Skip to content

Commit 7fc0024

Browse files
SamTebbs33SamTebbs33
committed
Removed isFixed check
1 parent 7fe67b2 commit 7fc0024

File tree

2 files changed

+57
-57
lines changed

2 files changed

+57
-57
lines changed

‎llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -5347,7 +5347,7 @@ InstructionCost AArch64TTIImpl::getPartialReductionCost(
53475347
// AArch64 supports lowering mixed fixed-width extensions to a usdot but only
53485348
// if the i8mm feature is available.
53495349
if (OpAExtend == TTI::PR_None || OpBExtend == TTI::PR_None ||
5350-
(OpAExtend != OpBExtend && VF.isFixed() && !ST->hasMatMulInt8()))
5350+
(OpAExtend != OpBExtend && !ST->hasMatMulInt8()))
53515351
return Invalid;
53525352

53535353
if (!BinOp || *BinOp != Instruction::Mul)

‎llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll‎

Lines changed: 56 additions & 56 deletions
Original file line numberDiff line numberDiff line change
@@ -10,11 +10,11 @@ define i32 @dotp_z_s(ptr %a, ptr %b) #0 {
1010
; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0:[0-9]+]] {
1111
; CHECK-NEXT: entry:
1212
; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
13-
; CHECK-NEXT: [[TMP13:%.*]] = mul i64 [[TMP0]], 16
13+
; CHECK-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 16
1414
; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
1515
; CHECK: vector.ph:
16-
; CHECK-NEXT: [[TMP14:%.*]] = call i64 @llvm.vscale.i64()
17-
; CHECK-NEXT: [[TMP3:%.*]] = mul i64 [[TMP14]], 16
16+
; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
17+
; CHECK-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 16
1818
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
1919
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
2020
; CHECK-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
@@ -24,21 +24,21 @@ define i32 @dotp_z_s(ptr %a, ptr %b) #0 {
2424
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
2525
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
2626
; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
27-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
28-
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 0
27+
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
28+
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
2929
; CHECK-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
3030
; CHECK-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 8
31-
; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP1]], i64 [[TMP9]]
32-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP2]], align 1
31+
; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP9]]
32+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
3333
; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
3434
; CHECK-NEXT: [[TMP11:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
3535
; CHECK-NEXT: [[TMP12:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
36-
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
37-
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
36+
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
37+
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[TMP13]], i32 0
3838
; CHECK-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
3939
; CHECK-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 8
40-
; CHECK-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP16]]
41-
; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
40+
; CHECK-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP13]], i64 [[TMP16]]
41+
; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP14]], align 1
4242
; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP17]], align 1
4343
; CHECK-NEXT: [[TMP18:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
4444
; CHECK-NEXT: [[TMP19:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
@@ -60,48 +60,48 @@ define i32 @dotp_z_s(ptr %a, ptr %b) #0 {
6060
; CHECK-NOI8MM-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0:[0-9]+]] {
6161
; CHECK-NOI8MM-NEXT: entry:
6262
; CHECK-NOI8MM-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
63-
; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = mul i64 [[TMP0]], 16
63+
; CHECK-NOI8MM-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 16
6464
; CHECK-NOI8MM-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
6565
; CHECK-NOI8MM: vector.ph:
66-
; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = call i64 @llvm.vscale.i64()
67-
; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = mul i64 [[TMP14]], 16
66+
; CHECK-NOI8MM-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
67+
; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 16
6868
; CHECK-NOI8MM-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
6969
; CHECK-NOI8MM-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
7070
; CHECK-NOI8MM-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
7171
; CHECK-NOI8MM-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 16
7272
; CHECK-NOI8MM-NEXT: br label [[VECTOR_BODY:%.*]]
7373
; CHECK-NOI8MM: vector.body:
7474
; CHECK-NOI8MM-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
75-
; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
76-
; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
77-
; CHECK-NOI8MM-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
78-
; CHECK-NOI8MM-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 0
75+
; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[VECTOR_BODY]] ]
76+
; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[VECTOR_BODY]] ]
77+
; CHECK-NOI8MM-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
78+
; CHECK-NOI8MM-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
7979
; CHECK-NOI8MM-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
8080
; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 8
81-
; CHECK-NOI8MM-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP1]], i64 [[TMP9]]
82-
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP2]], align 1
81+
; CHECK-NOI8MM-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP9]]
82+
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
8383
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
8484
; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
8585
; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
86-
; CHECK-NOI8MM-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
87-
; CHECK-NOI8MM-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
86+
; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
87+
; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[TMP13]], i32 0
8888
; CHECK-NOI8MM-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
8989
; CHECK-NOI8MM-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 8
90-
; CHECK-NOI8MM-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP16]]
91-
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
90+
; CHECK-NOI8MM-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP13]], i64 [[TMP16]]
91+
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP14]], align 1
9292
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP17]], align 1
9393
; CHECK-NOI8MM-NEXT: [[TMP18:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
9494
; CHECK-NOI8MM-NEXT: [[TMP19:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
9595
; CHECK-NOI8MM-NEXT: [[TMP20:%.*]] = mul <vscale x 8 x i32> [[TMP18]], [[TMP11]]
9696
; CHECK-NOI8MM-NEXT: [[TMP21:%.*]] = mul <vscale x 8 x i32> [[TMP19]], [[TMP12]]
97-
; 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]])
98-
; 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]])
97+
; CHECK-NOI8MM-NEXT: [[TMP22]] = add <vscale x 8 x i32> [[TMP20]], [[VEC_PHI]]
98+
; CHECK-NOI8MM-NEXT: [[TMP23]] = add <vscale x 8 x i32> [[TMP21]], [[VEC_PHI1]]
9999
; CHECK-NOI8MM-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
100100
; CHECK-NOI8MM-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
101101
; CHECK-NOI8MM-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
102102
; CHECK-NOI8MM: middle.block:
103-
; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 2 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
104-
; CHECK-NOI8MM-NEXT: [[TMP23:%.*]] = call i32 @llvm.vector.reduce.add.nxv2i32(<vscale x 2 x i32> [[BIN_RDX]])
103+
; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 8 x i32> [[TMP23]], [[TMP22]]
104+
; CHECK-NOI8MM-NEXT: [[TMP25:%.*]] = call i32 @llvm.vector.reduce.add.nxv8i32(<vscale x 8 x i32> [[BIN_RDX]])
105105
; CHECK-NOI8MM-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
106106
; CHECK-NOI8MM-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
107107
; CHECK-NOI8MM: scalar.ph:
@@ -133,11 +133,11 @@ define i32 @dotp_s_z(ptr %a, ptr %b) #0 {
133133
; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] {
134134
; CHECK-NEXT: entry:
135135
; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
136-
; CHECK-NEXT: [[TMP13:%.*]] = mul i64 [[TMP0]], 16
136+
; CHECK-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 16
137137
; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
138138
; CHECK: vector.ph:
139-
; CHECK-NEXT: [[TMP14:%.*]] = call i64 @llvm.vscale.i64()
140-
; CHECK-NEXT: [[TMP3:%.*]] = mul i64 [[TMP14]], 16
139+
; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
140+
; CHECK-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 16
141141
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
142142
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
143143
; CHECK-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
@@ -147,21 +147,21 @@ define i32 @dotp_s_z(ptr %a, ptr %b) #0 {
147147
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
148148
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
149149
; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
150-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
151-
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 0
150+
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
151+
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
152152
; CHECK-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
153153
; CHECK-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 8
154-
; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP1]], i64 [[TMP9]]
155-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP2]], align 1
154+
; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP9]]
155+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
156156
; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
157157
; CHECK-NEXT: [[TMP11:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
158158
; CHECK-NEXT: [[TMP12:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
159-
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
160-
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
159+
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
160+
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[TMP13]], i32 0
161161
; CHECK-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
162162
; CHECK-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 8
163-
; CHECK-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP16]]
164-
; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
163+
; CHECK-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP13]], i64 [[TMP16]]
164+
; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP14]], align 1
165165
; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP17]], align 1
166166
; CHECK-NEXT: [[TMP18:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
167167
; CHECK-NEXT: [[TMP19:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
@@ -183,48 +183,48 @@ define i32 @dotp_s_z(ptr %a, ptr %b) #0 {
183183
; CHECK-NOI8MM-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] {
184184
; CHECK-NOI8MM-NEXT: entry:
185185
; CHECK-NOI8MM-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
186-
; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = mul i64 [[TMP0]], 16
186+
; CHECK-NOI8MM-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 16
187187
; CHECK-NOI8MM-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
188188
; CHECK-NOI8MM: vector.ph:
189-
; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = call i64 @llvm.vscale.i64()
190-
; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = mul i64 [[TMP14]], 16
189+
; CHECK-NOI8MM-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
190+
; CHECK-NOI8MM-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 16
191191
; CHECK-NOI8MM-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
192192
; CHECK-NOI8MM-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
193193
; CHECK-NOI8MM-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
194194
; CHECK-NOI8MM-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 16
195195
; CHECK-NOI8MM-NEXT: br label [[VECTOR_BODY:%.*]]
196196
; CHECK-NOI8MM: vector.body:
197197
; CHECK-NOI8MM-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
198-
; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
199-
; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
200-
; CHECK-NOI8MM-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
201-
; CHECK-NOI8MM-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 0
198+
; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[VECTOR_BODY]] ]
199+
; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[VECTOR_BODY]] ]
200+
; CHECK-NOI8MM-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
201+
; CHECK-NOI8MM-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
202202
; CHECK-NOI8MM-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
203203
; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 8
204-
; CHECK-NOI8MM-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP1]], i64 [[TMP9]]
205-
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP2]], align 1
204+
; CHECK-NOI8MM-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP9]]
205+
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
206206
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
207207
; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
208208
; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
209-
; CHECK-NOI8MM-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
210-
; CHECK-NOI8MM-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
209+
; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
210+
; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[TMP13]], i32 0
211211
; CHECK-NOI8MM-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
212212
; CHECK-NOI8MM-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 8
213-
; CHECK-NOI8MM-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP16]]
214-
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
213+
; CHECK-NOI8MM-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP13]], i64 [[TMP16]]
214+
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP14]], align 1
215215
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP17]], align 1
216216
; CHECK-NOI8MM-NEXT: [[TMP18:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
217217
; CHECK-NOI8MM-NEXT: [[TMP19:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
218218
; CHECK-NOI8MM-NEXT: [[TMP20:%.*]] = mul <vscale x 8 x i32> [[TMP18]], [[TMP11]]
219219
; CHECK-NOI8MM-NEXT: [[TMP21:%.*]] = mul <vscale x 8 x i32> [[TMP19]], [[TMP12]]
220-
; 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]])
221-
; 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]])
220+
; CHECK-NOI8MM-NEXT: [[TMP22]] = add <vscale x 8 x i32> [[TMP20]], [[VEC_PHI]]
221+
; CHECK-NOI8MM-NEXT: [[TMP23]] = add <vscale x 8 x i32> [[TMP21]], [[VEC_PHI1]]
222222
; CHECK-NOI8MM-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
223223
; CHECK-NOI8MM-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
224224
; CHECK-NOI8MM-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
225225
; CHECK-NOI8MM: middle.block:
226-
; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 2 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
227-
; CHECK-NOI8MM-NEXT: [[TMP23:%.*]] = call i32 @llvm.vector.reduce.add.nxv2i32(<vscale x 2 x i32> [[BIN_RDX]])
226+
; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 8 x i32> [[TMP23]], [[TMP22]]
227+
; CHECK-NOI8MM-NEXT: [[TMP25:%.*]] = call i32 @llvm.vector.reduce.add.nxv8i32(<vscale x 8 x i32> [[BIN_RDX]])
228228
; CHECK-NOI8MM-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
229229
; CHECK-NOI8MM-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
230230
; CHECK-NOI8MM: scalar.ph:

0 commit comments

Comments
 (0)