raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-27 13:26:52 +01:00
Code Issues Packages Projects Releases Wiki Activity

shader: interpolate f16 parameter halves (V_INTERP_P1LL_F16, V_INTERP_P2_F16) (#1792)

Browse Source 
shader: interpolate packed f16 attributes for v_interp_p1ll_f16 and v_interp_p2_f16

Co-authored-by: nmzik <Nmzik@mail.ru>
This commit is contained in:
Nawid3333
2026-10-22 04:54:10 +02:00
committed by GitHub
parent 8502d8a77d
commit d7c1896647
10 changed files with 181 additions and 75 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/frontend/decode/ShaderDecoder.h
+2
View File
@@ -713,6 +713,8 @@ enum class Opcode {
V_INTERP_P1_F32,
V_INTERP_P2_F32,
V_INTERP_MOV_F32,
V_INTERP_P1LL_F16,
V_INTERP_P2_F16,
S_NOP,
S_WAITCNT,
src/graphics/shader/recompiler/frontend/decode/VectorAluOps.cpp
+26 -7
View File
@@ -391,6 +391,8 @@ constexpr OpcodeMap VOP3_OPCODE_LIST[] = {
{0x377u, Opcode::V_PERMLANE16_B32},
{0x378u, Opcode::V_PERMLANEX16_B32},
{0x34bu, Opcode::V_FMA_F16},
{0x342u, Opcode::V_INTERP_P1LL_F16},
{0x35au, Opcode::V_INTERP_P2_F16},
{0x36du, Opcode::V_ADD3_U32},
{0x14fu, Opcode::V_ALIGNBYTE_B32},
};
@@ -1553,6 +1555,9 @@ bool HasUnsupportedNativeVop3Modifiers(Opcode opcode, bool permlane, bool mad_mi
}
switch (opcode) {
case Opcode::V_LDEXP_F32: return (abs & ~1u) != 0u || op_sel != 0u || (neg & ~1u) != 0u;
case Opcode::V_INTERP_P1LL_F16:
case Opcode::V_INTERP_P2_F16:
return abs != 0u || (op_sel & 0x7u) != 0u || clamp != 0u || omod != 0u || neg != 0u;
case Opcode::V_CNDMASK_B32:
return (abs & ~0x3u) != 0u || op_sel != 0u || clamp != 0u || omod != 0u ||
(neg & ~0x3u) != 0u;
@@ -1664,6 +1669,17 @@ bool IsVopcCompareExec(Opcode opcode) {
}
}
// VINTRP and VOP3 interpolation share one layout: src0 = I/J VGPR, src1 = attr, src2 = chan.
void SetInterpolationAttribute(Instruction& inst, uint32_t attr, uint32_t chan) {
inst.src1.kind = OperandKind::IntegerInlineConstant;
inst.src1.value = attr;
inst.src1.signed_val = static_cast<int32_t>(attr);
inst.src2.kind = OperandKind::IntegerInlineConstant;
inst.src2.value = chan;
inst.src2.signed_val = static_cast<int32_t>(chan);
inst.src_count = 3;
}
} // namespace
void DecodeVop2(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
@@ -1830,6 +1846,15 @@ void DecodeVop3(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
inst.src_count = 0;
return;
}
if (inst.opcode == Opcode::V_INTERP_P1LL_F16 || inst.opcode == Opcode::V_INTERP_P2_F16) {
// SRC0 packs {attr_word, attr_chan, attr}; attr_word selects the high f16 of the dword.
DecodeVectorGpr(vdst, inst.dst);
DecodeScalarSource(src1, pc, inst.src0);
SetInterpolationAttribute(inst, src0 & 0x3fu, (src0 >> 6u) & 0x3u);
inst.src1.op_sel = (src0 & 0x100u) != 0u;
inst.dst.sdwa_sel = (op_sel & 0x8u) != 0u ? 5u : 4u;
return;
}
if (compare_exec) {
inst.dst.kind = OperandKind::ExecLo;
} else if (scalar_dst) {
@@ -2004,13 +2029,7 @@ void DecodeVintrp(uint32_t pc, std::span<const uint32_t> code, uint32_t word_ind
} else {
DecodeVectorGpr(vsrc, inst.src0);
}
inst.src1.kind = OperandKind::IntegerInlineConstant;
inst.src1.value = attr;
inst.src1.signed_val = static_cast<int32_t>(attr);
inst.src2.kind = OperandKind::IntegerInlineConstant;
inst.src2.value = chan;
inst.src2.signed_val = static_cast<int32_t>(chan);
inst.src_count = 3;
SetInterpolationAttribute(inst, attr, chan);
}
} // namespace Libs::Graphics::ShaderRecompiler::Decoder
src/graphics/shader/recompiler/frontend/translate/Attribute.cpp
+11 -3
View File
@@ -73,14 +73,19 @@ void Translator::TranslateEmbeddedFetch(const Decoder::Instruction& inst, uint32
}
}
void Translator::V_INTERP_P1_F32() {}
void Translator::V_INTERP_P2_F32(const Decoder::Instruction& inst) {
const auto value = ir.Emit(IR::ValueOpcode::GetAttribute,
{IR::Value(inst.src1.value), IR::Value(inst.src2.value)});
WriteOperand(inst.dst, value);
}
void Translator::V_INTERP_P2_F16(const Decoder::Instruction& inst) {
const auto value = ir.Emit(IR::ValueOpcode::GetAttributeF16,
{IR::Value(inst.src1.value), IR::Value(inst.src2.value),
IR::Value(uint32_t {inst.src1.op_sel})});
WriteF16(inst.dst, IR::F32(value));
}
void Translator::V_INTERP_MOV_F32(const Decoder::Instruction& inst) {
if (inst.src0.value >= 3u) {
EXIT("v_interp_mov_f32 mode %u is reserved at pc 0x%08x", inst.src0.value, inst.pc);
@@ -108,8 +113,11 @@ void Translator::EXP(const Decoder::Instruction& inst) {
void Translator::EmitInterpolation(const Decoder::Instruction& inst) {
switch (inst.opcode) {
case Decoder::Opcode::V_INTERP_P1_F32: V_INTERP_P1_F32(); return;
// P2 interpolates the whole attribute, so the P1 halves have nothing to compute.
case Decoder::Opcode::V_INTERP_P1_F32:
case Decoder::Opcode::V_INTERP_P1LL_F16: return;
case Decoder::Opcode::V_INTERP_P2_F32: V_INTERP_P2_F32(inst); return;
case Decoder::Opcode::V_INTERP_P2_F16: V_INTERP_P2_F16(inst); return;
case Decoder::Opcode::V_INTERP_MOV_F32: V_INTERP_MOV_F32(inst); return;
default: return FailMissingTranslation(inst);
}
src/graphics/shader/recompiler/frontend/translate/Translator.h
+1 -1
View File
@@ -240,8 +240,8 @@ private:
void V_READLANE_B32(const Decoder::Instruction& inst);
void V_WRITELANE_B32(const Decoder::Instruction& inst);
void V_PERMLANE16_B32(const Decoder::Instruction& inst, bool x16);
void V_INTERP_P1_F32();
void V_INTERP_P2_F32(const Decoder::Instruction& inst);
void V_INTERP_P2_F16(const Decoder::Instruction& inst);
void V_INTERP_MOV_F32(const Decoder::Instruction& inst);
void EXP(const Decoder::Instruction& inst);
src/graphics/shader/recompiler/frontend/translate/Vector.cpp
+2
View File
@@ -6,6 +6,8 @@ void Translator::EmitVector(const Decoder::Instruction& inst) {
using O = Decoder::Opcode;
switch (inst.opcode) {
case O::V_NOP: return;
case O::V_INTERP_P1LL_F16:
case O::V_INTERP_P2_F16: return EmitInterpolation(inst);
case O::V_ADD_I32: ADD_U32(inst, true, false); return;
case O::V_ADDC_U32: ADD_U32(inst, true, true); return;
case O::V_SUB_I32: SUB_U32(inst, true, false); return;