summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorbunnei <bunneidev@gmail.com>2018-09-15 21:27:12 +0200
committerGitHub <noreply@github.com>2018-09-15 21:27:12 +0200
commitba480ea2fbb71bb8daa184a9ab82c3aee2e24681 (patch)
treeff8bea04a8de8757f0e287a7ecbef14ab0dbbfc3
parentMerge pull request #1271 from Subv/kepler_engine (diff)
parentShaders: Implemented multiple-word loads and stores to and from attribute memory. (diff)
downloadyuzu-ba480ea2fbb71bb8daa184a9ab82c3aee2e24681.tar
yuzu-ba480ea2fbb71bb8daa184a9ab82c3aee2e24681.tar.gz
yuzu-ba480ea2fbb71bb8daa184a9ab82c3aee2e24681.tar.bz2
yuzu-ba480ea2fbb71bb8daa184a9ab82c3aee2e24681.tar.lz
yuzu-ba480ea2fbb71bb8daa184a9ab82c3aee2e24681.tar.xz
yuzu-ba480ea2fbb71bb8daa184a9ab82c3aee2e24681.tar.zst
yuzu-ba480ea2fbb71bb8daa184a9ab82c3aee2e24681.zip
Diffstat (limited to '')
-rw-r--r--src/video_core/engines/shader_bytecode.h10
-rw-r--r--src/video_core/renderer_opengl/gl_shader_decompiler.cpp55
2 files changed, 58 insertions, 7 deletions
diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h
index 58f2904ce..d6e2397f2 100644
--- a/src/video_core/engines/shader_bytecode.h
+++ b/src/video_core/engines/shader_bytecode.h
@@ -67,6 +67,13 @@ private:
u64 value{};
};
+enum class AttributeSize : u64 {
+ Word = 0,
+ DoubleWord = 1,
+ TripleWord = 2,
+ QuadWord = 3,
+};
+
union Attribute {
Attribute() = default;
@@ -87,9 +94,10 @@ union Attribute {
};
union {
+ BitField<20, 10, u64> immediate;
BitField<22, 2, u64> element;
BitField<24, 6, Index> index;
- BitField<47, 3, u64> size;
+ BitField<47, 3, AttributeSize> size;
} fmt20;
union {
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index 2d56370c7..81c0662d0 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -1772,13 +1772,34 @@ private:
case OpCode::Type::Memory: {
switch (opcode->GetId()) {
case OpCode::Id::LD_A: {
- ASSERT_MSG(instr.attribute.fmt20.size == 0, "untested");
// Note: Shouldn't this be interp mode flat? As in no interpolation made.
+ ASSERT_MSG(instr.gpr8.Value() == Register::ZeroIndex,
+ "Indirect attribute loads are not supported");
+ ASSERT_MSG((instr.attribute.fmt20.immediate.Value() % sizeof(u32)) == 0,
+ "Unaligned attribute loads are not supported");
Tegra::Shader::IpaMode input_mode{Tegra::Shader::IpaInterpMode::Perspective,
Tegra::Shader::IpaSampleMode::Default};
- regs.SetRegisterToInputAttibute(instr.gpr0, instr.attribute.fmt20.element,
- instr.attribute.fmt20.index, input_mode);
+
+ u32 next_element = instr.attribute.fmt20.element;
+ u32 next_index = static_cast<u32>(instr.attribute.fmt20.index.Value());
+
+ const auto LoadNextElement = [&](u32 reg_offset) {
+ regs.SetRegisterToInputAttibute(instr.gpr0.Value() + reg_offset, next_element,
+ static_cast<Attribute::Index>(next_index),
+ input_mode);
+
+ // Load the next attribute element into the following register. If the element
+ // to load goes beyond the vec4 size, load the first element of the next
+ // attribute.
+ next_element = (next_element + 1) % 4;
+ next_index = next_index + (next_element == 0 ? 1 : 0);
+ };
+
+ const u32 num_words = static_cast<u32>(instr.attribute.fmt20.size.Value()) + 1;
+ for (u32 reg_offset = 0; reg_offset < num_words; ++reg_offset) {
+ LoadNextElement(reg_offset);
+ }
break;
}
case OpCode::Id::LD_C: {
@@ -1820,9 +1841,31 @@ private:
break;
}
case OpCode::Id::ST_A: {
- ASSERT_MSG(instr.attribute.fmt20.size == 0, "untested");
- regs.SetOutputAttributeToRegister(instr.attribute.fmt20.index,
- instr.attribute.fmt20.element, instr.gpr0);
+ ASSERT_MSG(instr.gpr8.Value() == Register::ZeroIndex,
+ "Indirect attribute loads are not supported");
+ ASSERT_MSG((instr.attribute.fmt20.immediate.Value() % sizeof(u32)) == 0,
+ "Unaligned attribute loads are not supported");
+
+ u32 next_element = instr.attribute.fmt20.element;
+ u32 next_index = static_cast<u32>(instr.attribute.fmt20.index.Value());
+
+ const auto StoreNextElement = [&](u32 reg_offset) {
+ regs.SetOutputAttributeToRegister(static_cast<Attribute::Index>(next_index),
+ next_element,
+ instr.gpr0.Value() + reg_offset);
+
+ // Load the next attribute element into the following register. If the element
+ // to load goes beyond the vec4 size, load the first element of the next
+ // attribute.
+ next_element = (next_element + 1) % 4;
+ next_index = next_index + (next_element == 0 ? 1 : 0);
+ };
+
+ const u32 num_words = static_cast<u32>(instr.attribute.fmt20.size.Value()) + 1;
+ for (u32 reg_offset = 0; reg_offset < num_words; ++reg_offset) {
+ StoreNextElement(reg_offset);
+ }
+
break;
}
case OpCode::Id::TEX: {