Skip to content

Commit e36dd3e

Browse files
chuongg3davemgreen
chuongg3
authored and
davemgreen
committed
[AArch64] Fix cost modelling for SVE Min/Max Intrinsics
Add more legal types for SMIN, SMAX, UMIN, UMAX in cost modelling for AArch64 Differential Revision: https://reviews.llvm.org/D154622
1 parent a2426eb commit e36dd3e

File tree

3 files changed

+67
-65
lines changed

3 files changed

+67
-65
lines changed

‎llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -401,7 +401,9 @@ AArch64TTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
401401
case Intrinsic::smin:
402402
case Intrinsic::smax: {
403403
static const auto ValidMinMaxTys = {MVT::v8i8, MVT::v16i8, MVT::v4i16,
404-
MVT::v8i16, MVT::v2i32, MVT::v4i32};
404+
MVT::v8i16, MVT::v2i32, MVT::v4i32,
405+
MVT::nxv16i8, MVT::nxv8i16, MVT::nxv4i32,
406+
MVT::nxv2i64};
405407
auto LT = getTypeLegalizationCost(RetTy);
406408
// v2i64 types get converted to cmp+bif hence the cost of 2
407409
if (LT.second == MVT::v2i64)

‎llvm/test/Analysis/CostModel/AArch64/sve-intrinsics.ll

Lines changed: 8 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -44,13 +44,13 @@ define void @reductions(<vscale x 4 x i32> %v0, <vscale x 4 x i64> %v1, <vscale
4444
; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %xor_nxv4i32 = call i32 @llvm.vector.reduce.xor.nxv4i32(<vscale x 4 x i32> %v0)
4545
; CHECK-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %xor_nxv4i64 = call i64 @llvm.vector.reduce.xor.nxv4i64(<vscale x 4 x i64> %v1)
4646
; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %umin_nxv4i32 = call i32 @llvm.vector.reduce.umin.nxv4i32(<vscale x 4 x i32> %v0)
47-
; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %umin_nxv4i64 = call i64 @llvm.vector.reduce.umin.nxv4i64(<vscale x 4 x i64> %v1)
47+
; CHECK-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %umin_nxv4i64 = call i64 @llvm.vector.reduce.umin.nxv4i64(<vscale x 4 x i64> %v1)
4848
; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %smin_nxv4i32 = call i32 @llvm.vector.reduce.smin.nxv4i32(<vscale x 4 x i32> %v0)
49-
; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %smin_nxv4i64 = call i64 @llvm.vector.reduce.smin.nxv4i64(<vscale x 4 x i64> %v1)
49+
; CHECK-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %smin_nxv4i64 = call i64 @llvm.vector.reduce.smin.nxv4i64(<vscale x 4 x i64> %v1)
5050
; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %umax_nxv4i32 = call i32 @llvm.vector.reduce.umax.nxv4i32(<vscale x 4 x i32> %v0)
51-
; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %umax_nxv4i64 = call i64 @llvm.vector.reduce.umax.nxv4i64(<vscale x 4 x i64> %v1)
51+
; CHECK-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %umax_nxv4i64 = call i64 @llvm.vector.reduce.umax.nxv4i64(<vscale x 4 x i64> %v1)
5252
; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %smax_nxv4i32 = call i32 @llvm.vector.reduce.smax.nxv4i32(<vscale x 4 x i32> %v0)
53-
; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %smax_nxv4i64 = call i64 @llvm.vector.reduce.smax.nxv4i64(<vscale x 4 x i64> %v1)
53+
; CHECK-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %smax_nxv4i64 = call i64 @llvm.vector.reduce.smax.nxv4i64(<vscale x 4 x i64> %v1)
5454
; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %fadd_nxv4f32 = call fast float @llvm.vector.reduce.fadd.nxv4f32(float 0.000000e+00, <vscale x 4 x float> %v2)
5555
; CHECK-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %fadd_nxv4f64 = call fast double @llvm.vector.reduce.fadd.nxv4f64(double 0.000000e+00, <vscale x 4 x double> %v3)
5656
; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %fmin_nxv4f32 = call fast float @llvm.vector.reduce.fmin.nxv4f32(<vscale x 4 x float> %v2)
@@ -71,13 +71,13 @@ define void @reductions(<vscale x 4 x i32> %v0, <vscale x 4 x i64> %v1, <vscale
7171
; TYPE_BASED_ONLY-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %xor_nxv4i32 = call i32 @llvm.vector.reduce.xor.nxv4i32(<vscale x 4 x i32> %v0)
7272
; TYPE_BASED_ONLY-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %xor_nxv4i64 = call i64 @llvm.vector.reduce.xor.nxv4i64(<vscale x 4 x i64> %v1)
7373
; TYPE_BASED_ONLY-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %umin_nxv4i32 = call i32 @llvm.vector.reduce.umin.nxv4i32(<vscale x 4 x i32> %v0)
74-
; TYPE_BASED_ONLY-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %umin_nxv4i64 = call i64 @llvm.vector.reduce.umin.nxv4i64(<vscale x 4 x i64> %v1)
74+
; TYPE_BASED_ONLY-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %umin_nxv4i64 = call i64 @llvm.vector.reduce.umin.nxv4i64(<vscale x 4 x i64> %v1)
7575
; TYPE_BASED_ONLY-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %smin_nxv4i32 = call i32 @llvm.vector.reduce.smin.nxv4i32(<vscale x 4 x i32> %v0)
76-
; TYPE_BASED_ONLY-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %smin_nxv4i64 = call i64 @llvm.vector.reduce.smin.nxv4i64(<vscale x 4 x i64> %v1)
76+
; TYPE_BASED_ONLY-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %smin_nxv4i64 = call i64 @llvm.vector.reduce.smin.nxv4i64(<vscale x 4 x i64> %v1)
7777
; TYPE_BASED_ONLY-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %umax_nxv4i32 = call i32 @llvm.vector.reduce.umax.nxv4i32(<vscale x 4 x i32> %v0)
78-
; TYPE_BASED_ONLY-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %umax_nxv4i64 = call i64 @llvm.vector.reduce.umax.nxv4i64(<vscale x 4 x i64> %v1)
78+
; TYPE_BASED_ONLY-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %umax_nxv4i64 = call i64 @llvm.vector.reduce.umax.nxv4i64(<vscale x 4 x i64> %v1)
7979
; TYPE_BASED_ONLY-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %smax_nxv4i32 = call i32 @llvm.vector.reduce.smax.nxv4i32(<vscale x 4 x i32> %v0)
80-
; TYPE_BASED_ONLY-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %smax_nxv4i64 = call i64 @llvm.vector.reduce.smax.nxv4i64(<vscale x 4 x i64> %v1)
80+
; TYPE_BASED_ONLY-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %smax_nxv4i64 = call i64 @llvm.vector.reduce.smax.nxv4i64(<vscale x 4 x i64> %v1)
8181
; TYPE_BASED_ONLY-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %fadd_nxv4f32 = call fast float @llvm.vector.reduce.fadd.nxv4f32(float 0.000000e+00, <vscale x 4 x float> %v2)
8282
; TYPE_BASED_ONLY-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %fadd_nxv4f64 = call fast double @llvm.vector.reduce.fadd.nxv4f64(double 0.000000e+00, <vscale x 4 x double> %v3)
8383
; TYPE_BASED_ONLY-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %fmin_nxv4f32 = call fast float @llvm.vector.reduce.fmin.nxv4f32(<vscale x 4 x float> %v2)

0 commit comments

Comments
 (0)