summaryrefslogtreecommitdiffstats
path: root/src/video_core/engines
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core/engines')
-rw-r--r--src/video_core/engines/fermi_2d.cpp4
-rw-r--r--src/video_core/engines/fermi_2d.h32
-rw-r--r--src/video_core/engines/kepler_compute.cpp5
-rw-r--r--src/video_core/engines/kepler_compute.h18
-rw-r--r--src/video_core/engines/kepler_memory.cpp1
-rw-r--r--src/video_core/engines/kepler_memory.h4
-rw-r--r--src/video_core/engines/maxwell_3d.cpp89
-rw-r--r--src/video_core/engines/maxwell_3d.h166
-rw-r--r--src/video_core/engines/maxwell_dma.cpp3
-rw-r--r--src/video_core/engines/shader_header.h38
10 files changed, 174 insertions, 186 deletions
diff --git a/src/video_core/engines/fermi_2d.cpp b/src/video_core/engines/fermi_2d.cpp
index a01d334ad..0f640fdae 100644
--- a/src/video_core/engines/fermi_2d.cpp
+++ b/src/video_core/engines/fermi_2d.cpp
@@ -18,8 +18,8 @@ Fermi2D::Fermi2D() {
Fermi2D::~Fermi2D() = default;
-void Fermi2D::BindRasterizer(VideoCore::RasterizerInterface& rasterizer_) {
- rasterizer = &rasterizer_;
+void Fermi2D::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) {
+ rasterizer = rasterizer_;
}
void Fermi2D::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
diff --git a/src/video_core/engines/fermi_2d.h b/src/video_core/engines/fermi_2d.h
index 81522988e..c808a577d 100644
--- a/src/video_core/engines/fermi_2d.h
+++ b/src/video_core/engines/fermi_2d.h
@@ -38,7 +38,7 @@ public:
~Fermi2D();
/// Binds a rasterizer to this engine.
- void BindRasterizer(VideoCore::RasterizerInterface& rasterizer);
+ void BindRasterizer(VideoCore::RasterizerInterface* rasterizer);
/// Write the value to the register identified by method.
void CallMethod(u32 method, u32 method_argument, bool is_last_call) override;
@@ -171,30 +171,30 @@ public:
static constexpr std::size_t NUM_REGS = 0x258;
struct {
u32 object;
- INSERT_UNION_PADDING_WORDS(0x3F);
+ INSERT_PADDING_WORDS_NOINIT(0x3F);
u32 no_operation;
NotifyType notify;
- INSERT_UNION_PADDING_WORDS(0x2);
+ INSERT_PADDING_WORDS_NOINIT(0x2);
u32 wait_for_idle;
- INSERT_UNION_PADDING_WORDS(0xB);
+ INSERT_PADDING_WORDS_NOINIT(0xB);
u32 pm_trigger;
- INSERT_UNION_PADDING_WORDS(0xF);
+ INSERT_PADDING_WORDS_NOINIT(0xF);
u32 context_dma_notify;
u32 dst_context_dma;
u32 src_context_dma;
u32 semaphore_context_dma;
- INSERT_UNION_PADDING_WORDS(0x1C);
+ INSERT_PADDING_WORDS_NOINIT(0x1C);
Surface dst;
CpuIndexWrap pixels_from_cpu_index_wrap;
u32 kind2d_check_enable;
Surface src;
SectorPromotion pixels_from_memory_sector_promotion;
- INSERT_UNION_PADDING_WORDS(0x1);
+ INSERT_PADDING_WORDS_NOINIT(0x1);
NumTpcs num_tpcs;
u32 render_enable_addr_upper;
u32 render_enable_addr_lower;
RenderEnableMode render_enable_mode;
- INSERT_UNION_PADDING_WORDS(0x4);
+ INSERT_PADDING_WORDS_NOINIT(0x4);
u32 clip_x0;
u32 clip_y0;
u32 clip_width;
@@ -212,7 +212,7 @@ public:
BitField<8, 6, u32> y;
} pattern_offset;
BitField<0, 2, PatternSelect> pattern_select;
- INSERT_UNION_PADDING_WORDS(0xC);
+ INSERT_PADDING_WORDS_NOINIT(0xC);
struct {
BitField<0, 3, MonochromePatternColorFormat> color_format;
BitField<0, 1, MonochromePatternFormat> format;
@@ -227,15 +227,15 @@ public:
std::array<u32, 0x20> X1R5G5B5;
std::array<u32, 0x10> Y8;
} color_pattern;
- INSERT_UNION_PADDING_WORDS(0x10);
+ INSERT_PADDING_WORDS_NOINIT(0x10);
struct {
u32 prim_mode;
u32 prim_color_format;
u32 prim_color;
u32 line_tie_break_bits;
- INSERT_UNION_PADDING_WORDS(0x14);
+ INSERT_PADDING_WORDS_NOINIT(0x14);
u32 prim_point_xy;
- INSERT_UNION_PADDING_WORDS(0x7);
+ INSERT_PADDING_WORDS_NOINIT(0x7);
std::array<Point, 0x40> prim_point;
} render_solid;
struct {
@@ -247,7 +247,7 @@ public:
u32 color0;
u32 color1;
u32 mono_opacity;
- INSERT_UNION_PADDING_WORDS(0x6);
+ INSERT_PADDING_WORDS_NOINIT(0x6);
u32 src_width;
u32 src_height;
u32 dx_du_frac;
@@ -260,9 +260,9 @@ public:
u32 dst_y0_int;
u32 data;
} pixels_from_cpu;
- INSERT_UNION_PADDING_WORDS(0x3);
+ INSERT_PADDING_WORDS_NOINIT(0x3);
u32 big_endian_control;
- INSERT_UNION_PADDING_WORDS(0x3);
+ INSERT_PADDING_WORDS_NOINIT(0x3);
struct {
BitField<0, 3, u32> block_shape;
BitField<0, 5, u32> corral_size;
@@ -271,7 +271,7 @@ public:
BitField<0, 1, Origin> origin;
BitField<4, 1, Filter> filter;
} sample_mode;
- INSERT_UNION_PADDING_WORDS(0x8);
+ INSERT_PADDING_WORDS_NOINIT(0x8);
s32 dst_x0;
s32 dst_y0;
s32 dst_width;
diff --git a/src/video_core/engines/kepler_compute.cpp b/src/video_core/engines/kepler_compute.cpp
index ba387506e..a9b75091e 100644
--- a/src/video_core/engines/kepler_compute.cpp
+++ b/src/video_core/engines/kepler_compute.cpp
@@ -21,8 +21,8 @@ KeplerCompute::KeplerCompute(Core::System& system_, MemoryManager& memory_manage
KeplerCompute::~KeplerCompute() = default;
-void KeplerCompute::BindRasterizer(VideoCore::RasterizerInterface& rasterizer_) {
- rasterizer = &rasterizer_;
+void KeplerCompute::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) {
+ rasterizer = rasterizer_;
}
void KeplerCompute::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
@@ -39,7 +39,6 @@ void KeplerCompute::CallMethod(u32 method, u32 method_argument, bool is_last_cal
case KEPLER_COMPUTE_REG_INDEX(data_upload): {
upload_state.ProcessData(method_argument, is_last_call);
if (is_last_call) {
- system.GPU().Maxwell3D().OnMemoryWrite();
}
break;
}
diff --git a/src/video_core/engines/kepler_compute.h b/src/video_core/engines/kepler_compute.h
index 51a041202..7c40cba38 100644
--- a/src/video_core/engines/kepler_compute.h
+++ b/src/video_core/engines/kepler_compute.h
@@ -46,7 +46,7 @@ public:
~KeplerCompute();
/// Binds a rasterizer to this engine.
- void BindRasterizer(VideoCore::RasterizerInterface& rasterizer);
+ void BindRasterizer(VideoCore::RasterizerInterface* rasterizer);
static constexpr std::size_t NumConstBuffers = 8;
@@ -55,7 +55,7 @@ public:
union {
struct {
- INSERT_UNION_PADDING_WORDS(0x60);
+ INSERT_PADDING_WORDS_NOINIT(0x60);
Upload::Registers upload;
@@ -67,7 +67,7 @@ public:
u32 data_upload;
- INSERT_UNION_PADDING_WORDS(0x3F);
+ INSERT_PADDING_WORDS_NOINIT(0x3F);
struct {
u32 address;
@@ -76,11 +76,11 @@ public:
}
} launch_desc_loc;
- INSERT_UNION_PADDING_WORDS(0x1);
+ INSERT_PADDING_WORDS_NOINIT(0x1);
u32 launch;
- INSERT_UNION_PADDING_WORDS(0x4A7);
+ INSERT_PADDING_WORDS_NOINIT(0x4A7);
struct {
u32 address_high;
@@ -92,7 +92,7 @@ public:
}
} tsc;
- INSERT_UNION_PADDING_WORDS(0x3);
+ INSERT_PADDING_WORDS_NOINIT(0x3);
struct {
u32 address_high;
@@ -104,7 +104,7 @@ public:
}
} tic;
- INSERT_UNION_PADDING_WORDS(0x22);
+ INSERT_PADDING_WORDS_NOINIT(0x22);
struct {
u32 address_high;
@@ -115,11 +115,11 @@ public:
}
} code_loc;
- INSERT_UNION_PADDING_WORDS(0x3FE);
+ INSERT_PADDING_WORDS_NOINIT(0x3FE);
u32 tex_cb_index;
- INSERT_UNION_PADDING_WORDS(0x374);
+ INSERT_PADDING_WORDS_NOINIT(0x374);
};
std::array<u32, NUM_REGS> reg_array;
};
diff --git a/src/video_core/engines/kepler_memory.cpp b/src/video_core/engines/kepler_memory.cpp
index 9911140e9..560551157 100644
--- a/src/video_core/engines/kepler_memory.cpp
+++ b/src/video_core/engines/kepler_memory.cpp
@@ -33,7 +33,6 @@ void KeplerMemory::CallMethod(u32 method, u32 method_argument, bool is_last_call
case KEPLERMEMORY_REG_INDEX(data): {
upload_state.ProcessData(method_argument, is_last_call);
if (is_last_call) {
- system.GPU().Maxwell3D().OnMemoryWrite();
}
break;
}
diff --git a/src/video_core/engines/kepler_memory.h b/src/video_core/engines/kepler_memory.h
index 62483589e..19808a5c6 100644
--- a/src/video_core/engines/kepler_memory.h
+++ b/src/video_core/engines/kepler_memory.h
@@ -50,7 +50,7 @@ public:
union {
struct {
- INSERT_UNION_PADDING_WORDS(0x60);
+ INSERT_PADDING_WORDS_NOINIT(0x60);
Upload::Registers upload;
@@ -62,7 +62,7 @@ public:
u32 data;
- INSERT_UNION_PADDING_WORDS(0x11);
+ INSERT_PADDING_WORDS_NOINIT(0x11);
};
std::array<u32, NUM_REGS> reg_array;
};
diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp
index 9be651e24..75517a4f7 100644
--- a/src/video_core/engines/maxwell_3d.cpp
+++ b/src/video_core/engines/maxwell_3d.cpp
@@ -30,8 +30,8 @@ Maxwell3D::Maxwell3D(Core::System& system_, MemoryManager& memory_manager_)
Maxwell3D::~Maxwell3D() = default;
-void Maxwell3D::BindRasterizer(VideoCore::RasterizerInterface& rasterizer_) {
- rasterizer = &rasterizer_;
+void Maxwell3D::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) {
+ rasterizer = rasterizer_;
}
void Maxwell3D::InitializeRegisterDefaults() {
@@ -179,22 +179,22 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume
return ProcessMacroBind(argument);
case MAXWELL3D_REG_INDEX(firmware[4]):
return ProcessFirmwareCall4();
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[1]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[2]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[3]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[4]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[5]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[6]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[7]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[8]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[9]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[10]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[11]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[12]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[13]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[14]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[15]):
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data):
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 1:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 2:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 3:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 4:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 5:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 6:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 7:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 8:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 9:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 10:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 11:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 12:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 13:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 14:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 15:
return StartCBData(method);
case MAXWELL3D_REG_INDEX(cb_bind[0]):
return ProcessCBBind(0);
@@ -223,7 +223,6 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume
case MAXWELL3D_REG_INDEX(data_upload):
upload_state.ProcessData(argument, is_last_call);
if (is_last_call) {
- OnMemoryWrite();
}
return;
case MAXWELL3D_REG_INDEX(fragment_barrier):
@@ -287,22 +286,22 @@ void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
return;
}
switch (method) {
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[1]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[2]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[3]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[4]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[5]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[6]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[7]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[8]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[9]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[10]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[11]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[12]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[13]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[14]):
- case MAXWELL3D_REG_INDEX(const_buffer.cb_data[15]):
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data):
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 1:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 2:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 3:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 4:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 5:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 6:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 7:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 8:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 9:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 10:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 11:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 12:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 13:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 14:
+ case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 15:
ProcessCBMultiData(method, base_start, amount);
break;
default:
@@ -570,17 +569,18 @@ std::optional<u64> Maxwell3D::GetQueryResult() {
}
}
-void Maxwell3D::ProcessCBBind(std::size_t stage_index) {
+void Maxwell3D::ProcessCBBind(size_t stage_index) {
// Bind the buffer currently in CB_ADDRESS to the specified index in the desired shader stage.
- auto& shader = state.shader_stages[stage_index];
- auto& bind_data = regs.cb_bind[stage_index];
-
- ASSERT(bind_data.index < Regs::MaxConstBuffers);
- auto& buffer = shader.const_buffers[bind_data.index];
-
+ const auto& bind_data = regs.cb_bind[stage_index];
+ auto& buffer = state.shader_stages[stage_index].const_buffers[bind_data.index];
buffer.enabled = bind_data.valid.Value() != 0;
buffer.address = regs.const_buffer.BufferAddress();
buffer.size = regs.const_buffer.cb_size;
+
+ const bool is_enabled = bind_data.valid.Value() != 0;
+ const GPUVAddr gpu_addr = is_enabled ? regs.const_buffer.BufferAddress() : 0;
+ const u32 size = is_enabled ? regs.const_buffer.cb_size : 0;
+ rasterizer->BindGraphicsUniformBuffer(stage_index, bind_data.index, gpu_addr, size);
}
void Maxwell3D::ProcessCBData(u32 value) {
@@ -592,7 +592,7 @@ void Maxwell3D::ProcessCBData(u32 value) {
}
void Maxwell3D::StartCBData(u32 method) {
- constexpr u32 first_cb_data = MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]);
+ constexpr u32 first_cb_data = MAXWELL3D_REG_INDEX(const_buffer.cb_data);
cb_data_state.start_pos = regs.const_buffer.cb_pos;
cb_data_state.id = method - first_cb_data;
cb_data_state.current = method;
@@ -605,7 +605,7 @@ void Maxwell3D::ProcessCBMultiData(u32 method, const u32* start_base, u32 amount
if (cb_data_state.current != null_cb_data) {
FinishCBData();
}
- constexpr u32 first_cb_data = MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]);
+ constexpr u32 first_cb_data = MAXWELL3D_REG_INDEX(const_buffer.cb_data);
cb_data_state.start_pos = regs.const_buffer.cb_pos;
cb_data_state.id = method - first_cb_data;
cb_data_state.current = method;
@@ -635,7 +635,6 @@ void Maxwell3D::FinishCBData() {
const u32 id = cb_data_state.id;
memory_manager.WriteBlock(address, cb_data_state.buffer[id].data(), size);
- OnMemoryWrite();
cb_data_state.id = null_cb_data;
cb_data_state.current = null_cb_data;
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index bf9e07c9b..ffed42a29 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -55,7 +55,7 @@ public:
~Maxwell3D();
/// Binds a rasterizer to this engine.
- void BindRasterizer(VideoCore::RasterizerInterface& rasterizer);
+ void BindRasterizer(VideoCore::RasterizerInterface* rasterizer);
/// Register structure of the Maxwell3D engine.
/// TODO(Subv): This structure will need to be made bigger as more registers are discovered.
@@ -536,7 +536,7 @@ public:
Equation equation_a;
Factor factor_source_a;
Factor factor_dest_a;
- INSERT_UNION_PADDING_WORDS(1);
+ INSERT_PADDING_WORDS_NOINIT(1);
};
enum class TessellationPrimitive : u32 {
@@ -608,7 +608,7 @@ public:
};
u32 layer_stride;
u32 base_layer;
- INSERT_UNION_PADDING_WORDS(7);
+ INSERT_PADDING_WORDS_NOINIT(7);
GPUVAddr Address() const {
return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) |
@@ -640,7 +640,7 @@ public:
BitField<8, 3, ViewportSwizzle> z;
BitField<12, 3, ViewportSwizzle> w;
} swizzle;
- INSERT_UNION_PADDING_WORDS(1);
+ INSERT_PADDING_WORDS_NOINIT(1);
Common::Rectangle<f32> GetRect() const {
return {
@@ -700,7 +700,7 @@ public:
u32 address_low;
s32 buffer_size;
s32 buffer_offset;
- INSERT_UNION_PADDING_WORDS(3);
+ INSERT_PADDING_WORDS_NOINIT(3);
GPUVAddr Address() const {
return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) |
@@ -713,7 +713,7 @@ public:
u32 stream;
u32 varying_count;
u32 stride;
- INSERT_UNION_PADDING_WORDS(1);
+ INSERT_PADDING_WORDS_NOINIT(1);
};
static_assert(sizeof(TransformFeedbackLayout) == 16);
@@ -731,7 +731,7 @@ public:
union {
struct {
- INSERT_UNION_PADDING_WORDS(0x44);
+ INSERT_PADDING_WORDS_NOINIT(0x44);
u32 wait_for_idle;
@@ -744,7 +744,7 @@ public:
ShadowRamControl shadow_ram_control;
- INSERT_UNION_PADDING_WORDS(0x16);
+ INSERT_PADDING_WORDS_NOINIT(0x16);
Upload::Registers upload;
struct {
@@ -755,11 +755,11 @@ public:
u32 data_upload;
- INSERT_UNION_PADDING_WORDS(0x16);
+ INSERT_PADDING_WORDS_NOINIT(0x16);
u32 force_early_fragment_tests;
- INSERT_UNION_PADDING_WORDS(0x2D);
+ INSERT_PADDING_WORDS_NOINIT(0x2D);
struct {
union {
@@ -769,7 +769,7 @@ public:
};
} sync_info;
- INSERT_UNION_PADDING_WORDS(0x15);
+ INSERT_PADDING_WORDS_NOINIT(0x15);
union {
BitField<0, 2, TessellationPrimitive> prim;
@@ -781,21 +781,21 @@ public:
std::array<f32, 4> tess_level_outer;
std::array<f32, 2> tess_level_inner;
- INSERT_UNION_PADDING_WORDS(0x10);
+ INSERT_PADDING_WORDS_NOINIT(0x10);
u32 rasterize_enable;
std::array<TransformFeedbackBinding, NumTransformFeedbackBuffers> tfb_bindings;
- INSERT_UNION_PADDING_WORDS(0xC0);
+ INSERT_PADDING_WORDS_NOINIT(0xC0);
std::array<TransformFeedbackLayout, NumTransformFeedbackBuffers> tfb_layouts;
- INSERT_UNION_PADDING_WORDS(0x1);
+ INSERT_PADDING_WORDS_NOINIT(0x1);
u32 tfb_enabled;
- INSERT_UNION_PADDING_WORDS(0x2E);
+ INSERT_PADDING_WORDS_NOINIT(0x2E);
std::array<RenderTargetConfig, NumRenderTargets> rt;
@@ -803,7 +803,7 @@ public:
std::array<ViewPort, NumViewports> viewports;
- INSERT_UNION_PADDING_WORDS(0x1D);
+ INSERT_PADDING_WORDS_NOINIT(0x1D);
struct {
u32 first;
@@ -815,16 +815,16 @@ public:
float clear_color[4];
float clear_depth;
- INSERT_UNION_PADDING_WORDS(0x3);
+ INSERT_PADDING_WORDS_NOINIT(0x3);
s32 clear_stencil;
- INSERT_UNION_PADDING_WORDS(0x2);
+ INSERT_PADDING_WORDS_NOINIT(0x2);
PolygonMode polygon_mode_front;
PolygonMode polygon_mode_back;
- INSERT_UNION_PADDING_WORDS(0x3);
+ INSERT_PADDING_WORDS_NOINIT(0x3);
u32 polygon_offset_point_enable;
u32 polygon_offset_line_enable;
@@ -832,47 +832,47 @@ public:
u32 patch_vertices;
- INSERT_UNION_PADDING_WORDS(0x4);
+ INSERT_PADDING_WORDS_NOINIT(0x4);
u32 fragment_barrier;
- INSERT_UNION_PADDING_WORDS(0x7);
+ INSERT_PADDING_WORDS_NOINIT(0x7);
std::array<ScissorTest, NumViewports> scissor_test;
- INSERT_UNION_PADDING_WORDS(0x15);
+ INSERT_PADDING_WORDS_NOINIT(0x15);
s32 stencil_back_func_ref;
u32 stencil_back_mask;
u32 stencil_back_func_mask;
- INSERT_UNION_PADDING_WORDS(0x5);
+ INSERT_PADDING_WORDS_NOINIT(0x5);
u32 invalidate_texture_data_cache;
- INSERT_UNION_PADDING_WORDS(0x1);
+ INSERT_PADDING_WORDS_NOINIT(0x1);
u32 tiled_cache_barrier;
- INSERT_UNION_PADDING_WORDS(0x4);
+ INSERT_PADDING_WORDS_NOINIT(0x4);
u32 color_mask_common;
- INSERT_UNION_PADDING_WORDS(0x2);
+ INSERT_PADDING_WORDS_NOINIT(0x2);
f32 depth_bounds[2];
- INSERT_UNION_PADDING_WORDS(0x2);
+ INSERT_PADDING_WORDS_NOINIT(0x2);
u32 rt_separate_frag_data;
- INSERT_UNION_PADDING_WORDS(0x1);
+ INSERT_PADDING_WORDS_NOINIT(0x1);
u32 multisample_raster_enable;
u32 multisample_raster_samples;
std::array<u32, 4> multisample_sample_mask;
- INSERT_UNION_PADDING_WORDS(0x5);
+ INSERT_PADDING_WORDS_NOINIT(0x5);
struct {
u32 address_high;
@@ -898,7 +898,7 @@ public:
};
} render_area;
- INSERT_UNION_PADDING_WORDS(0x3F);
+ INSERT_PADDING_WORDS_NOINIT(0x3F);
union {
BitField<0, 4, u32> stencil;
@@ -907,24 +907,24 @@ public:
BitField<12, 4, u32> viewport;
} clear_flags;
- INSERT_UNION_PADDING_WORDS(0x10);
+ INSERT_PADDING_WORDS_NOINIT(0x10);
u32 fill_rectangle;
- INSERT_UNION_PADDING_WORDS(0x8);
+ INSERT_PADDING_WORDS_NOINIT(0x8);
std::array<VertexAttribute, NumVertexAttributes> vertex_attrib_format;
std::array<MsaaSampleLocation, 4> multisample_sample_locations;
- INSERT_UNION_PADDING_WORDS(0x2);
+ INSERT_PADDING_WORDS_NOINIT(0x2);
union {
BitField<0, 1, u32> enable;
BitField<4, 3, u32> target;
} multisample_coverage_to_color;
- INSERT_UNION_PADDING_WORDS(0x8);
+ INSERT_PADDING_WORDS_NOINIT(0x8);
struct {
union {
@@ -947,7 +947,7 @@ public:
}
} rt_control;
- INSERT_UNION_PADDING_WORDS(0x2);
+ INSERT_PADDING_WORDS_NOINIT(0x2);
u32 zeta_width;
u32 zeta_height;
@@ -958,11 +958,11 @@ public:
SamplerIndex sampler_index;
- INSERT_UNION_PADDING_WORDS(0x25);
+ INSERT_PADDING_WORDS_NOINIT(0x25);
u32 depth_test_enable;
- INSERT_UNION_PADDING_WORDS(0x5);
+ INSERT_PADDING_WORDS_NOINIT(0x5);
u32 independent_blend_enable;
@@ -970,7 +970,7 @@ public:
u32 alpha_test_enabled;
- INSERT_UNION_PADDING_WORDS(0x6);
+ INSERT_PADDING_WORDS_NOINIT(0x6);
u32 d3d_cull_mode;
@@ -985,7 +985,7 @@ public:
float a;
} blend_color;
- INSERT_UNION_PADDING_WORDS(0x4);
+ INSERT_PADDING_WORDS_NOINIT(0x4);
struct {
u32 separate_alpha;
@@ -994,7 +994,7 @@ public:
Blend::Factor factor_dest_rgb;
Blend::Equation equation_a;
Blend::Factor factor_source_a;
- INSERT_UNION_PADDING_WORDS(1);
+ INSERT_PADDING_WORDS_NOINIT(1);
Blend::Factor factor_dest_a;
u32 enable_common;
@@ -1010,7 +1010,7 @@ public:
u32 stencil_front_func_mask;
u32 stencil_front_mask;
- INSERT_UNION_PADDING_WORDS(0x2);
+ INSERT_PADDING_WORDS_NOINIT(0x2);
u32 frag_color_clamp;
@@ -1022,17 +1022,17 @@ public:
float line_width_smooth;
float line_width_aliased;
- INSERT_UNION_PADDING_WORDS(0x1B);
+ INSERT_PADDING_WORDS_NOINIT(0x1B);
u32 invalidate_sampler_cache_no_wfi;
u32 invalidate_texture_header_cache_no_wfi;
- INSERT_UNION_PADDING_WORDS(0x2);
+ INSERT_PADDING_WORDS_NOINIT(0x2);
u32 vb_element_base;
u32 vb_base_instance;
- INSERT_UNION_PADDING_WORDS(0x35);
+ INSERT_PADDING_WORDS_NOINIT(0x35);
u32 clip_distance_enabled;
@@ -1040,11 +1040,11 @@ public:
float point_size;
- INSERT_UNION_PADDING_WORDS(0x1);
+ INSERT_PADDING_WORDS_NOINIT(0x1);
u32 point_sprite_enable;
- INSERT_UNION_PADDING_WORDS(0x3);
+ INSERT_PADDING_WORDS_NOINIT(0x3);
CounterReset counter_reset;
@@ -1057,7 +1057,7 @@ public:
BitField<4, 1, u32> alpha_to_one;
} multisample_control;
- INSERT_UNION_PADDING_WORDS(0x4);
+ INSERT_PADDING_WORDS_NOINIT(0x4);
struct {
u32 address_high;
@@ -1081,7 +1081,7 @@ public:
}
} tsc;
- INSERT_UNION_PADDING_WORDS(0x1);
+ INSERT_PADDING_WORDS_NOINIT(0x1);
float polygon_offset_factor;
@@ -1098,7 +1098,7 @@ public:
}
} tic;
- INSERT_UNION_PADDING_WORDS(0x5);
+ INSERT_PADDING_WORDS_NOINIT(0x5);
u32 stencil_two_side_enable;
StencilOp stencil_back_op_fail;
@@ -1106,17 +1106,17 @@ public:
StencilOp stencil_back_op_zpass;
ComparisonOp stencil_back_func_func;
- INSERT_UNION_PADDING_WORDS(0x4);
+ INSERT_PADDING_WORDS_NOINIT(0x4);
u32 framebuffer_srgb;
float polygon_offset_units;
- INSERT_UNION_PADDING_WORDS(0x4);
+ INSERT_PADDING_WORDS_NOINIT(0x4);
Tegra::Texture::MsaaMode multisample_mode;
- INSERT_UNION_PADDING_WORDS(0xC);
+ INSERT_PADDING_WORDS_NOINIT(0xC);
union {
BitField<2, 1, u32> coord_origin;
@@ -1132,7 +1132,7 @@ public:
(static_cast<GPUVAddr>(code_address_high) << 32) | code_address_low);
}
} code_address;
- INSERT_UNION_PADDING_WORDS(1);
+ INSERT_PADDING_WORDS_NOINIT(1);
struct {
u32 vertex_end_gl;
@@ -1144,14 +1144,14 @@ public:
};
} draw;
- INSERT_UNION_PADDING_WORDS(0xA);
+ INSERT_PADDING_WORDS_NOINIT(0xA);
struct {
u32 enabled;
u32 index;
} primitive_restart;
- INSERT_UNION_PADDING_WORDS(0x5F);
+ INSERT_PADDING_WORDS_NOINIT(0x5F);
struct {
u32 start_addr_high;
@@ -1192,9 +1192,9 @@ public:
}
} index_array;
- INSERT_UNION_PADDING_WORDS(0x7);
+ INSERT_PADDING_WORDS_NOINIT(0x7);
- INSERT_UNION_PADDING_WORDS(0x1F);
+ INSERT_PADDING_WORDS_NOINIT(0x1F);
float polygon_offset_clamp;
@@ -1208,14 +1208,14 @@ public:
}
} instanced_arrays;
- INSERT_UNION_PADDING_WORDS(0x4);
+ INSERT_PADDING_WORDS_NOINIT(0x4);
union {
BitField<0, 1, u32> enable;
BitField<4, 8, u32> unk4;
} vp_point_size;
- INSERT_UNION_PADDING_WORDS(1);
+ INSERT_PADDING_WORDS_NOINIT(1);
u32 cull_test_enabled;
FrontFace front_face;
@@ -1223,11 +1223,11 @@ public:
u32 pixel_center_integer;
- INSERT_UNION_PADDING_WORDS(0x1);
+ INSERT_PADDING_WORDS_NOINIT(0x1);
u32 viewport_transform_enabled;
- INSERT_UNION_PADDING_WORDS(0x3);
+ INSERT_PADDING_WORDS_NOINIT(0x3);
union {
BitField<0, 1, u32> depth_range_0_1;
@@ -1236,18 +1236,18 @@ public:
BitField<11, 1, u32> depth_clamp_disabled;
} view_volume_clip_control;
- INSERT_UNION_PADDING_WORDS(0x1F);
+ INSERT_PADDING_WORDS_NOINIT(0x1F);
u32 depth_bounds_enable;
- INSERT_UNION_PADDING_WORDS(1);
+ INSERT_PADDING_WORDS_NOINIT(1);
struct {
u32 enable;
LogicOperation operation;
} logic_op;
- INSERT_UNION_PADDING_WORDS(0x1);
+ INSERT_PADDING_WORDS_NOINIT(0x1);
union {
u32 raw;
@@ -1260,9 +1260,9 @@ public:
BitField<6, 4, u32> RT;
BitField<10, 11, u32> layer;
} clear_buffers;
- INSERT_UNION_PADDING_WORDS(0xB);
+ INSERT_PADDING_WORDS_NOINIT(0xB);
std::array<ColorMask, NumRenderTargets> color_mask;
- INSERT_UNION_PADDING_WORDS(0x38);
+ INSERT_PADDING_WORDS_NOINIT(0x38);
struct {
u32 query_address_high;
@@ -1284,7 +1284,7 @@ public:
}
} query;
- INSERT_UNION_PADDING_WORDS(0x3C);
+ INSERT_PADDING_WORDS_NOINIT(0x3C);
struct {
union {
@@ -1314,8 +1314,7 @@ public:
GPUVAddr LimitAddress() const {
return static_cast<GPUVAddr>((static_cast<GPUVAddr>(limit_high) << 32) |
- limit_low) +
- 1;
+ limit_low);
}
} vertex_array_limit[NumVertexArrays];
@@ -1325,10 +1324,10 @@ public:
BitField<4, 4, ShaderProgram> program;
};
u32 offset;
- INSERT_UNION_PADDING_WORDS(14);
+ INSERT_PADDING_WORDS_NOINIT(14);
} shader_config[MaxShaderProgram];
- INSERT_UNION_PADDING_WORDS(0x60);
+ INSERT_PADDING_WORDS_NOINIT(0x60);
u32 firmware[0x20];
@@ -1337,7 +1336,7 @@ public:
u32 cb_address_high;
u32 cb_address_low;
u32 cb_pos;
- u32 cb_data[NumCBData];
+ std::array<u32, NumCBData> cb_data;
GPUVAddr BufferAddress() const {
return static_cast<GPUVAddr>(
@@ -1345,7 +1344,7 @@ public:
}
} const_buffer;
- INSERT_UNION_PADDING_WORDS(0x10);
+ INSERT_PADDING_WORDS_NOINIT(0x10);
struct {
union {
@@ -1353,18 +1352,18 @@ public:
BitField<0, 1, u32> valid;
BitField<4, 5, u32> index;
};
- INSERT_UNION_PADDING_WORDS(7);
+ INSERT_PADDING_WORDS_NOINIT(7);
} cb_bind[MaxShaderStage];
- INSERT_UNION_PADDING_WORDS(0x56);
+ INSERT_PADDING_WORDS_NOINIT(0x56);
u32 tex_cb_index;
- INSERT_UNION_PADDING_WORDS(0x7D);
+ INSERT_PADDING_WORDS_NOINIT(0x7D);
std::array<std::array<u8, 128>, NumTransformFeedbackBuffers> tfb_varying_locs;
- INSERT_UNION_PADDING_WORDS(0x298);
+ INSERT_PADDING_WORDS_NOINIT(0x298);
struct {
/// Compressed address of a buffer that holds information about bound SSBOs.
@@ -1376,14 +1375,14 @@ public:
}
} ssbo_info;
- INSERT_UNION_PADDING_WORDS(0x11);
+ INSERT_PADDING_WORDS_NOINIT(0x11);
struct {
u32 address[MaxShaderStage];
u32 size[MaxShaderStage];
} tex_info_buffers;
- INSERT_UNION_PADDING_WORDS(0xCC);
+ INSERT_PADDING_WORDS_NOINIT(0xCC);
};
std::array<u32, NUM_REGS> reg_array;
};
@@ -1403,6 +1402,7 @@ public:
};
std::array<ShaderStageInfo, Regs::MaxShaderStage> shader_stages;
+
u32 current_instance = 0; ///< Current instance to be used to simulate instanced rendering.
};
@@ -1452,11 +1452,6 @@ public:
return *rasterizer;
}
- /// Notify a memory write has happened.
- void OnMemoryWrite() {
- dirty.flags |= dirty.on_write_stores;
- }
-
enum class MMEDrawMode : u32 {
Undefined,
Array,
@@ -1478,7 +1473,6 @@ public:
using Tables = std::array<Table, 2>;
Flags flags;
- Flags on_write_stores;
Tables tables{};
} dirty;
@@ -1541,7 +1535,7 @@ private:
void FinishCBData();
/// Handles a write to the CB_BIND register.
- void ProcessCBBind(std::size_t stage_index);
+ void ProcessCBBind(size_t stage_index);
/// Handles a write to the VERTEX_END_GL register, triggering a draw.
void DrawArrays();
diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp
index ba750748c..a2f19559f 100644
--- a/src/video_core/engines/maxwell_dma.cpp
+++ b/src/video_core/engines/maxwell_dma.cpp
@@ -60,9 +60,6 @@ void MaxwellDMA::Launch() {
return;
}
- // All copies here update the main memory, so mark all rasterizer states as invalid.
- system.GPU().Maxwell3D().OnMemoryWrite();
-
if (is_src_pitch && is_dst_pitch) {
CopyPitchToPitch();
} else {
diff --git a/src/video_core/engines/shader_header.h b/src/video_core/engines/shader_header.h
index ceec05459..e0d7b89c5 100644
--- a/src/video_core/engines/shader_header.h
+++ b/src/video_core/engines/shader_header.h
@@ -68,10 +68,10 @@ struct Header {
union {
struct {
- INSERT_UNION_PADDING_BYTES(3); // ImapSystemValuesA
- INSERT_UNION_PADDING_BYTES(1); // ImapSystemValuesB
- INSERT_UNION_PADDING_BYTES(16); // ImapGenericVector[32]
- INSERT_UNION_PADDING_BYTES(2); // ImapColor
+ INSERT_PADDING_BYTES_NOINIT(3); // ImapSystemValuesA
+ INSERT_PADDING_BYTES_NOINIT(1); // ImapSystemValuesB
+ INSERT_PADDING_BYTES_NOINIT(16); // ImapGenericVector[32]
+ INSERT_PADDING_BYTES_NOINIT(2); // ImapColor
union {
BitField<0, 8, u16> clip_distances;
BitField<8, 1, u16> point_sprite_s;
@@ -82,20 +82,20 @@ struct Header {
BitField<14, 1, u16> instance_id;
BitField<15, 1, u16> vertex_id;
};
- INSERT_UNION_PADDING_BYTES(5); // ImapFixedFncTexture[10]
- INSERT_UNION_PADDING_BYTES(1); // ImapReserved
- INSERT_UNION_PADDING_BYTES(3); // OmapSystemValuesA
- INSERT_UNION_PADDING_BYTES(1); // OmapSystemValuesB
- INSERT_UNION_PADDING_BYTES(16); // OmapGenericVector[32]
- INSERT_UNION_PADDING_BYTES(2); // OmapColor
- INSERT_UNION_PADDING_BYTES(2); // OmapSystemValuesC
- INSERT_UNION_PADDING_BYTES(5); // OmapFixedFncTexture[10]
- INSERT_UNION_PADDING_BYTES(1); // OmapReserved
+ INSERT_PADDING_BYTES_NOINIT(5); // ImapFixedFncTexture[10]
+ INSERT_PADDING_BYTES_NOINIT(1); // ImapReserved
+ INSERT_PADDING_BYTES_NOINIT(3); // OmapSystemValuesA
+ INSERT_PADDING_BYTES_NOINIT(1); // OmapSystemValuesB
+ INSERT_PADDING_BYTES_NOINIT(16); // OmapGenericVector[32]
+ INSERT_PADDING_BYTES_NOINIT(2); // OmapColor
+ INSERT_PADDING_BYTES_NOINIT(2); // OmapSystemValuesC
+ INSERT_PADDING_BYTES_NOINIT(5); // OmapFixedFncTexture[10]
+ INSERT_PADDING_BYTES_NOINIT(1); // OmapReserved
} vtg;
struct {
- INSERT_UNION_PADDING_BYTES(3); // ImapSystemValuesA
- INSERT_UNION_PADDING_BYTES(1); // ImapSystemValuesB
+ INSERT_PADDING_BYTES_NOINIT(3); // ImapSystemValuesA
+ INSERT_PADDING_BYTES_NOINIT(1); // ImapSystemValuesB
union {
BitField<0, 2, PixelImap> x;
@@ -105,10 +105,10 @@ struct Header {
u8 raw;
} imap_generic_vector[32];
- INSERT_UNION_PADDING_BYTES(2); // ImapColor
- INSERT_UNION_PADDING_BYTES(2); // ImapSystemValuesC
- INSERT_UNION_PADDING_BYTES(10); // ImapFixedFncTexture[10]
- INSERT_UNION_PADDING_BYTES(2); // ImapReserved
+ INSERT_PADDING_BYTES_NOINIT(2); // ImapColor
+ INSERT_PADDING_BYTES_NOINIT(2); // ImapSystemValuesC
+ INSERT_PADDING_BYTES_NOINIT(10); // ImapFixedFncTexture[10]
+ INSERT_PADDING_BYTES_NOINIT(2); // ImapReserved
struct {
u32 target;