summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--src/video_core/engines/maxwell_3d.cpp6
-rw-r--r--src/video_core/engines/maxwell_3d.h4
-rw-r--r--src/video_core/engines/maxwell_compute.cpp6
-rw-r--r--src/video_core/engines/maxwell_dma.cpp17
-rw-r--r--src/video_core/engines/shader_bytecode.h10
-rw-r--r--src/video_core/renderer_opengl/gl_rasterizer.cpp25
-rw-r--r--src/video_core/renderer_opengl/gl_rasterizer.h6
-rw-r--r--src/video_core/renderer_opengl/gl_shader_decompiler.cpp79
-rw-r--r--src/video_core/renderer_opengl/gl_shader_gen.cpp29
-rw-r--r--src/video_core/renderer_opengl/gl_shader_manager.cpp11
-rw-r--r--src/video_core/renderer_opengl/gl_shader_manager.h8
11 files changed, 140 insertions, 61 deletions
diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp
index 8afd26fe9..bca014a4a 100644
--- a/src/video_core/engines/maxwell_3d.cpp
+++ b/src/video_core/engines/maxwell_3d.cpp
@@ -13,8 +13,7 @@
#include "video_core/renderer_base.h"
#include "video_core/textures/texture.h"
-namespace Tegra {
-namespace Engines {
+namespace Tegra::Engines {
/// First register id that is actually a Macro call.
constexpr u32 MacroRegistersStart = 0xE00;
@@ -408,5 +407,4 @@ void Maxwell3D::ProcessClearBuffers() {
rasterizer.Clear();
}
-} // namespace Engines
-} // namespace Tegra
+} // namespace Tegra::Engines
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index c8af1c6b6..0e09a7ee5 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -643,8 +643,10 @@ public:
u32 d3d_cull_mode;
ComparisonOp depth_test_func;
+ float alpha_test_ref;
+ ComparisonOp alpha_test_func;
- INSERT_PADDING_WORDS(0xB);
+ INSERT_PADDING_WORDS(0x9);
struct {
u32 separate_alpha;
diff --git a/src/video_core/engines/maxwell_compute.cpp b/src/video_core/engines/maxwell_compute.cpp
index 59e28b22d..8b5f08351 100644
--- a/src/video_core/engines/maxwell_compute.cpp
+++ b/src/video_core/engines/maxwell_compute.cpp
@@ -6,8 +6,7 @@
#include "core/core.h"
#include "video_core/engines/maxwell_compute.h"
-namespace Tegra {
-namespace Engines {
+namespace Tegra::Engines {
void MaxwellCompute::WriteReg(u32 method, u32 value) {
ASSERT_MSG(method < Regs::NUM_REGS,
@@ -26,5 +25,4 @@ void MaxwellCompute::WriteReg(u32 method, u32 value) {
}
}
-} // namespace Engines
-} // namespace Tegra
+} // namespace Tegra::Engines
diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp
index 103cd110e..b8a78cf82 100644
--- a/src/video_core/engines/maxwell_dma.cpp
+++ b/src/video_core/engines/maxwell_dma.cpp
@@ -7,8 +7,7 @@
#include "video_core/rasterizer_interface.h"
#include "video_core/textures/decoders.h"
-namespace Tegra {
-namespace Engines {
+namespace Tegra::Engines {
MaxwellDMA::MaxwellDMA(VideoCore::RasterizerInterface& rasterizer, MemoryManager& memory_manager)
: memory_manager(memory_manager), rasterizer{rasterizer} {}
@@ -78,9 +77,9 @@ void MaxwellDMA::HandleCopy() {
ASSERT(regs.exec.enable_2d == 1);
- std::size_t copy_size = regs.x_count * regs.y_count;
+ const std::size_t copy_size = regs.x_count * regs.y_count;
- const auto FlushAndInvalidate = [&](u32 src_size, u32 dst_size) {
+ const auto FlushAndInvalidate = [&](u32 src_size, u64 dst_size) {
// TODO(Subv): For now, manually flush the regions until we implement GPU-accelerated
// copying.
rasterizer.FlushRegion(source_cpu, src_size);
@@ -91,14 +90,11 @@ void MaxwellDMA::HandleCopy() {
rasterizer.InvalidateRegion(dest_cpu, dst_size);
};
- u8* src_buffer = Memory::GetPointer(source_cpu);
- u8* dst_buffer = Memory::GetPointer(dest_cpu);
-
if (regs.exec.is_dst_linear && !regs.exec.is_src_linear) {
ASSERT(regs.src_params.size_z == 1);
// If the input is tiled and the output is linear, deswizzle the input and copy it over.
- u32 src_bytes_per_pixel = regs.src_pitch / regs.src_params.size_x;
+ const u32 src_bytes_per_pixel = regs.src_pitch / regs.src_params.size_x;
FlushAndInvalidate(regs.src_pitch * regs.src_params.size_y,
copy_size * src_bytes_per_pixel);
@@ -111,7 +107,7 @@ void MaxwellDMA::HandleCopy() {
ASSERT(regs.dst_params.size_z == 1);
ASSERT(regs.src_pitch == regs.x_count);
- u32 src_bpp = regs.src_pitch / regs.x_count;
+ const u32 src_bpp = regs.src_pitch / regs.x_count;
FlushAndInvalidate(regs.src_pitch * regs.y_count,
regs.dst_params.size_x * regs.dst_params.size_y * src_bpp);
@@ -122,5 +118,4 @@ void MaxwellDMA::HandleCopy() {
}
}
-} // namespace Engines
-} // namespace Tegra
+} // namespace Tegra::Engines
diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h
index 67a0770dd..6cd08d28b 100644
--- a/src/video_core/engines/shader_bytecode.h
+++ b/src/video_core/engines/shader_bytecode.h
@@ -1223,11 +1223,13 @@ public:
KIL,
SSY,
SYNC,
+ BRK,
DEPBAR,
BFE_C,
BFE_R,
BFE_IMM,
BRA,
+ PBK,
LD_A,
LD_C,
ST_A,
@@ -1385,7 +1387,7 @@ public:
/// conditionally executed).
static bool IsPredicatedInstruction(Id opcode) {
// TODO(Subv): Add the rest of unpredicated instructions.
- return opcode != Id::SSY;
+ return opcode != Id::SSY && opcode != Id::PBK;
}
class Matcher {
@@ -1481,9 +1483,11 @@ private:
#define INST(bitstring, op, type, name) Detail::GetMatcher(bitstring, op, type, name)
INST("111000110011----", Id::KIL, Type::Flow, "KIL"),
INST("111000101001----", Id::SSY, Type::Flow, "SSY"),
+ INST("111000101010----", Id::PBK, Type::Flow, "PBK"),
INST("111000100100----", Id::BRA, Type::Flow, "BRA"),
+ INST("1111000011111---", Id::SYNC, Type::Flow, "SYNC"),
+ INST("111000110100---", Id::BRK, Type::Flow, "BRK"),
INST("1111000011110---", Id::DEPBAR, Type::Synch, "DEPBAR"),
- INST("1111000011111---", Id::SYNC, Type::Synch, "SYNC"),
INST("1110111111011---", Id::LD_A, Type::Memory, "LD_A"),
INST("1110111110010---", Id::LD_C, Type::Memory, "LD_C"),
INST("1110111111110---", Id::ST_A, Type::Memory, "ST_A"),
@@ -1622,4 +1626,4 @@ private:
}
};
-} // namespace Tegra::Shader
+} // namespace Tegra::Shader \ No newline at end of file
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index 3daccf82f..be51c5215 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -570,10 +570,11 @@ void RasterizerOpenGL::DrawArrays() {
SyncBlendState();
SyncLogicOpState();
SyncCullMode();
- SyncAlphaTest();
SyncScissorTest();
+ // Alpha Testing is synced on shaders.
SyncTransformFeedback();
SyncPointState();
+ CheckAlphaTests();
// TODO(bunnei): Sync framebuffer_scale uniform here
// TODO(bunnei): Sync scissorbox uniform(s) here
@@ -1007,17 +1008,6 @@ void RasterizerOpenGL::SyncLogicOpState() {
state.logic_op.operation = MaxwellToGL::LogicOp(regs.logic_op.operation);
}
-void RasterizerOpenGL::SyncAlphaTest() {
- const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
-
- // TODO(Rodrigo): Alpha testing is a legacy OpenGL feature, but it can be
- // implemented with a test+discard in fragment shaders.
- if (regs.alpha_test_enabled != 0) {
- LOG_CRITICAL(Render_OpenGL, "Alpha testing is not implemented");
- UNREACHABLE();
- }
-}
-
void RasterizerOpenGL::SyncScissorTest() {
const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
@@ -1052,4 +1042,15 @@ void RasterizerOpenGL::SyncPointState() {
state.point.size = regs.point_size == 0 ? 1 : regs.point_size;
}
+void RasterizerOpenGL::CheckAlphaTests() {
+ const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+
+ if (regs.alpha_test_enabled != 0 && regs.rt_control.count > 1) {
+ LOG_CRITICAL(
+ Render_OpenGL,
+ "Alpha Testing is enabled with Multiple Render Targets, this behavior is undefined.");
+ UNREACHABLE();
+ }
+}
+
} // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h
index b1f7ccc7e..0e90a31f5 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer.h
@@ -162,9 +162,6 @@ private:
/// Syncs the LogicOp state to match the guest state
void SyncLogicOpState();
- /// Syncs the alpha test state to match the guest state
- void SyncAlphaTest();
-
/// Syncs the scissor test state to match the guest state
void SyncScissorTest();
@@ -174,6 +171,9 @@ private:
/// Syncs the point state to match the guest state
void SyncPointState();
+ /// Check asserts for alpha testing.
+ void CheckAlphaTests();
+
bool has_ARB_direct_state_access = false;
bool has_ARB_multi_bind = false;
bool has_ARB_separate_shader_objects = false;
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index 42a072ed9..fe4d1bd83 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -163,10 +163,11 @@ private:
const ExitMethod jmp = Scan(target, end, labels);
return exit_method = ParallelExit(no_jmp, jmp);
}
- case OpCode::Id::SSY: {
- // The SSY instruction uses a similar encoding as the BRA instruction.
+ case OpCode::Id::SSY:
+ case OpCode::Id::PBK: {
+ // The SSY and PBK use a similar encoding as the BRA instruction.
ASSERT_MSG(instr.bra.constant_buffer == 0,
- "Constant buffer SSY is not supported");
+ "Constant buffer branching is not supported");
const u32 target = offset + instr.bra.GetBranchTarget();
labels.insert(target);
// Continue scanning for an exit method.
@@ -1233,27 +1234,27 @@ private:
}
/*
- * Emits code to push the input target address to the SSY address stack, incrementing the stack
+ * Emits code to push the input target address to the flow address stack, incrementing the stack
* top.
*/
- void EmitPushToSSYStack(u32 target) {
+ void EmitPushToFlowStack(u32 target) {
shader.AddLine('{');
++shader.scope;
- shader.AddLine("ssy_stack[ssy_stack_top] = " + std::to_string(target) + "u;");
- shader.AddLine("ssy_stack_top++;");
+ shader.AddLine("flow_stack[flow_stack_top] = " + std::to_string(target) + "u;");
+ shader.AddLine("flow_stack_top++;");
--shader.scope;
shader.AddLine('}');
}
/*
- * Emits code to pop an address from the SSY address stack, setting the jump address to the
+ * Emits code to pop an address from the flow address stack, setting the jump address to the
* popped address and decrementing the stack top.
*/
- void EmitPopFromSSYStack() {
+ void EmitPopFromFlowStack() {
shader.AddLine('{');
++shader.scope;
- shader.AddLine("ssy_stack_top--;");
- shader.AddLine("jmp_to = ssy_stack[ssy_stack_top];");
+ shader.AddLine("flow_stack_top--;");
+ shader.AddLine("jmp_to = flow_stack[flow_stack_top];");
shader.AddLine("break;");
--shader.scope;
shader.AddLine('}');
@@ -1265,9 +1266,29 @@ private:
ASSERT_MSG(header.ps.omap.sample_mask == 0, "Samplemask write is unimplemented");
+ shader.AddLine("if (alpha_test[0] != 0) {");
+ ++shader.scope;
+ // We start on the register containing the alpha value in the first RT.
+ u32 current_reg = 3;
+ for (u32 render_target = 0; render_target < Maxwell3D::Regs::NumRenderTargets;
+ ++render_target) {
+ // TODO(Blinkhawk): verify the behavior of alpha testing on hardware when
+ // multiple render targets are used.
+ if (header.ps.IsColorComponentOutputEnabled(render_target, 0) ||
+ header.ps.IsColorComponentOutputEnabled(render_target, 1) ||
+ header.ps.IsColorComponentOutputEnabled(render_target, 2) ||
+ header.ps.IsColorComponentOutputEnabled(render_target, 3)) {
+ shader.AddLine(fmt::format("if (!AlphaFunc({})) discard;",
+ regs.GetRegisterAsFloat(current_reg)));
+ current_reg += 4;
+ }
+ }
+ --shader.scope;
+ shader.AddLine('}');
+
// Write the color outputs using the data in the shader registers, disabled
// rendertargets/components are skipped in the register assignment.
- u32 current_reg = 0;
+ current_reg = 0;
for (u32 render_target = 0; render_target < Maxwell3D::Regs::NumRenderTargets;
++render_target) {
// TODO(Subv): Figure out how dual-source blending is configured in the Switch.
@@ -3322,16 +3343,32 @@ private:
// The SSY opcode tells the GPU where to re-converge divergent execution paths, it
// sets the target of the jump that the SYNC instruction will make. The SSY opcode
// has a similar structure to the BRA opcode.
- ASSERT_MSG(instr.bra.constant_buffer == 0, "Constant buffer SSY is not supported");
+ ASSERT_MSG(instr.bra.constant_buffer == 0, "Constant buffer flow is not supported");
const u32 target = offset + instr.bra.GetBranchTarget();
- EmitPushToSSYStack(target);
+ EmitPushToFlowStack(target);
+ break;
+ }
+ case OpCode::Id::PBK: {
+ // PBK pushes to a stack the address where BRK will jump to. This shares stack with
+ // SSY but using SYNC on a PBK address will kill the shader execution. We don't
+ // emulate this because it's very unlikely a driver will emit such invalid shader.
+ ASSERT_MSG(instr.bra.constant_buffer == 0, "Constant buffer PBK is not supported");
+
+ const u32 target = offset + instr.bra.GetBranchTarget();
+ EmitPushToFlowStack(target);
break;
}
case OpCode::Id::SYNC: {
// The SYNC opcode jumps to the address previously set by the SSY opcode
ASSERT(instr.flow.cond == Tegra::Shader::FlowCondition::Always);
- EmitPopFromSSYStack();
+ EmitPopFromFlowStack();
+ break;
+ }
+ case OpCode::Id::BRK: {
+ // The BRK opcode jumps to the address previously set by the PBK opcode
+ ASSERT(instr.flow.cond == Tegra::Shader::FlowCondition::Always);
+ EmitPopFromFlowStack();
break;
}
case OpCode::Id::DEPBAR: {
@@ -3449,11 +3486,11 @@ private:
labels.insert(subroutine.begin);
shader.AddLine("uint jmp_to = " + std::to_string(subroutine.begin) + "u;");
- // TODO(Subv): Figure out the actual depth of the SSY stack, for now it seems
- // unlikely that shaders will use 20 nested SSYs.
- constexpr u32 SSY_STACK_SIZE = 20;
- shader.AddLine("uint ssy_stack[" + std::to_string(SSY_STACK_SIZE) + "];");
- shader.AddLine("uint ssy_stack_top = 0u;");
+ // TODO(Subv): Figure out the actual depth of the flow stack, for now it seems
+ // unlikely that shaders will use 20 nested SSYs and PBKs.
+ constexpr u32 FLOW_STACK_SIZE = 20;
+ shader.AddLine("uint flow_stack[" + std::to_string(FLOW_STACK_SIZE) + "];");
+ shader.AddLine("uint flow_stack_top = 0u;");
shader.AddLine("while (true) {");
++shader.scope;
@@ -3520,7 +3557,7 @@ private:
// Declarations
std::set<std::string> declr_predicates;
-}; // namespace Decompiler
+}; // namespace OpenGL::GLShader::Decompiler
std::string GetCommonDeclarations() {
return fmt::format("#define MAX_CONSTBUFFER_ELEMENTS {}\n",
diff --git a/src/video_core/renderer_opengl/gl_shader_gen.cpp b/src/video_core/renderer_opengl/gl_shader_gen.cpp
index ecbc9d8ed..e883ffb1d 100644
--- a/src/video_core/renderer_opengl/gl_shader_gen.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_gen.cpp
@@ -29,6 +29,7 @@ layout(std140) uniform vs_config {
vec4 viewport_flip;
uvec4 instance_id;
uvec4 flip_stage;
+ uvec4 alpha_test;
};
)";
@@ -105,6 +106,7 @@ layout (std140) uniform gs_config {
vec4 viewport_flip;
uvec4 instance_id;
uvec4 flip_stage;
+ uvec4 alpha_test;
};
void main() {
@@ -142,8 +144,33 @@ layout (std140) uniform fs_config {
vec4 viewport_flip;
uvec4 instance_id;
uvec4 flip_stage;
+ uvec4 alpha_test;
};
+bool AlphaFunc(in float value) {
+ float ref = uintBitsToFloat(alpha_test[2]);
+ switch (alpha_test[1]) {
+ case 1:
+ return false;
+ case 2:
+ return value < ref;
+ case 3:
+ return value == ref;
+ case 4:
+ return value <= ref;
+ case 5:
+ return value > ref;
+ case 6:
+ return value != ref;
+ case 7:
+ return value >= ref;
+ case 8:
+ return true;
+ default:
+ return false;
+ }
+}
+
void main() {
exec_fragment();
}
@@ -152,4 +179,4 @@ void main() {
out += program.first;
return {out, program.second};
}
-} // namespace OpenGL::GLShader \ No newline at end of file
+} // namespace OpenGL::GLShader
diff --git a/src/video_core/renderer_opengl/gl_shader_manager.cpp b/src/video_core/renderer_opengl/gl_shader_manager.cpp
index 010857ec6..8b8869ecb 100644
--- a/src/video_core/renderer_opengl/gl_shader_manager.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_manager.cpp
@@ -16,6 +16,17 @@ void MaxwellUniformData::SetFromRegs(const Maxwell3D::State::ShaderStageInfo& sh
viewport_flip[0] = regs.viewport_transform[0].scale_x < 0.0 ? -1.0f : 1.0f;
viewport_flip[1] = regs.viewport_transform[0].scale_y < 0.0 ? -1.0f : 1.0f;
+ u32 func = static_cast<u32>(regs.alpha_test_func);
+ // Normalize the gl variants of opCompare to be the same as the normal variants
+ u32 op_gl_variant_base = static_cast<u32>(Tegra::Engines::Maxwell3D::Regs::ComparisonOp::Never);
+ if (func >= op_gl_variant_base) {
+ func = func - op_gl_variant_base + 1U;
+ }
+
+ alpha_test.enabled = regs.alpha_test_enabled;
+ alpha_test.func = func;
+ alpha_test.ref = regs.alpha_test_ref;
+
// We only assign the instance to the first component of the vector, the rest is just padding.
instance_id[0] = state.current_instance;
diff --git a/src/video_core/renderer_opengl/gl_shader_manager.h b/src/video_core/renderer_opengl/gl_shader_manager.h
index b3a191cf2..36fe1f04c 100644
--- a/src/video_core/renderer_opengl/gl_shader_manager.h
+++ b/src/video_core/renderer_opengl/gl_shader_manager.h
@@ -22,8 +22,14 @@ struct MaxwellUniformData {
alignas(16) GLvec4 viewport_flip;
alignas(16) GLuvec4 instance_id;
alignas(16) GLuvec4 flip_stage;
+ struct alignas(16) {
+ GLuint enabled;
+ GLuint func;
+ GLfloat ref;
+ GLuint padding;
+ } alpha_test;
};
-static_assert(sizeof(MaxwellUniformData) == 48, "MaxwellUniformData structure size is incorrect");
+static_assert(sizeof(MaxwellUniformData) == 64, "MaxwellUniformData structure size is incorrect");
static_assert(sizeof(MaxwellUniformData) < 16384,
"MaxwellUniformData structure must be less than 16kb as per the OpenGL spec");
generated by cgit v1.2.3 (git 2.25.1) at 2024-10-15 06:13:18 +0200