summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorbunnei <bunneidev@gmail.com>2019-09-10 17:56:17 +0200
committerGitHub <noreply@github.com>2019-09-10 17:56:17 +0200
commit34b2c60f95c964354f00546385684fc9c8726d76 (patch)
tree365887554f8b0ddf08bf98b00b1b6bad5f36f073
parentMerge pull request #2810 from ReinUsesLisp/mme-opt (diff)
parentshader/shift: Implement SHR wrapped and clamped variants (diff)
downloadyuzu-34b2c60f95c964354f00546385684fc9c8726d76.tar
yuzu-34b2c60f95c964354f00546385684fc9c8726d76.tar.gz
yuzu-34b2c60f95c964354f00546385684fc9c8726d76.tar.bz2
yuzu-34b2c60f95c964354f00546385684fc9c8726d76.tar.lz
yuzu-34b2c60f95c964354f00546385684fc9c8726d76.tar.xz
yuzu-34b2c60f95c964354f00546385684fc9c8726d76.tar.zst
yuzu-34b2c60f95c964354f00546385684fc9c8726d76.zip
-rw-r--r--src/video_core/engines/shader_bytecode.h4
-rw-r--r--src/video_core/shader/decode/shift.cpp19
2 files changed, 17 insertions, 6 deletions
diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h
index c3678b9ea..bd8c1ada0 100644
--- a/src/video_core/engines/shader_bytecode.h
+++ b/src/video_core/engines/shader_bytecode.h
@@ -675,6 +675,10 @@ union Instruction {
} shift;
union {
+ BitField<39, 1, u64> wrap;
+ } shr;
+
+ union {
BitField<39, 5, u64> shift_amount;
BitField<48, 1, u64> negate_b;
BitField<49, 1, u64> negate_a;
diff --git a/src/video_core/shader/decode/shift.cpp b/src/video_core/shader/decode/shift.cpp
index 2ac16eeb0..f6ee68a54 100644
--- a/src/video_core/shader/decode/shift.cpp
+++ b/src/video_core/shader/decode/shift.cpp
@@ -17,8 +17,8 @@ u32 ShaderIR::DecodeShift(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
- const Node op_a = GetRegister(instr.gpr8);
- const Node op_b = [&]() {
+ Node op_a = GetRegister(instr.gpr8);
+ Node op_b = [&]() {
if (instr.is_b_imm) {
return Immediate(instr.alu.GetSignedImm20_20());
} else if (instr.is_b_gpr) {
@@ -32,16 +32,23 @@ u32 ShaderIR::DecodeShift(NodeBlock& bb, u32 pc) {
case OpCode::Id::SHR_C:
case OpCode::Id::SHR_R:
case OpCode::Id::SHR_IMM: {
- const Node value = SignedOperation(OperationCode::IArithmeticShiftRight,
- instr.shift.is_signed, PRECISE, op_a, op_b);
+ if (instr.shr.wrap) {
+ op_b = Operation(OperationCode::UBitwiseAnd, std::move(op_b), Immediate(0x1f));
+ } else {
+ op_b = Operation(OperationCode::IMax, std::move(op_b), Immediate(0));
+ op_b = Operation(OperationCode::IMin, std::move(op_b), Immediate(31));
+ }
+
+ Node value = SignedOperation(OperationCode::IArithmeticShiftRight, instr.shift.is_signed,
+ std::move(op_a), std::move(op_b));
SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
- SetRegister(bb, instr.gpr0, value);
+ SetRegister(bb, instr.gpr0, std::move(value));
break;
}
case OpCode::Id::SHL_C:
case OpCode::Id::SHL_R:
case OpCode::Id::SHL_IMM: {
- const Node value = Operation(OperationCode::ILogicalShiftLeft, PRECISE, op_a, op_b);
+ const Node value = Operation(OperationCode::ILogicalShiftLeft, op_a, op_b);
SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
SetRegister(bb, instr.gpr0, value);
break;