Skip to content

Commit 2bb4e1e

Browse files
committed
AMDGPU: Fix adding m0 uses to gfx94/gfx12 ds atomics
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 aebfae0 commit 2bb4e1e

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
@@ -468,6 +468,56 @@ class DSAtomicRetPat<DS_Pseudo inst, ValueType vt, PatFrag frag, int complexity
468468
let AddedComplexity = complexity;
469469
}
470470

471+
multiclass DSAtomicRetPat_mc<DS_Pseudo inst, ValueType vt, string frag> {
472+
let OtherPredicates = [LDSRequiresM0Init] in {
473+
def : DSAtomicRetPat<inst, vt, !cast<PatFrag>(frag#"_local_m0_"#vt)>;
474+
}
475+
476+
let OtherPredicates = [NotLDSRequiresM0Init] in {
477+
def : DSAtomicRetPat<!cast<DS_Pseudo>(!cast<string>(inst)#"_gfx9"), vt,
478+
!cast<PatFrag>(frag#"_local_"#vt)>;
479+
}
480+
481+
let OtherPredicates = [HasGDS] in {
482+
def : DSAtomicRetPat<inst, vt, !cast<PatFrag>(frag#"_region_m0_"#vt),
483+
/* complexity */ 0, /* gds */ 1>;
484+
}
485+
}
486+
487+
multiclass DSAtomicRetNoRetPat_NoM0_mc<DS_Pseudo inst, DS_Pseudo noRetInst,
488+
ValueType vt, string frag> {
489+
def : DSAtomicRetPat<inst, vt,
490+
!cast<PatFrag>(frag#"_local_"#vt)>;
491+
def : DSAtomicRetPat<noRetInst, vt,
492+
!cast<PatFrag>(frag#"_local_noret_"#vt), /* complexity */ 1>;
493+
}
494+
495+
multiclass DSAtomicRetNoRetPat_mc<DS_Pseudo inst, DS_Pseudo noRetInst,
496+
ValueType vt, string frag> {
497+
let OtherPredicates = [LDSRequiresM0Init] in {
498+
def : DSAtomicRetPat<inst, vt,
499+
!cast<PatFrag>(frag#"_local_m0_"#vt)>;
500+
def : DSAtomicRetPat<noRetInst, vt,
501+
!cast<PatFrag>(frag#"_local_m0_noret_"#vt), /* complexity */ 1>;
502+
}
503+
504+
let OtherPredicates = [NotLDSRequiresM0Init] in {
505+
defm : DSAtomicRetNoRetPat_NoM0_mc<
506+
!cast<DS_Pseudo>(!cast<string>(inst)#"_gfx9"),
507+
!cast<DS_Pseudo>(!cast<string>(noRetInst)#"_gfx9"),
508+
vt, frag>;
509+
}
510+
511+
let OtherPredicates = [HasGDS] in {
512+
def : DSAtomicRetPat<inst, vt,
513+
!cast<PatFrag>(frag#"_region_m0_"#vt),
514+
/* complexity */ 0, /* gds */ 1>;
515+
def : DSAtomicRetPat<noRetInst, vt,
516+
!cast<PatFrag>(frag#"_region_m0_noret_"#vt),
517+
/* complexity */ 1, /* gds */ 1>;
518+
}
519+
}
520+
471521
defm DS_ADD_U32 : DS_1A1D_NORET_mc<"ds_add_u32">;
472522
defm DS_SUB_U32 : DS_1A1D_NORET_mc<"ds_sub_u32">;
473523
defm DS_RSUB_U32 : DS_1A1D_NORET_mc<"ds_rsub_u32">;
@@ -518,10 +568,10 @@ let SubtargetPredicate = HasLdsAtomicAddF64 in {
518568
} // End SubtargetPredicate = HasLdsAtomicAddF64
519569

520570
let SubtargetPredicate = HasAtomicDsPkAdd16Insts in {
521-
defm DS_PK_ADD_F16 : DS_1A1D_NORET_mc<"ds_pk_add_f16">;
522-
defm DS_PK_ADD_RTN_F16 : DS_1A1D_RET_mc<"ds_pk_add_rtn_f16", VGPR_32>;
523-
defm DS_PK_ADD_BF16 : DS_1A1D_NORET_mc<"ds_pk_add_bf16">;
524-
defm DS_PK_ADD_RTN_BF16 : DS_1A1D_RET_mc<"ds_pk_add_rtn_bf16", VGPR_32>;
571+
defm DS_PK_ADD_F16 : DS_1A1D_NORET_mc_gfx9<"ds_pk_add_f16">;
572+
defm DS_PK_ADD_RTN_F16 : DS_1A1D_RET_mc_gfx9<"ds_pk_add_rtn_f16", VGPR_32>;
573+
defm DS_PK_ADD_BF16 : DS_1A1D_NORET_mc_gfx9<"ds_pk_add_bf16">;
574+
defm DS_PK_ADD_RTN_BF16 : DS_1A1D_RET_mc_gfx9<"ds_pk_add_rtn_bf16", VGPR_32>;
525575
} // End SubtargetPredicate = HasAtomicDsPkAdd16Insts
526576

527577
defm DS_CMPSTORE_B32 : DS_1A2D_NORET_mc<"ds_cmpstore_b32">;
@@ -1088,50 +1138,6 @@ defm : DSWritePat_mc <DS_WRITE_B128, vt, "store_align_less_than_4_local">;
10881138

10891139
} // End AddedComplexity = 100
10901140

1091-
multiclass DSAtomicRetPat_mc<DS_Pseudo inst, ValueType vt, string frag> {
1092-
let OtherPredicates = [LDSRequiresM0Init] in {
1093-
def : DSAtomicRetPat<inst, vt, !cast<PatFrag>(frag#"_local_m0_"#vt)>;
1094-
}
1095-
1096-
let OtherPredicates = [NotLDSRequiresM0Init] in {
1097-
def : DSAtomicRetPat<!cast<DS_Pseudo>(!cast<string>(inst)#"_gfx9"), vt,
1098-
!cast<PatFrag>(frag#"_local_"#vt)>;
1099-
}
1100-
1101-
let OtherPredicates = [HasGDS] in {
1102-
def : DSAtomicRetPat<inst, vt, !cast<PatFrag>(frag#"_region_m0_"#vt),
1103-
/* complexity */ 0, /* gds */ 1>;
1104-
}
1105-
}
1106-
1107-
multiclass DSAtomicRetNoRetPat_mc<DS_Pseudo inst, DS_Pseudo noRetInst,
1108-
ValueType vt, string frag> {
1109-
let OtherPredicates = [LDSRequiresM0Init] in {
1110-
def : DSAtomicRetPat<inst, vt,
1111-
!cast<PatFrag>(frag#"_local_m0_"#vt)>;
1112-
def : DSAtomicRetPat<noRetInst, vt,
1113-
!cast<PatFrag>(frag#"_local_m0_noret_"#vt), /* complexity */ 1>;
1114-
}
1115-
1116-
let OtherPredicates = [NotLDSRequiresM0Init] in {
1117-
def : DSAtomicRetPat<!cast<DS_Pseudo>(!cast<string>(inst)#"_gfx9"), vt,
1118-
!cast<PatFrag>(frag#"_local_"#vt)>;
1119-
def : DSAtomicRetPat<!cast<DS_Pseudo>(!cast<string>(noRetInst)#"_gfx9"), vt,
1120-
!cast<PatFrag>(frag#"_local_noret_"#vt), /* complexity */ 1>;
1121-
}
1122-
1123-
let OtherPredicates = [HasGDS] in {
1124-
def : DSAtomicRetPat<inst, vt,
1125-
!cast<PatFrag>(frag#"_region_m0_"#vt),
1126-
/* complexity */ 0, /* gds */ 1>;
1127-
def : DSAtomicRetPat<noRetInst, vt,
1128-
!cast<PatFrag>(frag#"_region_m0_noret_"#vt),
1129-
/* complexity */ 1, /* gds */ 1>;
1130-
}
1131-
}
1132-
1133-
1134-
11351141
let SubtargetPredicate = isGFX6GFX7GFX8GFX9GFX10 in {
11361142
// Caution, the order of src and cmp is the *opposite* of the BUFFER_ATOMIC_CMPSWAP opcode.
11371143
class DSAtomicCmpXChgSwapped<DS_Pseudo inst, ValueType vt, PatFrag frag,
@@ -1209,8 +1215,8 @@ defm : DSAtomicRetNoRetPat_mc<DS_MAX_RTN_F32, DS_MAX_F32, f32, "atomic_load_fmax
12091215

12101216

12111217
let SubtargetPredicate = HasAtomicDsPkAdd16Insts in {
1212-
defm : DSAtomicRetNoRetPat_mc<DS_PK_ADD_RTN_F16, DS_PK_ADD_F16, v2f16, "atomic_load_fadd">;
1213-
defm : DSAtomicRetNoRetPat_mc<DS_PK_ADD_RTN_BF16, DS_PK_ADD_BF16, v2bf16, "atomic_load_fadd">;
1218+
defm : DSAtomicRetNoRetPat_NoM0_mc<DS_PK_ADD_RTN_F16, DS_PK_ADD_F16, v2f16, "atomic_load_fadd">;
1219+
defm : DSAtomicRetNoRetPat_NoM0_mc<DS_PK_ADD_RTN_BF16, DS_PK_ADD_BF16, v2bf16, "atomic_load_fadd">;
12141220
}
12151221

12161222
let SubtargetPredicate = isGFX6GFX7GFX8GFX9GFX10 in {
@@ -1262,7 +1268,7 @@ class DSAtomicRetPatIntrinsic<DS_Pseudo inst, ValueType vt, PatFrag frag,
12621268
} // End SubtargetPredicate = HasLdsAtomicAddF64
12631269

12641270
let SubtargetPredicate = HasAtomicDsPkAdd16Insts in {
1265-
defm : DSAtomicRetNoRetPat_mc<DS_PK_ADD_RTN_F16, DS_PK_ADD_F16, v2f16, "atomic_load_fadd">;
1271+
defm : DSAtomicRetNoRetPat_NoM0_mc<DS_PK_ADD_RTN_F16, DS_PK_ADD_F16, v2f16, "atomic_load_fadd">;
12661272
} // End SubtargetPredicate = HasAtomicDsPkAdd16Insts
12671273

12681274
let OtherPredicates = [HasGDS] in

0 commit comments

Comments
 (0)