raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-20 19:07:01 +02:00
Code Issues Packages Projects Releases Wiki Activity

Decode unsigned byte sources for SDWA half-float conversion

Browse Source
This commit is contained in:
nmzik
2026-09-21 11:13:21 +04:00
parent d792dfa885
commit cfb8f7b246
3 changed files with 47 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/frontend/decode/VectorAluOps.cpp
+12 -10
View File
@@ -527,8 +527,8 @@ constexpr Vop1SdwaRule VOP1_SDWA_RULES[] = {
{Opcode::V_CVT_F32_F16, SdwaSelWords() | SdwaSelFull(), 0, 0, true},
{Opcode::V_CVT_F16_F32, SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), SdwaSelWords(),
SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), true},
{Opcode::V_CVT_F16_U16, SdwaSelWords() | SdwaSelFull(), SdwaSelWords(),
SdwaSelWords() | SdwaSelFull(), false},
{Opcode::V_CVT_F16_U16, SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), SdwaSelWords(),
SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), false},
{Opcode::V_CVT_U16_F16, SdwaSelWords() | SdwaSelFull(), SdwaSelWords(),
SdwaSelWords() | SdwaSelFull(), true},
{Opcode::V_CVT_F16_I16, SdwaSelWords() | SdwaSelFull(), SdwaSelWords(),
@@ -618,13 +618,14 @@ bool SupportsVop1Clamp(Opcode opcode) {
}
bool ValidateVop1Sdwa(Instruction& inst, uint32_t opcode, uint32_t modifier) {
const auto dst_sel = (modifier >> 8u) & 0x7u;
const auto dst_u = (modifier >> 11u) & 0x3u;
const auto clamp = (modifier >> 13u) & 0x1u;
const auto omod = (modifier >> 14u) & 0x3u;
const auto src0_sel = (modifier >> 16u) & 0x7u;
const auto src0_neg = (modifier >> 20u) & 0x1u;
const auto src0_abs = (modifier >> 21u) & 0x1u;
const auto dst_sel = (modifier >> 8u) & 0x7u;
const auto dst_u = (modifier >> 11u) & 0x3u;
const auto clamp = (modifier >> 13u) & 0x1u;
const auto omod = (modifier >> 14u) & 0x3u;
const auto src0_sel = (modifier >> 16u) & 0x7u;
const auto src0_sext = (modifier >> 19u) & 0x1u;
const auto src0_neg = (modifier >> 20u) & 0x1u;
const auto src0_abs = (modifier >> 21u) & 0x1u;
if (src0_sel > 6u || dst_sel > 6u) {
SetUnsupported(inst, Family::VOP1, opcode, "VOP1 SDWA selector is invalid");
@@ -640,7 +641,8 @@ bool ValidateVop1Sdwa(Instruction& inst, uint32_t opcode, uint32_t modifier) {
"VOP1 SDWA destination selector is not supported");
return false;
}
if (!IsVop1SdwaSourceSupported(inst.opcode, src0_sel, src0_neg != 0u, src0_abs != 0u)) {
if (!IsVop1SdwaSourceSupported(inst.opcode, src0_sel, src0_neg != 0u, src0_abs != 0u) ||
(inst.opcode == Opcode::V_CVT_F16_U16 && src0_sel <= 3u && src0_sext != 0u)) {
SetUnsupported(inst, Family::VOP1, opcode, "VOP1 SDWA source selector is not supported");
return false;
}