summaryrefslogtreecommitdiffstats
path: root/src
diff options
context:
space:
mode:
authorReinUsesLisp <reinuseslisp@airmail.cc>2020-02-01 06:57:01 +0100
committerReinUsesLisp <reinuseslisp@airmail.cc>2020-02-02 01:19:44 +0100
commit017474c3f89e5afcb3ffdb9fac3e412c40986092 (patch)
tree4164fe82d9d5ee8abb9ac3ae7d7875d163195c9b /src
parentMerge pull request #3350 from ReinUsesLisp/atom (diff)
downloadyuzu-017474c3f89e5afcb3ffdb9fac3e412c40986092.tar
yuzu-017474c3f89e5afcb3ffdb9fac3e412c40986092.tar.gz
yuzu-017474c3f89e5afcb3ffdb9fac3e412c40986092.tar.bz2
yuzu-017474c3f89e5afcb3ffdb9fac3e412c40986092.tar.lz
yuzu-017474c3f89e5afcb3ffdb9fac3e412c40986092.tar.xz
yuzu-017474c3f89e5afcb3ffdb9fac3e412c40986092.tar.zst
yuzu-017474c3f89e5afcb3ffdb9fac3e412c40986092.zip
Diffstat (limited to '')
-rw-r--r--src/video_core/engines/shader_bytecode.h20
-rw-r--r--src/video_core/shader/decode/shift.cpp79
2 files changed, 89 insertions, 10 deletions
diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h
index f443ec0fe..901adb97b 100644
--- a/src/video_core/engines/shader_bytecode.h
+++ b/src/video_core/engines/shader_bytecode.h
@@ -624,6 +624,19 @@ enum class ShuffleOperation : u64 {
Bfly = 3, // shuffleXorNV
};
+enum class ShfType : u64 {
+ Bits32 = 0,
+ U64 = 2,
+ S64 = 3,
+};
+
+enum class ShfXmode : u64 {
+ None = 0,
+ HI = 1,
+ X = 2,
+ XHI = 3,
+};
+
union Instruction {
constexpr Instruction& operator=(const Instruction& instr) {
value = instr.value;
@@ -776,6 +789,13 @@ union Instruction {
} shr;
union {
+ BitField<37, 2, ShfType> type;
+ BitField<48, 2, ShfXmode> xmode;
+ BitField<50, 1, u64> wrap;
+ BitField<20, 6, u64> immediate;
+ } shf;
+
+ 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 d419e9c45..98e71b5f2 100644
--- a/src/video_core/shader/decode/shift.cpp
+++ b/src/video_core/shader/decode/shift.cpp
@@ -10,8 +10,30 @@
namespace VideoCommon::Shader {
+using std::move;
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
+using Tegra::Shader::ShfType;
+using Tegra::Shader::ShfXmode;
+
+namespace {
+
+Node Shift(OperationCode opcode, Node value, Node shift) {
+ Node is_full = Operation(OperationCode::LogicalIEqual, shift, Immediate(32));
+ Node shifted = Operation(opcode, move(value), move(shift));
+ return Operation(OperationCode::Select, move(is_full), Immediate(0), move(shifted));
+}
+
+Node ClampShift(Node shift, s32 size = 32) {
+ shift = Operation(OperationCode::IMax, move(shift), Immediate(0));
+ return Operation(OperationCode::IMin, move(shift), Immediate(size));
+}
+
+Node WrapShift(Node shift, s32 size = 32) {
+ return Operation(OperationCode::UBitwiseAnd, move(shift), Immediate(size - 1));
+}
+
+} // Anonymous namespace
u32 ShaderIR::DecodeShift(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
@@ -32,25 +54,62 @@ u32 ShaderIR::DecodeShift(NodeBlock& bb, u32 pc) {
case OpCode::Id::SHR_C:
case OpCode::Id::SHR_R:
case OpCode::Id::SHR_IMM: {
- 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));
- }
+ op_b = instr.shr.wrap ? WrapShift(move(op_b)) : ClampShift(move(op_b));
Node value = SignedOperation(OperationCode::IArithmeticShiftRight, instr.shift.is_signed,
- std::move(op_a), std::move(op_b));
+ move(op_a), move(op_b));
SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
- SetRegister(bb, instr.gpr0, std::move(value));
+ SetRegister(bb, instr.gpr0, move(value));
break;
}
case OpCode::Id::SHL_C:
case OpCode::Id::SHL_R:
case OpCode::Id::SHL_IMM: {
- const Node value = Operation(OperationCode::ILogicalShiftLeft, op_a, op_b);
+ Node value = Operation(OperationCode::ILogicalShiftLeft, op_a, op_b);
SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
- SetRegister(bb, instr.gpr0, value);
+ SetRegister(bb, instr.gpr0, move(value));
+ break;
+ }
+ case OpCode::Id::SHF_LEFT_R:
+ case OpCode::Id::SHF_LEFT_IMM: {
+ UNIMPLEMENTED_IF(instr.generates_cc);
+ UNIMPLEMENTED_IF_MSG(instr.shf.xmode != ShfXmode::None, "xmode={}",
+ static_cast<int>(instr.shf.xmode.Value()));
+
+ if (instr.is_b_imm) {
+ op_b = Immediate(static_cast<u32>(instr.shf.immediate));
+ }
+ const s32 size = instr.shf.type == ShfType::Bits32 ? 32 : 64;
+ Node shift = instr.shf.wrap ? WrapShift(move(op_b), size) : ClampShift(move(op_b), size);
+
+ Node negated_shift = Operation(OperationCode::INegate, shift);
+ Node low_shift = Operation(OperationCode::IAdd, move(negated_shift), Immediate(32));
+
+ Node low = move(op_a);
+ Node high = GetRegister(instr.gpr39);
+ Node value;
+ if (instr.shf.type == ShfType::Bits32) {
+ high = Shift(OperationCode::ILogicalShiftLeft, move(high), move(shift));
+ low = Shift(OperationCode::ILogicalShiftRight, move(op_a), move(low_shift));
+ value = Operation(OperationCode::IBitwiseOr, move(high), move(low));
+ } else {
+ // These values are used when the shift value is less than 32
+ Node less_low = Operation(OperationCode::ILogicalShiftRight, low, low_shift);
+ Node less_high = Operation(OperationCode::ILogicalShiftLeft, high, shift);
+ Node less = Operation(OperationCode::IBitwiseOr, move(less_low), move(less_high));
+
+ // And these when it's larger than or 32
+ Node reduced = Operation(OperationCode::IAdd, shift, Immediate(-32));
+ Node greater = Shift(OperationCode::ILogicalShiftLeft, move(low), move(reduced));
+
+ Node is_less = Operation(OperationCode::LogicalILessThan, shift, Immediate(32));
+ Node is_zero = Operation(OperationCode::LogicalIEqual, move(shift), Immediate(0));
+
+ value = Operation(OperationCode::Select, move(is_less), move(less), move(greater));
+ value = Operation(OperationCode::Select, move(is_zero), move(high), move(value));
+ }
+
+ SetRegister(bb, instr.gpr0, move(value));
break;
}
default: