diff options
| author | Takoroka <137927095+Takoroka@users.noreply.github.com> | 2026-10-30 11:51:27 -0400 |
|---|---|---|
| committer | Takoroka <137927095+Takoroka@users.noreply.github.com> | 2026-11-01 11:05:52 -0500 |
| commit | 2f2b0a746832e94ea244062fdffa92102e9c5b6c (patch) | |
| tree | 06ddd3c34284e90828df6042ef734cd729f02a39 /src | |
| parent | f9c6ae332a0fb36ae1e2f0534e09b3094d5eb675 (diff) | |
nx2: lower ATOMS.POPC.INC with an immediate offset
The shared-memory increment was only lowered at a zero offset; Nintendo
Switch Sports Resort increments [R10+0x4]. Maxwell ATOMS takes the word
offset in its immediate field.
Diffstat (limited to 'src')
| -rw-r--r-- | src/shader_recompiler/frontend/ampere/lower_to_maxwell.h | 8 |
1 files changed, 5 insertions, 3 deletions
diff --git a/src/shader_recompiler/frontend/ampere/lower_to_maxwell.h b/src/shader_recompiler/frontend/ampere/lower_to_maxwell.h index ce69f59161..9568951012 100644 --- a/src/shader_recompiler/frontend/ampere/lower_to_maxwell.h +++ b/src/shader_recompiler/frontend/ampere/lower_to_maxwell.h @@ -884,8 +884,9 @@ inline LoweredProgram LowerToMaxwell(std::span<const Instruction> code) { break; } case 0xf8c: { // ATOMS.POPC.INC.32 with no returned value: one increment per active lane. - const u32 stride = insn.Bits(78, 2); - if (dest != 255 || insn.words[1] || (insn.words[2] & ~0xc000U) != 0x0d00003f) + const u32 stride = insn.Bits(78, 2), offset = insn.Bits(40, 24); + if (dest != 255 || (insn.words[1] & 0xff) || (offset & 3) || + (insn.words[2] & ~0xc000U) != 0x0d00003f) return fail(index, "native shared increment mode"); u32 address = insn.Bits(24, 8); if (stride && address != 255) { @@ -894,7 +895,8 @@ inline LoweredProgram LowerToMaxwell(std::span<const Instruction> code) { address = 251; } mov_imm(250, 1); - emit(0xec00000000000000ULL | predicate | 255 | (u64{address} << 8) | (u64{250} << 20)); + emit(0xec00000000000000ULL | predicate | 255 | (u64{address} << 8) | (u64{250} << 20) | + (u64{offset >> 2} << 30)); break; } case 0x38c: { // ATOMS.ADD.U32, naturally aligned shared byte offset. |