[llvm] [AArch64] Don't allow mixed partial reductions without i8mm (PR #137602)
Sam Tebbs via llvm-commits
llvm-commits at lists.llvm.org
Thu May 1 02:08:17 PDT 2025
https://github.com/SamTebbs33 updated https://github.com/llvm/llvm-project/pull/137602
>From 01ea7d6439baaea6908f1b9007661cd97bb7c488 Mon Sep 17 00:00:00 2001
From: Samuel Tebbs <samuel.tebbs at arm.com>
Date: Wed, 23 Apr 2025 16:50:19 +0100
Subject: [PATCH 1/6] [AArch64] Don't allow mixed partial reductions without
i8mm
---
.../AArch64/AArch64TargetTransformInfo.cpp | 5 +-
.../partial-reduce-dot-product-mixed.ll | 114 +++++++++++-------
2 files changed, 72 insertions(+), 47 deletions(-)
diff --git a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
index c670b2ae71bf3..4c6803ed310b2 100644
--- a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
+++ b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
@@ -5345,10 +5345,9 @@ InstructionCost AArch64TTIImpl::getPartialReductionCost(
return Invalid;
// AArch64 supports lowering mixed extensions to a usdot but only if the
- // i8mm or sve/streaming features are available.
+ // i8mm feature is available.
if (OpAExtend == TTI::PR_None || OpBExtend == TTI::PR_None ||
- (OpAExtend != OpBExtend && !ST->hasMatMulInt8() &&
- !ST->isSVEorStreamingSVEAvailable()))
+ (OpAExtend != OpBExtend && !ST->hasMatMulInt8()))
return Invalid;
if (!BinOp || *BinOp != Instruction::Mul)
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
index cdd23b8f4d761..5950a3f84502b 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
@@ -46,38 +46,51 @@ define i32 @dotp_z_s(ptr %a, ptr %b) #0 {
; CHECK-NOI8MM-LABEL: define i32 @dotp_z_s(
; CHECK-NOI8MM-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NOI8MM-NEXT: entry:
+; CHECK-NOI8MM-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = mul i64 [[TMP0]], 8
; CHECK-NOI8MM-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-NOI8MM: vector.ph:
+; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = mul i64 [[TMP14]], 8
+; CHECK-NOI8MM-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
+; CHECK-NOI8MM-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
+; CHECK-NOI8MM-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NOI8MM-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 8
; CHECK-NOI8MM-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-NOI8MM: vector.body:
; CHECK-NOI8MM-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[VECTOR_BODY]] ]
; CHECK-NOI8MM-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-NOI8MM-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 0
-; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[TMP1]], i32 16
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
-; CHECK-NOI8MM-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 4
+; CHECK-NOI8MM-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP1]], i64 [[TMP9]]
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i8>, ptr [[TMP2]], align 1
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 4 x i8>, ptr [[TMP10]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = zext <vscale x 4 x i8> [[WIDE_LOAD]] to <vscale x 4 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = zext <vscale x 4 x i8> [[WIDE_LOAD2]] to <vscale x 4 x i32>
; CHECK-NOI8MM-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-NOI8MM-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
-; CHECK-NOI8MM-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[TMP6]], i32 16
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
-; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = sext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP10:%.*]] = sext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = mul <16 x i32> [[TMP9]], [[TMP4]]
-; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = mul <16 x i32> [[TMP10]], [[TMP5]]
-; CHECK-NOI8MM-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.experimental.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP11]])
-; CHECK-NOI8MM-NEXT: [[PARTIAL_REDUCE5]] = call <4 x i32> @llvm.experimental.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP12]])
-; CHECK-NOI8MM-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
-; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
-; CHECK-NOI8MM-NEXT: br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK-NOI8MM-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NOI8MM-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 4
+; CHECK-NOI8MM-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP16]]
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 4 x i8>, ptr [[TMP7]], align 1
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 4 x i8>, ptr [[TMP17]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP18:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD3]] to <vscale x 4 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP19:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD4]] to <vscale x 4 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP20:%.*]] = mul <vscale x 4 x i32> [[TMP18]], [[TMP11]]
+; CHECK-NOI8MM-NEXT: [[TMP21:%.*]] = mul <vscale x 4 x i32> [[TMP19]], [[TMP12]]
+; CHECK-NOI8MM-NEXT: [[TMP22]] = add <vscale x 4 x i32> [[TMP20]], [[VEC_PHI]]
+; CHECK-NOI8MM-NEXT: [[TMP23]] = add <vscale x 4 x i32> [[TMP21]], [[VEC_PHI1]]
+; CHECK-NOI8MM-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
+; CHECK-NOI8MM-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NOI8MM-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK-NOI8MM: middle.block:
-; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <4 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
-; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[BIN_RDX]])
-; CHECK-NOI8MM-NEXT: br i1 true, label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 4 x i32> [[TMP23]], [[TMP22]]
+; CHECK-NOI8MM-NEXT: [[TMP25:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[BIN_RDX]])
+; CHECK-NOI8MM-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
+; CHECK-NOI8MM-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
; CHECK-NOI8MM: scalar.ph:
;
entry:
@@ -143,38 +156,51 @@ define i32 @dotp_s_z(ptr %a, ptr %b) #0 {
; CHECK-NOI8MM-LABEL: define i32 @dotp_s_z(
; CHECK-NOI8MM-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] {
; CHECK-NOI8MM-NEXT: entry:
+; CHECK-NOI8MM-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = mul i64 [[TMP0]], 8
; CHECK-NOI8MM-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-NOI8MM: vector.ph:
+; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = mul i64 [[TMP14]], 8
+; CHECK-NOI8MM-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
+; CHECK-NOI8MM-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
+; CHECK-NOI8MM-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NOI8MM-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 8
; CHECK-NOI8MM-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-NOI8MM: vector.body:
; CHECK-NOI8MM-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[VECTOR_BODY]] ]
; CHECK-NOI8MM-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-NOI8MM-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 0
-; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[TMP1]], i32 16
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
-; CHECK-NOI8MM-NEXT: [[TMP4:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP5:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 4
+; CHECK-NOI8MM-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP1]], i64 [[TMP9]]
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i8>, ptr [[TMP2]], align 1
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 4 x i8>, ptr [[TMP10]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD]] to <vscale x 4 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD2]] to <vscale x 4 x i32>
; CHECK-NOI8MM-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-NOI8MM-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
-; CHECK-NOI8MM-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[TMP6]], i32 16
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
-; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = zext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP10:%.*]] = zext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = mul <16 x i32> [[TMP9]], [[TMP4]]
-; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = mul <16 x i32> [[TMP10]], [[TMP5]]
-; CHECK-NOI8MM-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.experimental.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP11]])
-; CHECK-NOI8MM-NEXT: [[PARTIAL_REDUCE5]] = call <4 x i32> @llvm.experimental.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP12]])
-; CHECK-NOI8MM-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
-; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
-; CHECK-NOI8MM-NEXT: br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK-NOI8MM-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NOI8MM-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 4
+; CHECK-NOI8MM-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP16]]
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 4 x i8>, ptr [[TMP7]], align 1
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 4 x i8>, ptr [[TMP17]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP18:%.*]] = zext <vscale x 4 x i8> [[WIDE_LOAD3]] to <vscale x 4 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP19:%.*]] = zext <vscale x 4 x i8> [[WIDE_LOAD4]] to <vscale x 4 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP20:%.*]] = mul <vscale x 4 x i32> [[TMP18]], [[TMP11]]
+; CHECK-NOI8MM-NEXT: [[TMP21:%.*]] = mul <vscale x 4 x i32> [[TMP19]], [[TMP12]]
+; CHECK-NOI8MM-NEXT: [[TMP22]] = add <vscale x 4 x i32> [[TMP20]], [[VEC_PHI]]
+; CHECK-NOI8MM-NEXT: [[TMP23]] = add <vscale x 4 x i32> [[TMP21]], [[VEC_PHI1]]
+; CHECK-NOI8MM-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
+; CHECK-NOI8MM-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NOI8MM-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK-NOI8MM: middle.block:
-; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <4 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
-; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[BIN_RDX]])
-; CHECK-NOI8MM-NEXT: br i1 true, label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 4 x i32> [[TMP23]], [[TMP22]]
+; CHECK-NOI8MM-NEXT: [[TMP25:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[BIN_RDX]])
+; CHECK-NOI8MM-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
+; CHECK-NOI8MM-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
; CHECK-NOI8MM: scalar.ph:
;
entry:
>From 960a50dcf0473f078505dab72561c073079357d8 Mon Sep 17 00:00:00 2001
From: Sam Tebbs <samuel.tebbs at arm.com>
Date: Tue, 29 Apr 2025 10:30:11 +0100
Subject: [PATCH 2/6] Add neon tests
---
.../partial-reduce-dot-product-mixed.ll | 195 ++++++++++++++++++
1 file changed, 195 insertions(+)
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
index 5950a3f84502b..7c538ce85bd78 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
@@ -225,8 +225,203 @@ for.exit: ; preds = %for.body
ret i32 %add
}
+define i32 @dotp_z_s_neon(ptr %a, ptr %b) #1 {
+; CHECK-LABEL: define i32 @dotp_z_s_neon(
+; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR1:[0-9]+]] {
+; CHECK-NEXT: entry:
+; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK: vector.ph:
+; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
+; CHECK: vector.body:
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP0:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
+; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[TMP0]], i32 0
+; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP0]], i32 16
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
+; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
+; CHECK-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
+; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[TMP5]], i32 0
+; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP5]], i32 16
+; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP6]], align 1
+; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
+; CHECK-NEXT: [[TMP8:%.*]] = sext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>
+; CHECK-NEXT: [[TMP9:%.*]] = sext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
+; CHECK-NEXT: [[TMP10:%.*]] = mul <16 x i32> [[TMP8]], [[TMP3]]
+; CHECK-NEXT: [[TMP11:%.*]] = mul <16 x i32> [[TMP9]], [[TMP4]]
+; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.experimental.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP10]])
+; CHECK-NEXT: [[PARTIAL_REDUCE5]] = call <4 x i32> @llvm.experimental.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP11]])
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
+; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
+; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
+; CHECK: middle.block:
+; CHECK-NEXT: [[BIN_RDX:%.*]] = add <4 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
+; CHECK-NEXT: [[TMP13:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[BIN_RDX]])
+; CHECK-NEXT: br i1 true, label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK: scalar.ph:
+;
+; CHECK-NOI8MM-LABEL: define i32 @dotp_z_s_neon(
+; CHECK-NOI8MM-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR1:[0-9]+]] {
+; CHECK-NOI8MM-NEXT: entry:
+; CHECK-NOI8MM-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK-NOI8MM: vector.ph:
+; CHECK-NOI8MM-NEXT: br label [[VECTOR_BODY:%.*]]
+; CHECK-NOI8MM: vector.body:
+; CHECK-NOI8MM-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <16 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP12:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <16 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP13:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[TMP0:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
+; CHECK-NOI8MM-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[TMP0]], i32 0
+; CHECK-NOI8MM-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP0]], i32 16
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
+; CHECK-NOI8MM-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[TMP5]], i32 0
+; CHECK-NOI8MM-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP5]], i32 16
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP6]], align 1
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP8:%.*]] = sext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = sext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP10:%.*]] = mul <16 x i32> [[TMP8]], [[TMP3]]
+; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = mul <16 x i32> [[TMP9]], [[TMP4]]
+; CHECK-NOI8MM-NEXT: [[TMP12]] = add <16 x i32> [[TMP10]], [[VEC_PHI]]
+; CHECK-NOI8MM-NEXT: [[TMP13]] = add <16 x i32> [[TMP11]], [[VEC_PHI1]]
+; CHECK-NOI8MM-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
+; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
+; CHECK-NOI8MM-NEXT: br i1 [[TMP14]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
+; CHECK-NOI8MM: middle.block:
+; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <16 x i32> [[TMP13]], [[TMP12]]
+; CHECK-NOI8MM-NEXT: [[TMP15:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[BIN_RDX]])
+; CHECK-NOI8MM-NEXT: br i1 true, label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK-NOI8MM: scalar.ph:
+;
+entry:
+ br label %for.body
+
+for.body: ; preds = %for.body, %entry
+ %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+ %accum = phi i32 [ 0, %entry ], [ %add, %for.body ]
+ %gep.a = getelementptr i8, ptr %a, i64 %iv
+ %load.a = load i8, ptr %gep.a, align 1
+ %ext.a = zext i8 %load.a to i32
+ %gep.b = getelementptr i8, ptr %b, i64 %iv
+ %load.b = load i8, ptr %gep.b, align 1
+ %ext.b = sext i8 %load.b to i32
+ %mul = mul i32 %ext.b, %ext.a
+ %add = add i32 %mul, %accum
+ %iv.next = add i64 %iv, 1
+ %exitcond.not = icmp eq i64 %iv.next, 1024
+ br i1 %exitcond.not, label %for.exit, label %for.body
+
+for.exit: ; preds = %for.body
+ ret i32 %add
+}
+
+define i32 @dotp_s_z_neon(ptr %a, ptr %b) #1 {
+; CHECK-LABEL: define i32 @dotp_s_z_neon(
+; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR1]] {
+; CHECK-NEXT: entry:
+; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK: vector.ph:
+; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
+; CHECK: vector.body:
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP0:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
+; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[TMP0]], i32 0
+; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP0]], i32 16
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
+; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
+; CHECK-NEXT: [[TMP3:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-NEXT: [[TMP4:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
+; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[TMP5]], i32 0
+; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP5]], i32 16
+; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP6]], align 1
+; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
+; CHECK-NEXT: [[TMP8:%.*]] = zext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>
+; CHECK-NEXT: [[TMP9:%.*]] = zext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
+; CHECK-NEXT: [[TMP10:%.*]] = mul <16 x i32> [[TMP8]], [[TMP3]]
+; CHECK-NEXT: [[TMP11:%.*]] = mul <16 x i32> [[TMP9]], [[TMP4]]
+; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.experimental.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP10]])
+; CHECK-NEXT: [[PARTIAL_REDUCE5]] = call <4 x i32> @llvm.experimental.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP11]])
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
+; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
+; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
+; CHECK: middle.block:
+; CHECK-NEXT: [[BIN_RDX:%.*]] = add <4 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
+; CHECK-NEXT: [[TMP13:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[BIN_RDX]])
+; CHECK-NEXT: br i1 true, label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK: scalar.ph:
+;
+; CHECK-NOI8MM-LABEL: define i32 @dotp_s_z_neon(
+; CHECK-NOI8MM-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR1]] {
+; CHECK-NOI8MM-NEXT: entry:
+; CHECK-NOI8MM-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK-NOI8MM: vector.ph:
+; CHECK-NOI8MM-NEXT: br label [[VECTOR_BODY:%.*]]
+; CHECK-NOI8MM: vector.body:
+; CHECK-NOI8MM-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <16 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP12:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <16 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP13:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[TMP0:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
+; CHECK-NOI8MM-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[TMP0]], i32 0
+; CHECK-NOI8MM-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP0]], i32 16
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP4:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
+; CHECK-NOI8MM-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[TMP5]], i32 0
+; CHECK-NOI8MM-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP5]], i32 16
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP6]], align 1
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP8:%.*]] = zext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = zext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP10:%.*]] = mul <16 x i32> [[TMP8]], [[TMP3]]
+; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = mul <16 x i32> [[TMP9]], [[TMP4]]
+; CHECK-NOI8MM-NEXT: [[TMP12]] = add <16 x i32> [[TMP10]], [[VEC_PHI]]
+; CHECK-NOI8MM-NEXT: [[TMP13]] = add <16 x i32> [[TMP11]], [[VEC_PHI1]]
+; CHECK-NOI8MM-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
+; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
+; CHECK-NOI8MM-NEXT: br i1 [[TMP14]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
+; CHECK-NOI8MM: middle.block:
+; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <16 x i32> [[TMP13]], [[TMP12]]
+; CHECK-NOI8MM-NEXT: [[TMP15:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[BIN_RDX]])
+; CHECK-NOI8MM-NEXT: br i1 true, label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK-NOI8MM: scalar.ph:
+;
+entry:
+ br label %for.body
+
+for.body: ; preds = %for.body, %entry
+ %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+ %accum = phi i32 [ 0, %entry ], [ %add, %for.body ]
+ %gep.a = getelementptr i8, ptr %a, i64 %iv
+ %load.a = load i8, ptr %gep.a, align 1
+ %ext.a = sext i8 %load.a to i32
+ %gep.b = getelementptr i8, ptr %b, i64 %iv
+ %load.b = load i8, ptr %gep.b, align 1
+ %ext.b = zext i8 %load.b to i32
+ %mul = mul i32 %ext.b, %ext.a
+ %add = add i32 %mul, %accum
+ %iv.next = add i64 %iv, 1
+ %exitcond.not = icmp eq i64 %iv.next, 1024
+ br i1 %exitcond.not, label %for.exit, label %for.body
+
+for.exit: ; preds = %for.body
+ ret i32 %add
+}
+
!7 = distinct !{!7, !8, !9, !10}
!8 = !{!"llvm.loop.mustprogress"}
!9 = !{!"llvm.loop.vectorize.predicate.enable", i1 true}
!10 = !{!"llvm.loop.vectorize.enable", i1 true}
attributes #0 = { vscale_range(1,16) "target-features"="+sve" }
+attributes #1 = { "target-features"="+neon" }
>From 7fe67b24cef4da4a1112c990a60c87518c420e13 Mon Sep 17 00:00:00 2001
From: Sam Tebbs <samuel.tebbs at arm.com>
Date: Tue, 29 Apr 2025 14:03:39 +0100
Subject: [PATCH 3/6] Only care about imm8 if the VF is fixed-width
---
.../AArch64/AArch64TargetTransformInfo.cpp | 6 +-
.../partial-reduce-dot-product-mixed.ll | 202 ++++++++++--------
2 files changed, 117 insertions(+), 91 deletions(-)
diff --git a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
index 4c6803ed310b2..e0b4a39944c72 100644
--- a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
+++ b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
@@ -5344,10 +5344,10 @@ InstructionCost AArch64TTIImpl::getPartialReductionCost(
} else
return Invalid;
- // AArch64 supports lowering mixed extensions to a usdot but only if the
- // i8mm feature is available.
+ // AArch64 supports lowering mixed fixed-width extensions to a usdot but only
+ // if the i8mm feature is available.
if (OpAExtend == TTI::PR_None || OpBExtend == TTI::PR_None ||
- (OpAExtend != OpBExtend && !ST->hasMatMulInt8()))
+ (OpAExtend != OpBExtend && VF.isFixed() && !ST->hasMatMulInt8()))
return Invalid;
if (!BinOp || *BinOp != Instruction::Mul)
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
index 7c538ce85bd78..5a9d6a59bb1e8 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
@@ -1,6 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --filter-out-after "^scalar.ph:" --version 4
-; RUN: opt -passes=loop-vectorize -enable-epilogue-vectorization=false -mattr=+i8mm,+dotprod -S < %s | FileCheck %s
-; RUN: opt -passes=loop-vectorize -enable-epilogue-vectorization=false -mattr=+dotprod -S < %s | FileCheck %s --check-prefix=CHECK-NOI8MM
+; RUN: opt -passes=loop-vectorize -enable-epilogue-vectorization=false -mattr=+i8mm,+dotprod --vectorizer-maximize-bandwidth -S < %s | FileCheck %s
+; RUN: opt -passes=loop-vectorize -enable-epilogue-vectorization=false -mattr=+dotprod -S --vectorizer-maximize-bandwidth < %s | FileCheck %s --check-prefix=CHECK-NOI8MM
target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
target triple = "aarch64-none-unknown-elf"
@@ -9,86 +9,99 @@ define i32 @dotp_z_s(ptr %a, ptr %b) #0 {
; CHECK-LABEL: define i32 @dotp_z_s(
; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NEXT: entry:
+; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP13:%.*]] = mul i64 [[TMP0]], 16
; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
+; CHECK-NEXT: [[TMP14:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP3:%.*]] = mul i64 [[TMP14]], 16
+; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
+; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
+; CHECK-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 16
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 0
-; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[TMP1]], i32 16
-; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
-; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
-; CHECK-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; CHECK-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 8
+; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP1]], i64 [[TMP9]]
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP2]], align 1
+; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
+; CHECK-NEXT: [[TMP11:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
+; CHECK-NEXT: [[TMP12:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
-; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[TMP6]], i32 16
-; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
-; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
-; CHECK-NEXT: [[TMP9:%.*]] = sext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>
-; CHECK-NEXT: [[TMP10:%.*]] = sext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
-; CHECK-NEXT: [[TMP11:%.*]] = mul <16 x i32> [[TMP9]], [[TMP4]]
-; CHECK-NEXT: [[TMP12:%.*]] = mul <16 x i32> [[TMP10]], [[TMP5]]
-; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.experimental.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP11]])
-; CHECK-NEXT: [[PARTIAL_REDUCE5]] = call <4 x i32> @llvm.experimental.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP12]])
-; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
-; CHECK-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
-; CHECK-NEXT: br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 8
+; CHECK-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP16]]
+; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
+; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP17]], align 1
+; CHECK-NEXT: [[TMP18:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
+; CHECK-NEXT: [[TMP19:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
+; CHECK-NEXT: [[TMP20:%.*]] = mul <vscale x 8 x i32> [[TMP18]], [[TMP11]]
+; CHECK-NEXT: [[TMP21:%.*]] = mul <vscale x 8 x i32> [[TMP19]], [[TMP12]]
+; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 2 x i32> @llvm.experimental.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI]], <vscale x 8 x i32> [[TMP20]])
+; CHECK-NEXT: [[PARTIAL_REDUCE5]] = call <vscale x 2 x i32> @llvm.experimental.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI1]], <vscale x 8 x i32> [[TMP21]])
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
+; CHECK-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK: middle.block:
-; CHECK-NEXT: [[BIN_RDX:%.*]] = add <4 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
-; CHECK-NEXT: [[TMP14:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[BIN_RDX]])
-; CHECK-NEXT: br i1 true, label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK-NEXT: [[BIN_RDX:%.*]] = add <vscale x 2 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
+; CHECK-NEXT: [[TMP23:%.*]] = call i32 @llvm.vector.reduce.add.nxv2i32(<vscale x 2 x i32> [[BIN_RDX]])
+; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
+; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
;
; CHECK-NOI8MM-LABEL: define i32 @dotp_z_s(
; CHECK-NOI8MM-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NOI8MM-NEXT: entry:
; CHECK-NOI8MM-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = mul i64 [[TMP0]], 8
+; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = mul i64 [[TMP0]], 16
; CHECK-NOI8MM-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-NOI8MM: vector.ph:
; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = mul i64 [[TMP14]], 8
+; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = mul i64 [[TMP14]], 16
; CHECK-NOI8MM-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
; CHECK-NOI8MM-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
; CHECK-NOI8MM-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 8
+; CHECK-NOI8MM-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 16
; CHECK-NOI8MM-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-NOI8MM: vector.body:
; CHECK-NOI8MM-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
; CHECK-NOI8MM-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-NOI8MM-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 0
; CHECK-NOI8MM-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 4
+; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 8
; CHECK-NOI8MM-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP1]], i64 [[TMP9]]
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i8>, ptr [[TMP2]], align 1
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 4 x i8>, ptr [[TMP10]], align 1
-; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = zext <vscale x 4 x i8> [[WIDE_LOAD]] to <vscale x 4 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = zext <vscale x 4 x i8> [[WIDE_LOAD2]] to <vscale x 4 x i32>
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP2]], align 1
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
; CHECK-NOI8MM-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-NOI8MM-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
; CHECK-NOI8MM-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 4
+; CHECK-NOI8MM-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 8
; CHECK-NOI8MM-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP16]]
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 4 x i8>, ptr [[TMP7]], align 1
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 4 x i8>, ptr [[TMP17]], align 1
-; CHECK-NOI8MM-NEXT: [[TMP18:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD3]] to <vscale x 4 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP19:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD4]] to <vscale x 4 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP20:%.*]] = mul <vscale x 4 x i32> [[TMP18]], [[TMP11]]
-; CHECK-NOI8MM-NEXT: [[TMP21:%.*]] = mul <vscale x 4 x i32> [[TMP19]], [[TMP12]]
-; CHECK-NOI8MM-NEXT: [[TMP22]] = add <vscale x 4 x i32> [[TMP20]], [[VEC_PHI]]
-; CHECK-NOI8MM-NEXT: [[TMP23]] = add <vscale x 4 x i32> [[TMP21]], [[VEC_PHI1]]
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP17]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP18:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP19:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP20:%.*]] = mul <vscale x 8 x i32> [[TMP18]], [[TMP11]]
+; CHECK-NOI8MM-NEXT: [[TMP21:%.*]] = mul <vscale x 8 x i32> [[TMP19]], [[TMP12]]
+; CHECK-NOI8MM-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 2 x i32> @llvm.experimental.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI]], <vscale x 8 x i32> [[TMP20]])
+; CHECK-NOI8MM-NEXT: [[PARTIAL_REDUCE5]] = call <vscale x 2 x i32> @llvm.experimental.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI1]], <vscale x 8 x i32> [[TMP21]])
; CHECK-NOI8MM-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
; CHECK-NOI8MM-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NOI8MM-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK-NOI8MM: middle.block:
-; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 4 x i32> [[TMP23]], [[TMP22]]
-; CHECK-NOI8MM-NEXT: [[TMP25:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[BIN_RDX]])
+; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 2 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
+; CHECK-NOI8MM-NEXT: [[TMP23:%.*]] = call i32 @llvm.vector.reduce.add.nxv2i32(<vscale x 2 x i32> [[BIN_RDX]])
; CHECK-NOI8MM-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
; CHECK-NOI8MM-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
; CHECK-NOI8MM: scalar.ph:
@@ -119,86 +132,99 @@ define i32 @dotp_s_z(ptr %a, ptr %b) #0 {
; CHECK-LABEL: define i32 @dotp_s_z(
; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: entry:
+; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP13:%.*]] = mul i64 [[TMP0]], 16
; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
+; CHECK-NEXT: [[TMP14:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP3:%.*]] = mul i64 [[TMP14]], 16
+; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
+; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
+; CHECK-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 16
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 0
-; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[TMP1]], i32 16
-; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
-; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
-; CHECK-NEXT: [[TMP4:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; CHECK-NEXT: [[TMP5:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 8
+; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP1]], i64 [[TMP9]]
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP2]], align 1
+; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
+; CHECK-NEXT: [[TMP11:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
+; CHECK-NEXT: [[TMP12:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
-; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[TMP6]], i32 16
-; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
-; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
-; CHECK-NEXT: [[TMP9:%.*]] = zext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>
-; CHECK-NEXT: [[TMP10:%.*]] = zext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
-; CHECK-NEXT: [[TMP11:%.*]] = mul <16 x i32> [[TMP9]], [[TMP4]]
-; CHECK-NEXT: [[TMP12:%.*]] = mul <16 x i32> [[TMP10]], [[TMP5]]
-; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.experimental.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP11]])
-; CHECK-NEXT: [[PARTIAL_REDUCE5]] = call <4 x i32> @llvm.experimental.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP12]])
-; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
-; CHECK-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
-; CHECK-NEXT: br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 8
+; CHECK-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP16]]
+; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
+; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP17]], align 1
+; CHECK-NEXT: [[TMP18:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
+; CHECK-NEXT: [[TMP19:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
+; CHECK-NEXT: [[TMP20:%.*]] = mul <vscale x 8 x i32> [[TMP18]], [[TMP11]]
+; CHECK-NEXT: [[TMP21:%.*]] = mul <vscale x 8 x i32> [[TMP19]], [[TMP12]]
+; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 2 x i32> @llvm.experimental.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI]], <vscale x 8 x i32> [[TMP20]])
+; CHECK-NEXT: [[PARTIAL_REDUCE5]] = call <vscale x 2 x i32> @llvm.experimental.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI1]], <vscale x 8 x i32> [[TMP21]])
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
+; CHECK-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK: middle.block:
-; CHECK-NEXT: [[BIN_RDX:%.*]] = add <4 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
-; CHECK-NEXT: [[TMP14:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[BIN_RDX]])
-; CHECK-NEXT: br i1 true, label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK-NEXT: [[BIN_RDX:%.*]] = add <vscale x 2 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
+; CHECK-NEXT: [[TMP23:%.*]] = call i32 @llvm.vector.reduce.add.nxv2i32(<vscale x 2 x i32> [[BIN_RDX]])
+; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
+; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
;
; CHECK-NOI8MM-LABEL: define i32 @dotp_s_z(
; CHECK-NOI8MM-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] {
; CHECK-NOI8MM-NEXT: entry:
; CHECK-NOI8MM-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = mul i64 [[TMP0]], 8
+; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = mul i64 [[TMP0]], 16
; CHECK-NOI8MM-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-NOI8MM: vector.ph:
; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = mul i64 [[TMP14]], 8
+; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = mul i64 [[TMP14]], 16
; CHECK-NOI8MM-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
; CHECK-NOI8MM-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
; CHECK-NOI8MM-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 8
+; CHECK-NOI8MM-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 16
; CHECK-NOI8MM-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-NOI8MM: vector.body:
; CHECK-NOI8MM-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
; CHECK-NOI8MM-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-NOI8MM-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 0
; CHECK-NOI8MM-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 4
+; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 8
; CHECK-NOI8MM-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP1]], i64 [[TMP9]]
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i8>, ptr [[TMP2]], align 1
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 4 x i8>, ptr [[TMP10]], align 1
-; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD]] to <vscale x 4 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD2]] to <vscale x 4 x i32>
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP2]], align 1
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
; CHECK-NOI8MM-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-NOI8MM-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
; CHECK-NOI8MM-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 4
+; CHECK-NOI8MM-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 8
; CHECK-NOI8MM-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP16]]
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 4 x i8>, ptr [[TMP7]], align 1
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 4 x i8>, ptr [[TMP17]], align 1
-; CHECK-NOI8MM-NEXT: [[TMP18:%.*]] = zext <vscale x 4 x i8> [[WIDE_LOAD3]] to <vscale x 4 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP19:%.*]] = zext <vscale x 4 x i8> [[WIDE_LOAD4]] to <vscale x 4 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP20:%.*]] = mul <vscale x 4 x i32> [[TMP18]], [[TMP11]]
-; CHECK-NOI8MM-NEXT: [[TMP21:%.*]] = mul <vscale x 4 x i32> [[TMP19]], [[TMP12]]
-; CHECK-NOI8MM-NEXT: [[TMP22]] = add <vscale x 4 x i32> [[TMP20]], [[VEC_PHI]]
-; CHECK-NOI8MM-NEXT: [[TMP23]] = add <vscale x 4 x i32> [[TMP21]], [[VEC_PHI1]]
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP17]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP18:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP19:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP20:%.*]] = mul <vscale x 8 x i32> [[TMP18]], [[TMP11]]
+; CHECK-NOI8MM-NEXT: [[TMP21:%.*]] = mul <vscale x 8 x i32> [[TMP19]], [[TMP12]]
+; CHECK-NOI8MM-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 2 x i32> @llvm.experimental.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI]], <vscale x 8 x i32> [[TMP20]])
+; CHECK-NOI8MM-NEXT: [[PARTIAL_REDUCE5]] = call <vscale x 2 x i32> @llvm.experimental.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI1]], <vscale x 8 x i32> [[TMP21]])
; CHECK-NOI8MM-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
; CHECK-NOI8MM-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NOI8MM-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK-NOI8MM: middle.block:
-; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 4 x i32> [[TMP23]], [[TMP22]]
-; CHECK-NOI8MM-NEXT: [[TMP25:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[BIN_RDX]])
+; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 2 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
+; CHECK-NOI8MM-NEXT: [[TMP23:%.*]] = call i32 @llvm.vector.reduce.add.nxv2i32(<vscale x 2 x i32> [[BIN_RDX]])
; CHECK-NOI8MM-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
; CHECK-NOI8MM-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
; CHECK-NOI8MM: scalar.ph:
>From 7fc0024935be27ce27b122de366c538c47a75892 Mon Sep 17 00:00:00 2001
From: Samuel Tebbs <samuel.tebbs at arm.com>
Date: Wed, 30 Apr 2025 15:21:05 +0100
Subject: [PATCH 4/6] Removed isFixed check
---
.../AArch64/AArch64TargetTransformInfo.cpp | 2 +-
.../partial-reduce-dot-product-mixed.ll | 112 +++++++++---------
2 files changed, 57 insertions(+), 57 deletions(-)
diff --git a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
index e0b4a39944c72..74281614d77a8 100644
--- a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
+++ b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
@@ -5347,7 +5347,7 @@ InstructionCost AArch64TTIImpl::getPartialReductionCost(
// AArch64 supports lowering mixed fixed-width extensions to a usdot but only
// if the i8mm feature is available.
if (OpAExtend == TTI::PR_None || OpBExtend == TTI::PR_None ||
- (OpAExtend != OpBExtend && VF.isFixed() && !ST->hasMatMulInt8()))
+ (OpAExtend != OpBExtend && !ST->hasMatMulInt8()))
return Invalid;
if (!BinOp || *BinOp != Instruction::Mul)
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
index 5a9d6a59bb1e8..04fe9ce96754a 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
@@ -10,11 +10,11 @@ define i32 @dotp_z_s(ptr %a, ptr %b) #0 {
; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP13:%.*]] = mul i64 [[TMP0]], 16
+; CHECK-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 16
; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
-; CHECK-NEXT: [[TMP14:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP3:%.*]] = mul i64 [[TMP14]], 16
+; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 16
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
; CHECK-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
@@ -24,21 +24,21 @@ define i32 @dotp_z_s(ptr %a, ptr %b) #0 {
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
-; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 0
+; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
+; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
; CHECK-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 8
-; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP1]], i64 [[TMP9]]
-; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP2]], align 1
+; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP9]]
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
; CHECK-NEXT: [[TMP11:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
; CHECK-NEXT: [[TMP12:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
-; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
-; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
+; CHECK-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
+; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[TMP13]], i32 0
; CHECK-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 8
-; CHECK-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP16]]
-; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
+; CHECK-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP13]], i64 [[TMP16]]
+; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP14]], align 1
; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP17]], align 1
; CHECK-NEXT: [[TMP18:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
; CHECK-NEXT: [[TMP19:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
@@ -60,11 +60,11 @@ define i32 @dotp_z_s(ptr %a, ptr %b) #0 {
; CHECK-NOI8MM-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NOI8MM-NEXT: entry:
; CHECK-NOI8MM-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = mul i64 [[TMP0]], 16
+; CHECK-NOI8MM-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 16
; CHECK-NOI8MM-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-NOI8MM: vector.ph:
-; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = mul i64 [[TMP14]], 16
+; CHECK-NOI8MM-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 16
; CHECK-NOI8MM-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
; CHECK-NOI8MM-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
; CHECK-NOI8MM-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
@@ -72,36 +72,36 @@ define i32 @dotp_z_s(ptr %a, ptr %b) #0 {
; CHECK-NOI8MM-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-NOI8MM: vector.body:
; CHECK-NOI8MM-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
-; CHECK-NOI8MM-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 0
+; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
+; CHECK-NOI8MM-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
; CHECK-NOI8MM-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 8
-; CHECK-NOI8MM-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP1]], i64 [[TMP9]]
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP2]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP9]]
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
-; CHECK-NOI8MM-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
+; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
+; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[TMP13]], i32 0
; CHECK-NOI8MM-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NOI8MM-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 8
-; CHECK-NOI8MM-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP16]]
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP13]], i64 [[TMP16]]
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP14]], align 1
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP17]], align 1
; CHECK-NOI8MM-NEXT: [[TMP18:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
; CHECK-NOI8MM-NEXT: [[TMP19:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
; CHECK-NOI8MM-NEXT: [[TMP20:%.*]] = mul <vscale x 8 x i32> [[TMP18]], [[TMP11]]
; CHECK-NOI8MM-NEXT: [[TMP21:%.*]] = mul <vscale x 8 x i32> [[TMP19]], [[TMP12]]
-; CHECK-NOI8MM-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 2 x i32> @llvm.experimental.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI]], <vscale x 8 x i32> [[TMP20]])
-; CHECK-NOI8MM-NEXT: [[PARTIAL_REDUCE5]] = call <vscale x 2 x i32> @llvm.experimental.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI1]], <vscale x 8 x i32> [[TMP21]])
+; CHECK-NOI8MM-NEXT: [[TMP22]] = add <vscale x 8 x i32> [[TMP20]], [[VEC_PHI]]
+; CHECK-NOI8MM-NEXT: [[TMP23]] = add <vscale x 8 x i32> [[TMP21]], [[VEC_PHI1]]
; CHECK-NOI8MM-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
; CHECK-NOI8MM-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NOI8MM-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK-NOI8MM: middle.block:
-; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 2 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
-; CHECK-NOI8MM-NEXT: [[TMP23:%.*]] = call i32 @llvm.vector.reduce.add.nxv2i32(<vscale x 2 x i32> [[BIN_RDX]])
+; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 8 x i32> [[TMP23]], [[TMP22]]
+; CHECK-NOI8MM-NEXT: [[TMP25:%.*]] = call i32 @llvm.vector.reduce.add.nxv8i32(<vscale x 8 x i32> [[BIN_RDX]])
; CHECK-NOI8MM-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
; CHECK-NOI8MM-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
; CHECK-NOI8MM: scalar.ph:
@@ -133,11 +133,11 @@ define i32 @dotp_s_z(ptr %a, ptr %b) #0 {
; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP13:%.*]] = mul i64 [[TMP0]], 16
+; CHECK-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 16
; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
-; CHECK-NEXT: [[TMP14:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP3:%.*]] = mul i64 [[TMP14]], 16
+; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 16
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
; CHECK-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
@@ -147,21 +147,21 @@ define i32 @dotp_s_z(ptr %a, ptr %b) #0 {
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
-; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 0
+; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
+; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
; CHECK-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 8
-; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP1]], i64 [[TMP9]]
-; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP2]], align 1
+; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP9]]
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
; CHECK-NEXT: [[TMP11:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
; CHECK-NEXT: [[TMP12:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
-; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
-; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
+; CHECK-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
+; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[TMP13]], i32 0
; CHECK-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 8
-; CHECK-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP16]]
-; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
+; CHECK-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP13]], i64 [[TMP16]]
+; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP14]], align 1
; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP17]], align 1
; CHECK-NEXT: [[TMP18:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
; CHECK-NEXT: [[TMP19:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
@@ -183,11 +183,11 @@ define i32 @dotp_s_z(ptr %a, ptr %b) #0 {
; CHECK-NOI8MM-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] {
; CHECK-NOI8MM-NEXT: entry:
; CHECK-NOI8MM-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = mul i64 [[TMP0]], 16
+; CHECK-NOI8MM-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 16
; CHECK-NOI8MM-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-NOI8MM: vector.ph:
-; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = mul i64 [[TMP14]], 16
+; CHECK-NOI8MM-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 16
; CHECK-NOI8MM-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
; CHECK-NOI8MM-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
; CHECK-NOI8MM-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
@@ -195,36 +195,36 @@ define i32 @dotp_s_z(ptr %a, ptr %b) #0 {
; CHECK-NOI8MM-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-NOI8MM: vector.body:
; CHECK-NOI8MM-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
-; CHECK-NOI8MM-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 0
+; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
+; CHECK-NOI8MM-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
; CHECK-NOI8MM-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 8
-; CHECK-NOI8MM-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP1]], i64 [[TMP9]]
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP2]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP9]]
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
-; CHECK-NOI8MM-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
+; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
+; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[TMP13]], i32 0
; CHECK-NOI8MM-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NOI8MM-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 8
-; CHECK-NOI8MM-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP16]]
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP13]], i64 [[TMP16]]
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP14]], align 1
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP17]], align 1
; CHECK-NOI8MM-NEXT: [[TMP18:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
; CHECK-NOI8MM-NEXT: [[TMP19:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
; CHECK-NOI8MM-NEXT: [[TMP20:%.*]] = mul <vscale x 8 x i32> [[TMP18]], [[TMP11]]
; CHECK-NOI8MM-NEXT: [[TMP21:%.*]] = mul <vscale x 8 x i32> [[TMP19]], [[TMP12]]
-; CHECK-NOI8MM-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 2 x i32> @llvm.experimental.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI]], <vscale x 8 x i32> [[TMP20]])
-; CHECK-NOI8MM-NEXT: [[PARTIAL_REDUCE5]] = call <vscale x 2 x i32> @llvm.experimental.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI1]], <vscale x 8 x i32> [[TMP21]])
+; CHECK-NOI8MM-NEXT: [[TMP22]] = add <vscale x 8 x i32> [[TMP20]], [[VEC_PHI]]
+; CHECK-NOI8MM-NEXT: [[TMP23]] = add <vscale x 8 x i32> [[TMP21]], [[VEC_PHI1]]
; CHECK-NOI8MM-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
; CHECK-NOI8MM-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NOI8MM-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK-NOI8MM: middle.block:
-; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 2 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
-; CHECK-NOI8MM-NEXT: [[TMP23:%.*]] = call i32 @llvm.vector.reduce.add.nxv2i32(<vscale x 2 x i32> [[BIN_RDX]])
+; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 8 x i32> [[TMP23]], [[TMP22]]
+; CHECK-NOI8MM-NEXT: [[TMP25:%.*]] = call i32 @llvm.vector.reduce.add.nxv8i32(<vscale x 8 x i32> [[BIN_RDX]])
; CHECK-NOI8MM-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
; CHECK-NOI8MM-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
; CHECK-NOI8MM: scalar.ph:
>From 41684b9643fd0b3d09c9a7b2ecfa703d8aaca0f9 Mon Sep 17 00:00:00 2001
From: Samuel Tebbs <samuel.tebbs at arm.com>
Date: Wed, 30 Apr 2025 15:21:15 +0100
Subject: [PATCH 5/6] Renamed tests
---
.../partial-reduce-dot-product-mixed.ll | 24 +++++++++----------
1 file changed, 12 insertions(+), 12 deletions(-)
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
index 04fe9ce96754a..eceff5ede34b3 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
@@ -5,8 +5,8 @@
target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
target triple = "aarch64-none-unknown-elf"
-define i32 @dotp_z_s(ptr %a, ptr %b) #0 {
-; CHECK-LABEL: define i32 @dotp_z_s(
+define i32 @sudot(ptr %a, ptr %b) #0 {
+; CHECK-LABEL: define i32 @sudot(
; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
@@ -56,7 +56,7 @@ define i32 @dotp_z_s(ptr %a, ptr %b) #0 {
; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
;
-; CHECK-NOI8MM-LABEL: define i32 @dotp_z_s(
+; CHECK-NOI8MM-LABEL: define i32 @sudot(
; CHECK-NOI8MM-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NOI8MM-NEXT: entry:
; CHECK-NOI8MM-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
@@ -128,8 +128,8 @@ for.exit: ; preds = %for.body
ret i32 %add
}
-define i32 @dotp_s_z(ptr %a, ptr %b) #0 {
-; CHECK-LABEL: define i32 @dotp_s_z(
+define i32 @usdot(ptr %a, ptr %b) #0 {
+; CHECK-LABEL: define i32 @usdot(
; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
@@ -179,7 +179,7 @@ define i32 @dotp_s_z(ptr %a, ptr %b) #0 {
; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
;
-; CHECK-NOI8MM-LABEL: define i32 @dotp_s_z(
+; CHECK-NOI8MM-LABEL: define i32 @usdot(
; CHECK-NOI8MM-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] {
; CHECK-NOI8MM-NEXT: entry:
; CHECK-NOI8MM-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
@@ -251,8 +251,8 @@ for.exit: ; preds = %for.body
ret i32 %add
}
-define i32 @dotp_z_s_neon(ptr %a, ptr %b) #1 {
-; CHECK-LABEL: define i32 @dotp_z_s_neon(
+define i32 @sudot_neon(ptr %a, ptr %b) #1 {
+; CHECK-LABEL: define i32 @sudot_neon(
; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR1:[0-9]+]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
@@ -289,7 +289,7 @@ define i32 @dotp_z_s_neon(ptr %a, ptr %b) #1 {
; CHECK-NEXT: br i1 true, label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
;
-; CHECK-NOI8MM-LABEL: define i32 @dotp_z_s_neon(
+; CHECK-NOI8MM-LABEL: define i32 @sudot_neon(
; CHECK-NOI8MM-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR1:[0-9]+]] {
; CHECK-NOI8MM-NEXT: entry:
; CHECK-NOI8MM-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
@@ -348,8 +348,8 @@ for.exit: ; preds = %for.body
ret i32 %add
}
-define i32 @dotp_s_z_neon(ptr %a, ptr %b) #1 {
-; CHECK-LABEL: define i32 @dotp_s_z_neon(
+define i32 @usdot_neon(ptr %a, ptr %b) #1 {
+; CHECK-LABEL: define i32 @usdot_neon(
; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR1]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
@@ -386,7 +386,7 @@ define i32 @dotp_s_z_neon(ptr %a, ptr %b) #1 {
; CHECK-NEXT: br i1 true, label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
;
-; CHECK-NOI8MM-LABEL: define i32 @dotp_s_z_neon(
+; CHECK-NOI8MM-LABEL: define i32 @usdot_neon(
; CHECK-NOI8MM-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR1]] {
; CHECK-NOI8MM-NEXT: entry:
; CHECK-NOI8MM-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
>From 486aa4ca815823000d9422b447947f2cabf4b50c Mon Sep 17 00:00:00 2001
From: Sam Tebbs <samuel.tebbs at arm.com>
Date: Thu, 1 May 2025 10:05:01 +0100
Subject: [PATCH 6/6] Remove --vectorizer-maximize-bandwidth from tests
---
.../partial-reduce-dot-product-mixed.ll | 202 ++++++++----------
1 file changed, 88 insertions(+), 114 deletions(-)
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
index eceff5ede34b3..74b358dd76976 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
@@ -1,6 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --filter-out-after "^scalar.ph:" --version 4
-; RUN: opt -passes=loop-vectorize -enable-epilogue-vectorization=false -mattr=+i8mm,+dotprod --vectorizer-maximize-bandwidth -S < %s | FileCheck %s
-; RUN: opt -passes=loop-vectorize -enable-epilogue-vectorization=false -mattr=+dotprod -S --vectorizer-maximize-bandwidth < %s | FileCheck %s --check-prefix=CHECK-NOI8MM
+; RUN: opt -passes=loop-vectorize -enable-epilogue-vectorization=false -mattr=+i8mm,+dotprod -S < %s | FileCheck %s
+; RUN: opt -passes=loop-vectorize -enable-epilogue-vectorization=false -mattr=+dotprod -S < %s | FileCheck %s --check-prefix=CHECK-NOI8MM
target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
target triple = "aarch64-none-unknown-elf"
@@ -9,99 +9,86 @@ define i32 @sudot(ptr %a, ptr %b) #0 {
; CHECK-LABEL: define i32 @sudot(
; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 16
; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
-; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 16
-; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
-; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
-; CHECK-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 16
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
-; CHECK-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 8
-; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP9]]
-; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
-; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
-; CHECK-NEXT: [[TMP11:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
-; CHECK-NEXT: [[TMP12:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
+; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP6]], i32 16
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
+; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
+; CHECK-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[TMP13]], i32 0
-; CHECK-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 8
-; CHECK-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP13]], i64 [[TMP16]]
-; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP14]], align 1
-; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP17]], align 1
-; CHECK-NEXT: [[TMP18:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
-; CHECK-NEXT: [[TMP19:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
-; CHECK-NEXT: [[TMP20:%.*]] = mul <vscale x 8 x i32> [[TMP18]], [[TMP11]]
-; CHECK-NEXT: [[TMP21:%.*]] = mul <vscale x 8 x i32> [[TMP19]], [[TMP12]]
-; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 2 x i32> @llvm.experimental.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI]], <vscale x 8 x i32> [[TMP20]])
-; CHECK-NEXT: [[PARTIAL_REDUCE5]] = call <vscale x 2 x i32> @llvm.experimental.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI1]], <vscale x 8 x i32> [[TMP21]])
-; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
-; CHECK-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK-NEXT: [[TMP15:%.*]] = getelementptr i8, ptr [[TMP13]], i32 16
+; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP14]], align 1
+; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP15]], align 1
+; CHECK-NEXT: [[TMP8:%.*]] = sext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>
+; CHECK-NEXT: [[TMP9:%.*]] = sext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
+; CHECK-NEXT: [[TMP10:%.*]] = mul <16 x i32> [[TMP8]], [[TMP3]]
+; CHECK-NEXT: [[TMP11:%.*]] = mul <16 x i32> [[TMP9]], [[TMP4]]
+; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.experimental.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP10]])
+; CHECK-NEXT: [[PARTIAL_REDUCE5]] = call <4 x i32> @llvm.experimental.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP11]])
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
+; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
+; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK: middle.block:
-; CHECK-NEXT: [[BIN_RDX:%.*]] = add <vscale x 2 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
-; CHECK-NEXT: [[TMP23:%.*]] = call i32 @llvm.vector.reduce.add.nxv2i32(<vscale x 2 x i32> [[BIN_RDX]])
-; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
-; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK-NEXT: [[BIN_RDX:%.*]] = add <4 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
+; CHECK-NEXT: [[TMP16:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[BIN_RDX]])
+; CHECK-NEXT: br i1 true, label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
;
; CHECK-NOI8MM-LABEL: define i32 @sudot(
; CHECK-NOI8MM-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NOI8MM-NEXT: entry:
; CHECK-NOI8MM-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 16
+; CHECK-NOI8MM-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 8
; CHECK-NOI8MM-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-NOI8MM: vector.ph:
; CHECK-NOI8MM-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 16
+; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 8
; CHECK-NOI8MM-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
; CHECK-NOI8MM-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
; CHECK-NOI8MM-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 16
+; CHECK-NOI8MM-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 8
; CHECK-NOI8MM-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-NOI8MM: vector.body:
; CHECK-NOI8MM-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[VECTOR_BODY]] ]
; CHECK-NOI8MM-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-NOI8MM-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
; CHECK-NOI8MM-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 8
+; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 4
; CHECK-NOI8MM-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP9]]
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
-; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i8>, ptr [[TMP7]], align 1
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 4 x i8>, ptr [[TMP10]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = zext <vscale x 4 x i8> [[WIDE_LOAD]] to <vscale x 4 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = zext <vscale x 4 x i8> [[WIDE_LOAD2]] to <vscale x 4 x i32>
; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[TMP13]], i32 0
; CHECK-NOI8MM-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 8
+; CHECK-NOI8MM-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 4
; CHECK-NOI8MM-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP13]], i64 [[TMP16]]
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP14]], align 1
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP17]], align 1
-; CHECK-NOI8MM-NEXT: [[TMP18:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP19:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP20:%.*]] = mul <vscale x 8 x i32> [[TMP18]], [[TMP11]]
-; CHECK-NOI8MM-NEXT: [[TMP21:%.*]] = mul <vscale x 8 x i32> [[TMP19]], [[TMP12]]
-; CHECK-NOI8MM-NEXT: [[TMP22]] = add <vscale x 8 x i32> [[TMP20]], [[VEC_PHI]]
-; CHECK-NOI8MM-NEXT: [[TMP23]] = add <vscale x 8 x i32> [[TMP21]], [[VEC_PHI1]]
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 4 x i8>, ptr [[TMP14]], align 1
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 4 x i8>, ptr [[TMP17]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP18:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD3]] to <vscale x 4 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP19:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD4]] to <vscale x 4 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP20:%.*]] = mul <vscale x 4 x i32> [[TMP18]], [[TMP11]]
+; CHECK-NOI8MM-NEXT: [[TMP21:%.*]] = mul <vscale x 4 x i32> [[TMP19]], [[TMP12]]
+; CHECK-NOI8MM-NEXT: [[TMP22]] = add <vscale x 4 x i32> [[TMP20]], [[VEC_PHI]]
+; CHECK-NOI8MM-NEXT: [[TMP23]] = add <vscale x 4 x i32> [[TMP21]], [[VEC_PHI1]]
; CHECK-NOI8MM-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
; CHECK-NOI8MM-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NOI8MM-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK-NOI8MM: middle.block:
-; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 8 x i32> [[TMP23]], [[TMP22]]
-; CHECK-NOI8MM-NEXT: [[TMP25:%.*]] = call i32 @llvm.vector.reduce.add.nxv8i32(<vscale x 8 x i32> [[BIN_RDX]])
+; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 4 x i32> [[TMP23]], [[TMP22]]
+; CHECK-NOI8MM-NEXT: [[TMP25:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[BIN_RDX]])
; CHECK-NOI8MM-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
; CHECK-NOI8MM-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
; CHECK-NOI8MM: scalar.ph:
@@ -132,99 +119,86 @@ define i32 @usdot(ptr %a, ptr %b) #0 {
; CHECK-LABEL: define i32 @usdot(
; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 16
; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
-; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 16
-; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
-; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
-; CHECK-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 16
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
-; CHECK-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 8
-; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP9]]
-; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
-; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
-; CHECK-NEXT: [[TMP11:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
-; CHECK-NEXT: [[TMP12:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
+; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP6]], i32 16
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
+; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
+; CHECK-NEXT: [[TMP3:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-NEXT: [[TMP4:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[TMP13]], i32 0
-; CHECK-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 8
-; CHECK-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP13]], i64 [[TMP16]]
-; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP14]], align 1
-; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP17]], align 1
-; CHECK-NEXT: [[TMP18:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
-; CHECK-NEXT: [[TMP19:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
-; CHECK-NEXT: [[TMP20:%.*]] = mul <vscale x 8 x i32> [[TMP18]], [[TMP11]]
-; CHECK-NEXT: [[TMP21:%.*]] = mul <vscale x 8 x i32> [[TMP19]], [[TMP12]]
-; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 2 x i32> @llvm.experimental.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI]], <vscale x 8 x i32> [[TMP20]])
-; CHECK-NEXT: [[PARTIAL_REDUCE5]] = call <vscale x 2 x i32> @llvm.experimental.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI1]], <vscale x 8 x i32> [[TMP21]])
-; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
-; CHECK-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK-NEXT: [[TMP15:%.*]] = getelementptr i8, ptr [[TMP13]], i32 16
+; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP14]], align 1
+; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP15]], align 1
+; CHECK-NEXT: [[TMP8:%.*]] = zext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>
+; CHECK-NEXT: [[TMP9:%.*]] = zext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
+; CHECK-NEXT: [[TMP10:%.*]] = mul <16 x i32> [[TMP8]], [[TMP3]]
+; CHECK-NEXT: [[TMP11:%.*]] = mul <16 x i32> [[TMP9]], [[TMP4]]
+; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.experimental.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP10]])
+; CHECK-NEXT: [[PARTIAL_REDUCE5]] = call <4 x i32> @llvm.experimental.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP11]])
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
+; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
+; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK: middle.block:
-; CHECK-NEXT: [[BIN_RDX:%.*]] = add <vscale x 2 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
-; CHECK-NEXT: [[TMP23:%.*]] = call i32 @llvm.vector.reduce.add.nxv2i32(<vscale x 2 x i32> [[BIN_RDX]])
-; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
-; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK-NEXT: [[BIN_RDX:%.*]] = add <4 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
+; CHECK-NEXT: [[TMP16:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[BIN_RDX]])
+; CHECK-NEXT: br i1 true, label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
;
; CHECK-NOI8MM-LABEL: define i32 @usdot(
; CHECK-NOI8MM-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] {
; CHECK-NOI8MM-NEXT: entry:
; CHECK-NOI8MM-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 16
+; CHECK-NOI8MM-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 8
; CHECK-NOI8MM-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-NOI8MM: vector.ph:
; CHECK-NOI8MM-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 16
+; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 8
; CHECK-NOI8MM-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
; CHECK-NOI8MM-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
; CHECK-NOI8MM-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 16
+; CHECK-NOI8MM-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 8
; CHECK-NOI8MM-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-NOI8MM: vector.body:
; CHECK-NOI8MM-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[VECTOR_BODY]] ]
; CHECK-NOI8MM-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-NOI8MM-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
; CHECK-NOI8MM-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 8
+; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 4
; CHECK-NOI8MM-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP9]]
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
-; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i8>, ptr [[TMP7]], align 1
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 4 x i8>, ptr [[TMP10]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD]] to <vscale x 4 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD2]] to <vscale x 4 x i32>
; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[TMP13]], i32 0
; CHECK-NOI8MM-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NOI8MM-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 8
+; CHECK-NOI8MM-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 4
; CHECK-NOI8MM-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP13]], i64 [[TMP16]]
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP14]], align 1
-; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP17]], align 1
-; CHECK-NOI8MM-NEXT: [[TMP18:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP19:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP20:%.*]] = mul <vscale x 8 x i32> [[TMP18]], [[TMP11]]
-; CHECK-NOI8MM-NEXT: [[TMP21:%.*]] = mul <vscale x 8 x i32> [[TMP19]], [[TMP12]]
-; CHECK-NOI8MM-NEXT: [[TMP22]] = add <vscale x 8 x i32> [[TMP20]], [[VEC_PHI]]
-; CHECK-NOI8MM-NEXT: [[TMP23]] = add <vscale x 8 x i32> [[TMP21]], [[VEC_PHI1]]
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 4 x i8>, ptr [[TMP14]], align 1
+; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 4 x i8>, ptr [[TMP17]], align 1
+; CHECK-NOI8MM-NEXT: [[TMP18:%.*]] = zext <vscale x 4 x i8> [[WIDE_LOAD3]] to <vscale x 4 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP19:%.*]] = zext <vscale x 4 x i8> [[WIDE_LOAD4]] to <vscale x 4 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP20:%.*]] = mul <vscale x 4 x i32> [[TMP18]], [[TMP11]]
+; CHECK-NOI8MM-NEXT: [[TMP21:%.*]] = mul <vscale x 4 x i32> [[TMP19]], [[TMP12]]
+; CHECK-NOI8MM-NEXT: [[TMP22]] = add <vscale x 4 x i32> [[TMP20]], [[VEC_PHI]]
+; CHECK-NOI8MM-NEXT: [[TMP23]] = add <vscale x 4 x i32> [[TMP21]], [[VEC_PHI1]]
; CHECK-NOI8MM-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
; CHECK-NOI8MM-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NOI8MM-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK-NOI8MM: middle.block:
-; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 8 x i32> [[TMP23]], [[TMP22]]
-; CHECK-NOI8MM-NEXT: [[TMP25:%.*]] = call i32 @llvm.vector.reduce.add.nxv8i32(<vscale x 8 x i32> [[BIN_RDX]])
+; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 4 x i32> [[TMP23]], [[TMP22]]
+; CHECK-NOI8MM-NEXT: [[TMP25:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[BIN_RDX]])
; CHECK-NOI8MM-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
; CHECK-NOI8MM-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
; CHECK-NOI8MM: scalar.ph:
More information about the llvm-commits
mailing list