line_length_in; offset += 16) {
+ memory_manager.ReadBlockUnsafe(
+ convert_linear_2_blocklinear_addr(regs.offset_in + offset),
+ tmp_buffer.data(), tmp_buffer.size());
+ memory_manager.WriteBlock(regs.offset_out + offset, tmp_buffer.data(),
+ tmp_buffer.size());
}
- memory_manager.WriteBlock(regs.offset_out, dst_buffer.data(), regs.line_length_in);
} else if (is_src_pitch && !is_dst_pitch) {
- std::vector<u8> tmp_buffer(regs.line_length_in);
- std::vector<u8> dst_buffer(regs.line_length_in);
- memory_manager.ReadBlockUnsafe(regs.offset_in, tmp_buffer.data(),
- regs.line_length_in);
- for (u32 offset = 0; offset < regs.line_length_in; ++offset) {
- dst_buffer[convert_linear_2_blocklinear_addr(regs.offset_out + offset) -
- regs.offset_out] = tmp_buffer[offset];
+ UNIMPLEMENTED_IF(regs.line_length_in % 16 != 0);
+ UNIMPLEMENTED_IF(regs.offset_in % 16 != 0);
+ UNIMPLEMENTED_IF(regs.offset_out % 16 != 0);
+ std::vector<u8> tmp_buffer(16);
+ for (u32 offset = 0; offset < regs.line_length_in; offset += 16) {
+ memory_manager.ReadBlockUnsafe(regs.offset_in + offset, tmp_buffer.data(),
+ tmp_buffer.size());
+ memory_manager.WriteBlock(
+ convert_linear_2_blocklinear_addr(regs.offset_out + offset),
+ tmp_buffer.data(), tmp_buffer.size());
}
- memory_manager.WriteBlock(regs.offset_out, dst_buffer.data(), regs.line_length_in);
} else {
if (!accelerate.BufferCopy(regs.offset_in, regs.offset_out, regs.line_length_in)) {
std::vector<u8> tmp_buffer(regs.line_length_in);
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index 8a8b5ce54..d05a5f60b 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/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));
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp
index 4221c2774..3fe04a115 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp
@@ -76,7 +76,8 @@ Shader::RuntimeInfo MakeRuntimeInfo(const GraphicsPipelineKey& key,
}
break;
case Shader::Stage::TessellationEval:
- info.tess_clockwise = key.tessellation_clockwise != 0;
+ // Flip the face, as OpenGL's drawing is flipped.
+ info.tess_clockwise = key.tessellation_clockwise == 0;
info.tess_primitive = [&key] {
switch (key.tessellation_primitive) {
case Maxwell::Tessellation::DomainType::Isolines:
diff --git a/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp b/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp
index f85ed8e5b..98cc26679 100644
--- a/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp
+++ b/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) {
diff --git a/src/video_core/renderer_vulkan/fixed_pipeline_state.h b/src/video_core/renderer_vulkan/fixed_pipeline_state.h
index 43441209c..1afdef329 100644
--- a/src/video_core/renderer_vulkan/fixed_pipeline_state.h
+++ b/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;
diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
index 1aa116cea..ef75c126c 100644
--- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
+++ b/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);
}
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
index e216b90d9..d4b0a542a 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
@@ -166,6 +166,7 @@ Shader::RuntimeInfo MakeRuntimeInfo(std::span<const Shader::IR::Program> program
}
break;
case Shader::Stage::TessellationEval:
+ info.tess_clockwise = key.state.tessellation_clockwise != 0;
info.tess_primitive = [&key] {
const u32 raw{key.state.tessellation_primitive.Value()};
switch (static_cast<Maxwell::Tessellation::DomainType>(raw)) {
generated by cgit v1.2.3 (git 2.25.1) at 2024-11-22 03:15:31 +0100