Skip to content

Commit 442b14d

Browse files
authored
AMDGPU: Fix adding m0 uses to gfx94/gfx12 ds atomics (#156402)
This was using the legacy multiclass which assumes the base form has an m0 use. Use the versions which assume no m0 as the base name. Most of the diff is shuffling around the pattern classes to avoid trying to match the nonexistent m0-having form.
1 parent c7b26bd commit 442b14d

File tree

1 file changed

+57
-51
lines changed

1 file changed

+57
-51
lines changed

llvm/lib/Target/AMDGPU/DSInstructions.td

Lines changed: 57 additions & 51 deletions
Original file line numberDiff line numberDiff line change
@@ -469,6 +469,56 @@ class DSAtomicRetPat<DS_Pseudo inst, ValueType vt, PatFrag frag, int complexity
469469
let AddedComplexity = complexity;
470470
}
471471

472+
multiclass DSAtomicRetPat_mc<DS_Pseudo inst, ValueType vt, string frag> {
473+
let OtherPredicates = [LDSRequiresM0Init] in {
474+
def : DSAtomicRetPat<inst, vt, !cast<PatFrag>(frag#"_local_m0_"#vt)>;
475+
}
476+
477+
let OtherPredicates = [NotLDSRequiresM0Init] in {
478+
def : DSAtomicRetPat<!cast<DS_Pseudo>(!cast<string>(inst)#"_gfx9"), vt,
479+
!cast<PatFrag>(frag#"_local_"#vt)>;
480+
}
481+
482+
let OtherPredicates = [HasGDS] in {
483+
def : DSAtomicRetPat<inst, vt, !cast<PatFrag>(frag#"_region_m0_"#vt),
484+
/* complexity */ 0, /* gds */ 1>;
485+
}
486+
}
487+
488+
multiclass DSAtomicRetNoRetPat_NoM0_mc<DS_Pseudo inst, DS_Pseudo noRetInst,
489+
ValueType vt, string frag> {
490+
def : DSAtomicRetPat<inst, vt,
491+
!cast<PatFrag>(frag#"_local_"#vt)>;
492+
def : DSAtomicRetPat<noRetInst, vt,
493+
!cast<PatFrag>(frag#"_local_noret_"#vt), /* complexity */ 1>;
494+
}
495+
496+
multiclass DSAtomicRetNoRetPat_mc<DS_Pseudo inst, DS_Pseudo noRetInst,
497+
ValueType vt, string frag> {
498+
let OtherPredicates = [LDSRequiresM0Init] in {
499+
def : DSAtomicRetPat<inst, vt,
500+
!cast<PatFrag>(frag#"_local_m0_"#vt)>;
501+
def : DSAtomicRetPat<noRetInst, vt,
502+
!cast<PatFrag>(frag#"_local_m0_noret_"#vt), /* complexity */ 1>;
503+
}
504+
505+
let OtherPredicates = [NotLDSRequiresM0Init] in {
506+
defm : DSAtomicRetNoRetPat_NoM0_mc<
507+
!cast<DS_Pseudo>(!cast<string>(inst)#"_gfx9"),
508+
!cast<DS_Pseudo>(!cast<string>(noRetInst)#"_gfx9"),
509+
vt, frag>;
510+
}
511+
512+
let OtherPredicates = [HasGDS] in {
513+
def : DSAtomicRetPat<inst, vt,
514+
!cast<PatFrag>(frag#"_region_m0_"#vt),
515+
/* complexity */ 0, /* gds */ 1>;
516+
def : DSAtomicRetPat<noRetInst, vt,
517+
!cast<PatFrag>(frag#"_region_m0_noret_"#vt),
518+
/* complexity */ 1, /* gds */ 1>;
519+
}
520+
}
521+
472522
defm DS_ADD_U32 : DS_1A1D_NORET_mc<"ds_add_u32">;
473523
defm DS_SUB_U32 : DS_1A1D_NORET_mc<"ds_sub_u32">;
474524
defm DS_RSUB_U32 : DS_1A1D_NORET_mc<"ds_rsub_u32">;
@@ -519,10 +569,10 @@ let SubtargetPredicate = HasLdsAtomicAddF64 in {
519569
} // End SubtargetPredicate = HasLdsAtomicAddF64
520570

521571
let SubtargetPredicate = HasAtomicDsPkAdd16Insts in {
522-
defm DS_PK_ADD_F16 : DS_1A1D_NORET_mc<"ds_pk_add_f16">;
523-
defm DS_PK_ADD_RTN_F16 : DS_1A1D_RET_mc<"ds_pk_add_rtn_f16", VGPR_32>;
524-
defm DS_PK_ADD_BF16 : DS_1A1D_NORET_mc<"ds_pk_add_bf16">;
525-
defm DS_PK_ADD_RTN_BF16 : DS_1A1D_RET_mc<"ds_pk_add_rtn_bf16", VGPR_32>;
572+
defm DS_PK_ADD_F16 : DS_1A1D_NORET_mc_gfx9<"ds_pk_add_f16">;
573+
defm DS_PK_ADD_RTN_F16 : DS_1A1D_RET_mc_gfx9<"ds_pk_add_rtn_f16", VGPR_32>;
574+
defm DS_PK_ADD_BF16 : DS_1A1D_NORET_mc_gfx9<"ds_pk_add_bf16">;
575+
defm DS_PK_ADD_RTN_BF16 : DS_1A1D_RET_mc_gfx9<"ds_pk_add_rtn_bf16", VGPR_32>;
526576
} // End SubtargetPredicate = HasAtomicDsPkAdd16Insts
527577

528578
defm DS_CMPSTORE_B32 : DS_1A2D_NORET_mc<"ds_cmpstore_b32">;
@@ -1089,50 +1139,6 @@ defm : DSWritePat_mc <DS_WRITE_B128, vt, "store_align_less_than_4_local">;
10891139

10901140
} // End AddedComplexity = 100
10911141

1092-
multiclass DSAtomicRetPat_mc<DS_Pseudo inst, ValueType vt, string frag> {
1093-
let OtherPredicates = [LDSRequiresM0Init] in {
1094-
def : DSAtomicRetPat<inst, vt, !cast<PatFrag>(frag#"_local_m0_"#vt)>;
1095-
}
1096-
1097-
let OtherPredicates = [NotLDSRequiresM0Init] in {
1098-
def : DSAtomicRetPat<!cast<DS_Pseudo>(!cast<string>(inst)#"_gfx9"), vt,
1099-
!cast<PatFrag>(frag#"_local_"#vt)>;
1100-
}
1101-
1102-
let OtherPredicates = [HasGDS] in {
1103-
def : DSAtomicRetPat<inst, vt, !cast<PatFrag>(frag#"_region_m0_"#vt),
1104-
/* complexity */ 0, /* gds */ 1>;
1105-
}
1106-
}
1107-
1108-
multiclass DSAtomicRetNoRetPat_mc<DS_Pseudo inst, DS_Pseudo noRetInst,
1109-
ValueType vt, string frag> {
1110-
let OtherPredicates = [LDSRequiresM0Init] in {
1111-
def : DSAtomicRetPat<inst, vt,
1112-
!cast<PatFrag>(frag#"_local_m0_"#vt)>;
1113-
def : DSAtomicRetPat<noRetInst, vt,
1114-
!cast<PatFrag>(frag#"_local_m0_noret_"#vt), /* complexity */ 1>;
1115-
}
1116-
1117-
let OtherPredicates = [NotLDSRequiresM0Init] in {
1118-
def : DSAtomicRetPat<!cast<DS_Pseudo>(!cast<string>(inst)#"_gfx9"), vt,
1119-
!cast<PatFrag>(frag#"_local_"#vt)>;
1120-
def : DSAtomicRetPat<!cast<DS_Pseudo>(!cast<string>(noRetInst)#"_gfx9"), vt,
1121-
!cast<PatFrag>(frag#"_local_noret_"#vt), /* complexity */ 1>;
1122-
}
1123-
1124-
let OtherPredicates = [HasGDS] in {
1125-
def : DSAtomicRetPat<inst, vt,
1126-
!cast<PatFrag>(frag#"_region_m0_"#vt),
1127-
/* complexity */ 0, /* gds */ 1>;
1128-
def : DSAtomicRetPat<noRetInst, vt,
1129-
!cast<PatFrag>(frag#"_region_m0_noret_"#vt),
1130-
/* complexity */ 1, /* gds */ 1>;
1131-
}
1132-
}
1133-
1134-
1135-
11361142
let SubtargetPredicate = isGFX6GFX7GFX8GFX9GFX10 in {
11371143
// Caution, the order of src and cmp is the *opposite* of the BUFFER_ATOMIC_CMPSWAP opcode.
11381144
class DSAtomicCmpXChgSwapped<DS_Pseudo inst, ValueType vt, PatFrag frag,
@@ -1210,8 +1216,8 @@ defm : DSAtomicRetNoRetPat_mc<DS_MAX_RTN_F32, DS_MAX_F32, f32, "atomic_load_fmax
12101216

12111217

12121218
let SubtargetPredicate = HasAtomicDsPkAdd16Insts in {
1213-
defm : DSAtomicRetNoRetPat_mc<DS_PK_ADD_RTN_F16, DS_PK_ADD_F16, v2f16, "atomic_load_fadd">;
1214-
defm : DSAtomicRetNoRetPat_mc<DS_PK_ADD_RTN_BF16, DS_PK_ADD_BF16, v2bf16, "atomic_load_fadd">;
1219+
defm : DSAtomicRetNoRetPat_NoM0_mc<DS_PK_ADD_RTN_F16, DS_PK_ADD_F16, v2f16, "atomic_load_fadd">;
1220+
defm : DSAtomicRetNoRetPat_NoM0_mc<DS_PK_ADD_RTN_BF16, DS_PK_ADD_BF16, v2bf16, "atomic_load_fadd">;
12151221
}
12161222

12171223
let SubtargetPredicate = isGFX6GFX7GFX8GFX9GFX10 in {
@@ -1263,7 +1269,7 @@ class DSAtomicRetPatIntrinsic<DS_Pseudo inst, ValueType vt, PatFrag frag,
12631269
} // End SubtargetPredicate = HasLdsAtomicAddF64
12641270

12651271
let SubtargetPredicate = HasAtomicDsPkAdd16Insts in {
1266-
defm : DSAtomicRetNoRetPat_mc<DS_PK_ADD_RTN_F16, DS_PK_ADD_F16, v2f16, "atomic_load_fadd">;
1272+
defm : DSAtomicRetNoRetPat_NoM0_mc<DS_PK_ADD_RTN_F16, DS_PK_ADD_F16, v2f16, "atomic_load_fadd">;
12671273
} // End SubtargetPredicate = HasAtomicDsPkAdd16Insts
12681274

12691275
let OtherPredicates = [HasGDS] in

0 commit comments

Comments
 (0)