diff options
| author | Takoroka <137927095+Takoroka@users.noreply.github.com> | 2026-10-30 14:22:05 -0400 |
|---|---|---|
| committer | Takoroka <137927095+Takoroka@users.noreply.github.com> | 2026-11-01 11:09:34 -0500 |
| commit | f4d79570e995c62aca41becf9d1cfa8c6cbec9ce (patch) | |
| tree | 7a09c9e373de1b30bb1c5ab1cf4680e2fc303eeb /src/shader_recompiler | |
| parent | f74256da81abebb0e413da7e0f1828ae3d54322e (diff) | |
nx2: lower MATCH.ALL into P6
MATCH.ALL used P6 as its scratch predicate and so rejected P6 as the
result. Kirby Air Riders' compute shaders write P6 (5 programs); use P5
as the scratch predicate for those.
Diffstat (limited to 'src/shader_recompiler')
| -rw-r--r-- | src/shader_recompiler/frontend/ampere/lower_to_maxwell.h | 15 |
1 files changed, 8 insertions, 7 deletions
diff --git a/src/shader_recompiler/frontend/ampere/lower_to_maxwell.h b/src/shader_recompiler/frontend/ampere/lower_to_maxwell.h index affc0f8207..6974a2e2dc 100644 --- a/src/shader_recompiler/frontend/ampere/lower_to_maxwell.h +++ b/src/shader_recompiler/frontend/ampere/lower_to_maxwell.h @@ -868,10 +868,11 @@ inline LoweredProgram LowerToMaxwell(std::span<const Instruction> code) { case 0x3a1: { // MATCH.ALL (.U32, or .U64 on bit 73) with only the all-equal predicate retained. const u32 source = insn.Bits(24, 8), target = insn.Bits(81, 3); const u32 words = insn.Bits(73, 1) ? 2 : 1; - if (dest != 255 || source >= 254 || (words == 2 && (source & 1)) || target >= 6 || + const u32 scratch = target == 6 ? 5U : 6U; // the comparison, saved around the match + if (dest != 255 || source >= 254 || (words == 2 && (source & 1)) || target == 7 || native_predicate != 7 || insn.words[1] || (insn.words[2] & ~0xe0200U) != 0) return fail(index, "native warp match mode"); - save_predicate(); + save_predicate(1U << scratch); // Compare every lane with the lowest active one, then require all to agree. emit(0x50d8000000070000ULL | 250 | (u64{7} << 39) | (u64{7} << 45) | (u64{1} << 48)); emit(0x5c10000000070000ULL | 251 | (u64{250} << 8) | (u64{255} << 20) | (u64{1} << 49)); @@ -880,13 +881,13 @@ inline LoweredProgram LowerToMaxwell(std::span<const Instruction> code) { for (u32 word = 0; word < words; ++word) emit(0xef10000000070000ULL | (251 + word) | (u64{source + word} << 8) | (u64{250} << 20) | (u64{1} << 29) | (u64{31} << 34) | (u64{7} << 48)); - emit(0x5b60000000070000ULL | 7 | (u64{6} << 3) | (u64{source} << 8) | + emit(0x5b60000000070000ULL | 7 | (u64{scratch} << 3) | (u64{source} << 8) | (u64{251} << 20) | (u64{7} << 39) | (u64{2} << 49)); if (words == 2) - emit(0x5b60000000070000ULL | 7 | (u64{6} << 3) | (u64{source + 1} << 8) | - (u64{252} << 20) | (u64{6} << 39) | (u64{2} << 49)); - emit(0x50d8000000070000ULL | 255 | (u64{6} << 39) | (u64{target} << 45)); - restore_predicate(); + emit(0x5b60000000070000ULL | 7 | (u64{scratch} << 3) | (u64{source + 1} << 8) | + (u64{252} << 20) | (u64{scratch} << 39) | (u64{2} << 49)); + emit(0x50d8000000070000ULL | 255 | (u64{scratch} << 39) | (u64{target} << 45)); + restore_predicate(1U << scratch); break; } case 0x80d: { // CSMTEST: a test of shading-mode state Eden does not model, taken as false. |