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: decode FLAT, GLOBAL and SCRATCH D16 high-half loads (#1924)

Browse Source 
FLAT-family opcode 0x25 (FLAT/GLOBAL/SCRATCH_LOAD_SHORT_D16_HI) loads an
unsigned halfword into bits [31:16] of the destination and keeps bits [15:0].
Decode it beside the low-half variant and let it select the high half of the
existing partial destination write, so FLAT_LOAD needs no new path.

Ghost of Yotei CS 0x79b0066f0cd1f3cd stopped at
global_load_short_d16_hi v16, v18, s[0:1] offset:196 (0xdc9480c4 0x10000012).
This commit is contained in:
MrFr3di
2026-10-27 02:34:54 +05:00
committed by GitHub
parent f90a8b1589
commit 986721f5bc
6 changed files with 358 additions and 37 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/frontend/decode/MemoryOps.cpp
+12 -11
View File
@@ -95,14 +95,14 @@ constexpr MemoryOpcodeInfo MTBUF_OPCODE_LIST[] = {
};
constexpr MemoryOpcodeInfo FLAT_OPCODE_LIST[] = {
{0x08u, Opcode::FLAT_LOAD_UBYTE, 1, 8}, {0x09u, Opcode::FLAT_LOAD_SBYTE, 1, 8, true},
{0x0au, Opcode::FLAT_LOAD_USHORT, 1, 16}, {0x0bu, Opcode::FLAT_LOAD_SSHORT, 1, 16, true},
{0x0cu, Opcode::FLAT_LOAD_DWORD, 1, 32}, {0x0du, Opcode::FLAT_LOAD_DWORDX2, 2, 32},
{0x0eu, Opcode::FLAT_LOAD_DWORDX4, 4, 32}, {0x0fu, Opcode::FLAT_LOAD_DWORDX3, 3, 32},
{0x18u, Opcode::FLAT_STORE_BYTE, 1, 8}, {0x1au, Opcode::FLAT_STORE_SHORT, 1, 16},
{0x1cu, Opcode::FLAT_STORE_DWORD, 1, 32}, {0x1du, Opcode::FLAT_STORE_DWORDX2, 2, 32},
{0x1eu, Opcode::FLAT_STORE_DWORDX4, 4, 32}, {0x1fu, Opcode::FLAT_STORE_DWORDX3, 3, 32},
{0x24u, Opcode::FLAT_LOAD_SHORT_D16, 1, 16},
{0x08u, Opcode::FLAT_LOAD_UBYTE, 1, 8}, {0x09u, Opcode::FLAT_LOAD_SBYTE, 1, 8, true},
{0x0au, Opcode::FLAT_LOAD_USHORT, 1, 16}, {0x0bu, Opcode::FLAT_LOAD_SSHORT, 1, 16, true},
{0x0cu, Opcode::FLAT_LOAD_DWORD, 1, 32}, {0x0du, Opcode::FLAT_LOAD_DWORDX2, 2, 32},
{0x0eu, Opcode::FLAT_LOAD_DWORDX4, 4, 32}, {0x0fu, Opcode::FLAT_LOAD_DWORDX3, 3, 32},
{0x18u, Opcode::FLAT_STORE_BYTE, 1, 8}, {0x1au, Opcode::FLAT_STORE_SHORT, 1, 16},
{0x1cu, Opcode::FLAT_STORE_DWORD, 1, 32}, {0x1du, Opcode::FLAT_STORE_DWORDX2, 2, 32},
{0x1eu, Opcode::FLAT_STORE_DWORDX4, 4, 32}, {0x1fu, Opcode::FLAT_STORE_DWORDX3, 3, 32},
{0x24u, Opcode::FLAT_LOAD_SHORT_D16, 1, 16}, {0x25u, Opcode::FLAT_LOAD_SHORT_D16_HI, 1, 16},
};
constexpr MemoryOpcodeInfo DS_OPCODE_LIST[] = {
@@ -419,9 +419,10 @@ void DecodeFlat(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
}
DecodeVectorGpr(IsFlatStoreOpcode(inst.opcode) ? data : vdst, inst.dst);
if (inst.opcode == Opcode::FLAT_LOAD_SHORT_D16) {
// D16 loads reuse partial destinations to preserve the untouched high half.
inst.dst.sdwa_sel = 4u;
if (inst.opcode == Opcode::FLAT_LOAD_SHORT_D16 ||
inst.opcode == Opcode::FLAT_LOAD_SHORT_D16_HI) {
// D16 loads reuse partial destinations to preserve the untouched half.
inst.dst.sdwa_sel = inst.opcode == Opcode::FLAT_LOAD_SHORT_D16_HI ? 5u : 4u;
}
DecodeVectorGpr(addr, inst.src0);
inst.src_count = 1;
src/graphics/shader/recompiler/frontend/decode/ShaderDecoder.cpp
+1
View File
@@ -701,6 +701,7 @@ std::string InstructionToString(const Instruction& inst) {
case Opcode::FLAT_LOAD_USHORT:
case Opcode::FLAT_LOAD_SSHORT:
case Opcode::FLAT_LOAD_SHORT_D16:
case Opcode::FLAT_LOAD_SHORT_D16_HI:
case Opcode::FLAT_LOAD_DWORD:
case Opcode::FLAT_LOAD_DWORDX2:
case Opcode::FLAT_LOAD_DWORDX3:
src/graphics/shader/recompiler/frontend/decode/ShaderDecoder.h
+1
View File
@@ -612,6 +612,7 @@ enum class Opcode {
FLAT_LOAD_USHORT,
FLAT_LOAD_SSHORT,
FLAT_LOAD_SHORT_D16,
FLAT_LOAD_SHORT_D16_HI,
FLAT_LOAD_DWORD,
FLAT_LOAD_DWORDX2,
FLAT_LOAD_DWORDX3,
src/graphics/shader/recompiler/frontend/translate/Memory.cpp
+1
View File
@@ -1146,6 +1146,7 @@ void Translator::EmitMemory(const Decoder::Instruction& inst) {
case Decoder::Opcode::FLAT_LOAD_USHORT:
case Decoder::Opcode::FLAT_LOAD_SSHORT:
case Decoder::Opcode::FLAT_LOAD_SHORT_D16:
case Decoder::Opcode::FLAT_LOAD_SHORT_D16_HI:
case Decoder::Opcode::FLAT_LOAD_DWORD:
case Decoder::Opcode::FLAT_LOAD_DWORDX2:
case Decoder::Opcode::FLAT_LOAD_DWORDX3: