Skip to content

Commit d990f79

Browse files
arsenmarsenm
committed
AMDGPU: Fix cost model for 16-bit operations on gfx8
We should only divide the number of pieces to fit the packed instructions if we actually have pk instructions. This increases the cost of copysign, but is closer to the current codegen output. It could be much cheaper than it is now.
1 parent 82b3b3e commit d990f79

File tree

6 files changed

+522
-304
lines changed

6 files changed

+522
-304
lines changed

‎llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -721,7 +721,7 @@ GCNTTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
721721
if (SLT == MVT::f64)
722722
return LT.first * NElts * get64BitInstrCost(CostKind);
723723

724-
if ((ST->has16BitInsts() && (SLT == MVT::f16 || SLT == MVT::i16)) ||
724+
if ((ST->hasVOP3PInsts() && (SLT == MVT::f16 || SLT == MVT::i16)) ||
725725
(ST->hasPackedFP32Ops() && SLT == MVT::f32))
726726
NElts = (NElts + 1) / 2;
727727

‎llvm/test/Analysis/CostModel/AMDGPU/canonicalize.ll

Lines changed: 12 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -22,12 +22,12 @@ define void @canonicalize_f16() {
2222
;
2323
; GFX8-LABEL: 'canonicalize_f16'
2424
; GFX8-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %f16 = call half @llvm.canonicalize.f16(half undef)
25-
; GFX8-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2f16 = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> undef)
26-
; GFX8-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v3f16 = call <3 x half> @llvm.canonicalize.v3f16(<3 x half> undef)
27-
; GFX8-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4f16 = call <4 x half> @llvm.canonicalize.v4f16(<4 x half> undef)
28-
; GFX8-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v5f16 = call <5 x half> @llvm.canonicalize.v5f16(<5 x half> undef)
29-
; GFX8-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v16f16 = call <16 x half> @llvm.canonicalize.v16f16(<16 x half> undef)
30-
; GFX8-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %v17f16 = call <17 x half> @llvm.canonicalize.v17f16(<17 x half> undef)
25+
; GFX8-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2f16 = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> undef)
26+
; GFX8-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v3f16 = call <3 x half> @llvm.canonicalize.v3f16(<3 x half> undef)
27+
; GFX8-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v4f16 = call <4 x half> @llvm.canonicalize.v4f16(<4 x half> undef)
28+
; GFX8-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v5f16 = call <5 x half> @llvm.canonicalize.v5f16(<5 x half> undef)
29+
; GFX8-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %v16f16 = call <16 x half> @llvm.canonicalize.v16f16(<16 x half> undef)
30+
; GFX8-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %v17f16 = call <17 x half> @llvm.canonicalize.v17f16(<17 x half> undef)
3131
; GFX8-NEXT: Cost Model: Found an estimated cost of 10 for instruction: ret void
3232
;
3333
; GFX9-LABEL: 'canonicalize_f16'
@@ -62,12 +62,12 @@ define void @canonicalize_f16() {
6262
;
6363
; GFX8-SIZE-LABEL: 'canonicalize_f16'
6464
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %f16 = call half @llvm.canonicalize.f16(half undef)
65-
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2f16 = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> undef)
66-
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v3f16 = call <3 x half> @llvm.canonicalize.v3f16(<3 x half> undef)
67-
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4f16 = call <4 x half> @llvm.canonicalize.v4f16(<4 x half> undef)
68-
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v5f16 = call <5 x half> @llvm.canonicalize.v5f16(<5 x half> undef)
69-
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v16f16 = call <16 x half> @llvm.canonicalize.v16f16(<16 x half> undef)
70-
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %v17f16 = call <17 x half> @llvm.canonicalize.v17f16(<17 x half> undef)
65+
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2f16 = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> undef)
66+
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v3f16 = call <3 x half> @llvm.canonicalize.v3f16(<3 x half> undef)
67+
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v4f16 = call <4 x half> @llvm.canonicalize.v4f16(<4 x half> undef)
68+
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v5f16 = call <5 x half> @llvm.canonicalize.v5f16(<5 x half> undef)
69+
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %v16f16 = call <16 x half> @llvm.canonicalize.v16f16(<16 x half> undef)
70+
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %v17f16 = call <17 x half> @llvm.canonicalize.v17f16(<17 x half> undef)
7171
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
7272
;
7373
; GFX9-SIZE-LABEL: 'canonicalize_f16'

‎llvm/test/Analysis/CostModel/AMDGPU/copysign.ll

Lines changed: 14 additions & 14 deletions
Original file line numberDiff line numberDiff line change
@@ -23,13 +23,13 @@ define void @copysign_f16() {
2323
;
2424
; GFX8-LABEL: 'copysign_f16'
2525
; GFX8-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %f16 = call half @llvm.copysign.f16(half undef, half undef)
26-
; GFX8-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2f16 = call <2 x half> @llvm.copysign.v2f16(<2 x half> undef, <2 x half> undef)
27-
; GFX8-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v3f16 = call <3 x half> @llvm.copysign.v3f16(<3 x half> undef, <3 x half> undef)
28-
; GFX8-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4f16 = call <4 x half> @llvm.copysign.v4f16(<4 x half> undef, <4 x half> undef)
29-
; GFX8-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v5f16 = call <5 x half> @llvm.copysign.v5f16(<5 x half> undef, <5 x half> undef)
30-
; GFX8-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v8f16 = call <8 x half> @llvm.copysign.v8f16(<8 x half> undef, <8 x half> undef)
31-
; GFX8-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v9f16 = call <9 x half> @llvm.copysign.v9f16(<9 x half> undef, <9 x half> undef)
32-
; GFX8-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v16f16 = call <16 x half> @llvm.copysign.v16f16(<16 x half> undef, <16 x half> undef)
26+
; GFX8-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2f16 = call <2 x half> @llvm.copysign.v2f16(<2 x half> undef, <2 x half> undef)
27+
; GFX8-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v3f16 = call <3 x half> @llvm.copysign.v3f16(<3 x half> undef, <3 x half> undef)
28+
; GFX8-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v4f16 = call <4 x half> @llvm.copysign.v4f16(<4 x half> undef, <4 x half> undef)
29+
; GFX8-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v5f16 = call <5 x half> @llvm.copysign.v5f16(<5 x half> undef, <5 x half> undef)
30+
; GFX8-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v8f16 = call <8 x half> @llvm.copysign.v8f16(<8 x half> undef, <8 x half> undef)
31+
; GFX8-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %v9f16 = call <9 x half> @llvm.copysign.v9f16(<9 x half> undef, <9 x half> undef)
32+
; GFX8-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %v16f16 = call <16 x half> @llvm.copysign.v16f16(<16 x half> undef, <16 x half> undef)
3333
; GFX8-NEXT: Cost Model: Found an estimated cost of 10 for instruction: ret void
3434
;
3535
; GFX9-LABEL: 'copysign_f16'
@@ -67,13 +67,13 @@ define void @copysign_f16() {
6767
;
6868
; GFX8-SIZE-LABEL: 'copysign_f16'
6969
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %f16 = call half @llvm.copysign.f16(half undef, half undef)
70-
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2f16 = call <2 x half> @llvm.copysign.v2f16(<2 x half> undef, <2 x half> undef)
71-
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v3f16 = call <3 x half> @llvm.copysign.v3f16(<3 x half> undef, <3 x half> undef)
72-
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4f16 = call <4 x half> @llvm.copysign.v4f16(<4 x half> undef, <4 x half> undef)
73-
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v5f16 = call <5 x half> @llvm.copysign.v5f16(<5 x half> undef, <5 x half> undef)
74-
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v8f16 = call <8 x half> @llvm.copysign.v8f16(<8 x half> undef, <8 x half> undef)
75-
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v9f16 = call <9 x half> @llvm.copysign.v9f16(<9 x half> undef, <9 x half> undef)
76-
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v16f16 = call <16 x half> @llvm.copysign.v16f16(<16 x half> undef, <16 x half> undef)
70+
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2f16 = call <2 x half> @llvm.copysign.v2f16(<2 x half> undef, <2 x half> undef)
71+
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v3f16 = call <3 x half> @llvm.copysign.v3f16(<3 x half> undef, <3 x half> undef)
72+
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v4f16 = call <4 x half> @llvm.copysign.v4f16(<4 x half> undef, <4 x half> undef)
73+
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v5f16 = call <5 x half> @llvm.copysign.v5f16(<5 x half> undef, <5 x half> undef)
74+
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v8f16 = call <8 x half> @llvm.copysign.v8f16(<8 x half> undef, <8 x half> undef)
75+
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %v9f16 = call <9 x half> @llvm.copysign.v9f16(<9 x half> undef, <9 x half> undef)
76+
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %v16f16 = call <16 x half> @llvm.copysign.v16f16(<16 x half> undef, <16 x half> undef)
7777
; GFX8-SIZE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
7878
;
7979
; GFX9-SIZE-LABEL: 'copysign_f16'

0 commit comments

Comments
 (0)