summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorbunnei <bunneidev@gmail.com>2019-09-04 04:35:33 +0200
committerGitHub <noreply@github.com>2019-09-04 04:35:33 +0200
commit81fbc5370d8f09a3a167c96556fb3a4bf211673a (patch)
tree496c370578de2ef494faab0fe8c5b1e4d93593bf
parentMerge pull request #2811 from ReinUsesLisp/fsetp-fix (diff)
parentshader_ir/conversion: Split int and float selector and implement F2F H1 (diff)
downloadyuzu-81fbc5370d8f09a3a167c96556fb3a4bf211673a.tar
yuzu-81fbc5370d8f09a3a167c96556fb3a4bf211673a.tar.gz
yuzu-81fbc5370d8f09a3a167c96556fb3a4bf211673a.tar.bz2
yuzu-81fbc5370d8f09a3a167c96556fb3a4bf211673a.tar.lz
yuzu-81fbc5370d8f09a3a167c96556fb3a4bf211673a.tar.xz
yuzu-81fbc5370d8f09a3a167c96556fb3a4bf211673a.tar.zst
yuzu-81fbc5370d8f09a3a167c96556fb3a4bf211673a.zip
-rw-r--r--src/video_core/engines/shader_bytecode.h8
-rw-r--r--src/video_core/shader/decode/conversion.cpp22
2 files changed, 23 insertions, 7 deletions
diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h
index ba28ff51c..c3678b9ea 100644
--- a/src/video_core/engines/shader_bytecode.h
+++ b/src/video_core/engines/shader_bytecode.h
@@ -1020,7 +1020,6 @@ union Instruction {
} iset;
union {
- BitField<41, 2, u64> selector; // i2i and i2f only
BitField<45, 1, u64> negate_a;
BitField<49, 1, u64> abs_a;
BitField<10, 2, Register::Size> src_size;
@@ -1046,6 +1045,13 @@ union Instruction {
}
} f2f;
+ union {
+ BitField<41, 2, u64> selector;
+ } int_src;
+
+ union {
+ BitField<41, 1, u64> selector;
+ } float_src;
} conversion;
union {
diff --git a/src/video_core/shader/decode/conversion.cpp b/src/video_core/shader/decode/conversion.cpp
index 8973fbefa..32facd6ba 100644
--- a/src/video_core/shader/decode/conversion.cpp
+++ b/src/video_core/shader/decode/conversion.cpp
@@ -14,6 +14,12 @@ using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::Register;
+namespace {
+constexpr OperationCode GetFloatSelector(u64 selector) {
+ return selector == 0 ? OperationCode::FCastHalf0 : OperationCode::FCastHalf1;
+}
+} // Anonymous namespace
+
u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
@@ -22,7 +28,7 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
case OpCode::Id::I2I_R:
case OpCode::Id::I2I_C:
case OpCode::Id::I2I_IMM: {
- UNIMPLEMENTED_IF(instr.conversion.selector);
+ UNIMPLEMENTED_IF(instr.conversion.int_src.selector != 0);
UNIMPLEMENTED_IF(instr.conversion.dst_size != Register::Size::Word);
UNIMPLEMENTED_IF(instr.alu.saturate_d);
@@ -57,8 +63,8 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
case OpCode::Id::I2F_R:
case OpCode::Id::I2F_C:
case OpCode::Id::I2F_IMM: {
+ UNIMPLEMENTED_IF(instr.conversion.int_src.selector != 0);
UNIMPLEMENTED_IF(instr.conversion.dst_size == Register::Size::Long);
- UNIMPLEMENTED_IF(instr.conversion.selector);
UNIMPLEMENTED_IF_MSG(instr.generates_cc,
"Condition codes generation in I2F is not implemented");
@@ -113,8 +119,10 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
}();
if (instr.conversion.src_size == Register::Size::Short) {
- // TODO: figure where extract is sey in the encoding
- value = Operation(OperationCode::FCastHalf0, PRECISE, value);
+ value = Operation(GetFloatSelector(instr.conversion.float_src.selector), NO_PRECISE,
+ std::move(value));
+ } else {
+ ASSERT(instr.conversion.float_src.selector == 0);
}
value = GetOperandAbsNegFloat(value, instr.conversion.abs_a, instr.conversion.negate_a);
@@ -169,8 +177,10 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
}();
if (instr.conversion.src_size == Register::Size::Short) {
- // TODO: figure where extract is sey in the encoding
- value = Operation(OperationCode::FCastHalf0, PRECISE, value);
+ value = Operation(GetFloatSelector(instr.conversion.float_src.selector), NO_PRECISE,
+ std::move(value));
+ } else {
+ ASSERT(instr.conversion.float_src.selector == 0);
}
value = GetOperandAbsNegFloat(value, instr.conversion.abs_a, instr.conversion.negate_a);