Skip to content

Commit a105877

Browse files
nikicnikic
authored
[InstCombine] Remove some of the complexity-based canonicalization (#91185)
The idea behind this canonicalization is that it allows us to handle less patterns, because we know that some will be canonicalized away. This is indeed very useful to e.g. know that constants are always on the right. However, this is only useful if the canonicalization is actually reliable. This is the case for constants, but not for arguments: Moving these to the right makes it look like the "more complex" expression is guaranteed to be on the left, but this is not actually the case in practice. It fails as soon as you replace the argument with another instruction. The end result is that it looks like things correctly work in tests, while they actually don't. We use the "thwart complexity-based canonicalization" trick to handle this in tests, but it's often a challenge for new contributors to get this right, and based on the regressions this PR originally exposed, we clearly don't get this right in many cases. For this reason, I think that it's better to remove this complexity canonicalization. It will make it much easier to write tests for commuted cases and make sure that they are handled.
1 parent 297bb46 commit a105877

File tree

289 files changed

+2680
-2681
lines changed

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

289 files changed

+2680
-2681
lines changed

‎clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld1-bfloat.c

+2-2
Original file line numberDiff line numberDiff line change
@@ -44,7 +44,7 @@ svbfloat16_t test_svld1_bf16(svbool_t pg, const bfloat16_t *base) MODE_ATTR
4444
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
4545
// CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
4646
// CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 4
47-
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
47+
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
4848
// CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
4949
// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.masked.load.nxv8bf16.p0(ptr [[TMP3]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x bfloat> zeroinitializer)
5050
// CHECK-NEXT: ret <vscale x 8 x bfloat> [[TMP4]]
@@ -54,7 +54,7 @@ svbfloat16_t test_svld1_bf16(svbool_t pg, const bfloat16_t *base) MODE_ATTR
5454
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
5555
// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
5656
// CPP-CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 4
57-
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
57+
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
5858
// CPP-CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
5959
// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.masked.load.nxv8bf16.p0(ptr [[TMP3]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x bfloat> zeroinitializer)
6060
// CPP-CHECK-NEXT: ret <vscale x 8 x bfloat> [[TMP4]]

‎clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld1.c

+22-22
Large diffs are not rendered by default.

‎clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld1sb.c

+13-13
Original file line numberDiff line numberDiff line change
@@ -141,7 +141,7 @@ svuint64_t test_svld1sb_u64(svbool_t pg, const int8_t *base) MODE_ATTR
141141
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
142142
// CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
143143
// CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 3
144-
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
144+
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
145145
// CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
146146
// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x i8> @llvm.masked.load.nxv8i8.p0(ptr [[TMP3]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x i8> zeroinitializer)
147147
// CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 8 x i8> [[TMP4]] to <vscale x 8 x i16>
@@ -152,7 +152,7 @@ svuint64_t test_svld1sb_u64(svbool_t pg, const int8_t *base) MODE_ATTR
152152
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
153153
// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
154154
// CPP-CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 3
155-
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
155+
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
156156
// CPP-CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
157157
// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x i8> @llvm.masked.load.nxv8i8.p0(ptr [[TMP3]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x i8> zeroinitializer)
158158
// CPP-CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 8 x i8> [[TMP4]] to <vscale x 8 x i16>
@@ -168,7 +168,7 @@ svint16_t test_svld1sb_vnum_s16(svbool_t pg, const int8_t *base, int64_t vnum) M
168168
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
169169
// CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
170170
// CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 2
171-
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
171+
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
172172
// CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
173173
// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 4 x i8> @llvm.masked.load.nxv4i8.p0(ptr [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i8> zeroinitializer)
174174
// CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 4 x i8> [[TMP4]] to <vscale x 4 x i32>
@@ -179,7 +179,7 @@ svint16_t test_svld1sb_vnum_s16(svbool_t pg, const int8_t *base, int64_t vnum) M
179179
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
180180
// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
181181
// CPP-CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 2
182-
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
182+
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
183183
// CPP-CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
184184
// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 4 x i8> @llvm.masked.load.nxv4i8.p0(ptr [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i8> zeroinitializer)
185185
// CPP-CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 4 x i8> [[TMP4]] to <vscale x 4 x i32>
@@ -195,7 +195,7 @@ svint32_t test_svld1sb_vnum_s32(svbool_t pg, const int8_t *base, int64_t vnum) M
195195
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
196196
// CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
197197
// CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 1
198-
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
198+
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
199199
// CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
200200
// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 2 x i8> @llvm.masked.load.nxv2i8.p0(ptr [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i8> zeroinitializer)
201201
// CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 2 x i8> [[TMP4]] to <vscale x 2 x i64>
@@ -206,7 +206,7 @@ svint32_t test_svld1sb_vnum_s32(svbool_t pg, const int8_t *base, int64_t vnum) M
206206
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
207207
// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
208208
// CPP-CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 1
209-
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
209+
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
210210
// CPP-CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
211211
// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 2 x i8> @llvm.masked.load.nxv2i8.p0(ptr [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i8> zeroinitializer)
212212
// CPP-CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 2 x i8> [[TMP4]] to <vscale x 2 x i64>
@@ -222,7 +222,7 @@ svint64_t test_svld1sb_vnum_s64(svbool_t pg, const int8_t *base, int64_t vnum) M
222222
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
223223
// CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
224224
// CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 3
225-
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
225+
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
226226
// CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
227227
// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x i8> @llvm.masked.load.nxv8i8.p0(ptr [[TMP3]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x i8> zeroinitializer)
228228
// CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 8 x i8> [[TMP4]] to <vscale x 8 x i16>
@@ -233,7 +233,7 @@ svint64_t test_svld1sb_vnum_s64(svbool_t pg, const int8_t *base, int64_t vnum) M
233233
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
234234
// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
235235
// CPP-CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 3
236-
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
236+
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
237237
// CPP-CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
238238
// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x i8> @llvm.masked.load.nxv8i8.p0(ptr [[TMP3]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x i8> zeroinitializer)
239239
// CPP-CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 8 x i8> [[TMP4]] to <vscale x 8 x i16>
@@ -249,7 +249,7 @@ svuint16_t test_svld1sb_vnum_u16(svbool_t pg, const int8_t *base, int64_t vnum)
249249
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
250250
// CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
251251
// CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 2
252-
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
252+
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
253253
// CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
254254
// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 4 x i8> @llvm.masked.load.nxv4i8.p0(ptr [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i8> zeroinitializer)
255255
// CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 4 x i8> [[TMP4]] to <vscale x 4 x i32>
@@ -260,7 +260,7 @@ svuint16_t test_svld1sb_vnum_u16(svbool_t pg, const int8_t *base, int64_t vnum)
260260
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
261261
// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
262262
// CPP-CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 2
263-
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
263+
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
264264
// CPP-CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
265265
// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 4 x i8> @llvm.masked.load.nxv4i8.p0(ptr [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i8> zeroinitializer)
266266
// CPP-CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 4 x i8> [[TMP4]] to <vscale x 4 x i32>
@@ -276,7 +276,7 @@ svuint32_t test_svld1sb_vnum_u32(svbool_t pg, const int8_t *base, int64_t vnum)
276276
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
277277
// CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
278278
// CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 1
279-
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
279+
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
280280
// CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
281281
// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 2 x i8> @llvm.masked.load.nxv2i8.p0(ptr [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i8> zeroinitializer)
282282
// CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 2 x i8> [[TMP4]] to <vscale x 2 x i64>
@@ -287,7 +287,7 @@ svuint32_t test_svld1sb_vnum_u32(svbool_t pg, const int8_t *base, int64_t vnum)
287287
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
288288
// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
289289
// CPP-CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 1
290-
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
290+
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
291291
// CPP-CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
292292
// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 2 x i8> @llvm.masked.load.nxv2i8.p0(ptr [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i8> zeroinitializer)
293293
// CPP-CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 2 x i8> [[TMP4]] to <vscale x 2 x i64>
@@ -298,7 +298,7 @@ svuint64_t test_svld1sb_vnum_u64(svbool_t pg, const int8_t *base, int64_t vnum)
298298
return svld1sb_vnum_u64(pg, base, vnum);
299299
}
300300

301-
#ifndef __ARM_FEATURE_SME
301+
#ifndef __ARM_FEATURE_SME
302302

303303
// CHECK-LABEL: @test_svld1sb_gather_u32base_s32(
304304
// CHECK-NEXT: entry:

‎clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld1sh.c

+9-9
Original file line numberDiff line numberDiff line change
@@ -103,7 +103,7 @@ svuint64_t test_svld1sh_u64(svbool_t pg, const int16_t *base) MODE_ATTR
103103
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
104104
// CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
105105
// CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 3
106-
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
106+
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
107107
// CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
108108
// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 4 x i16> @llvm.masked.load.nxv4i16.p0(ptr [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i16> zeroinitializer)
109109
// CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 4 x i16> [[TMP4]] to <vscale x 4 x i32>
@@ -114,7 +114,7 @@ svuint64_t test_svld1sh_u64(svbool_t pg, const int16_t *base) MODE_ATTR
114114
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
115115
// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
116116
// CPP-CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 3
117-
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
117+
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
118118
// CPP-CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
119119
// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 4 x i16> @llvm.masked.load.nxv4i16.p0(ptr [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i16> zeroinitializer)
120120
// CPP-CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 4 x i16> [[TMP4]] to <vscale x 4 x i32>
@@ -130,7 +130,7 @@ svint32_t test_svld1sh_vnum_s32(svbool_t pg, const int16_t *base, int64_t vnum)
130130
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
131131
// CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
132132
// CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 2
133-
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
133+
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
134134
// CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
135135
// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 2 x i16> @llvm.masked.load.nxv2i16.p0(ptr [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i16> zeroinitializer)
136136
// CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 2 x i16> [[TMP4]] to <vscale x 2 x i64>
@@ -141,7 +141,7 @@ svint32_t test_svld1sh_vnum_s32(svbool_t pg, const int16_t *base, int64_t vnum)
141141
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
142142
// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
143143
// CPP-CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 2
144-
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
144+
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
145145
// CPP-CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
146146
// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 2 x i16> @llvm.masked.load.nxv2i16.p0(ptr [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i16> zeroinitializer)
147147
// CPP-CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 2 x i16> [[TMP4]] to <vscale x 2 x i64>
@@ -157,7 +157,7 @@ svint64_t test_svld1sh_vnum_s64(svbool_t pg, const int16_t *base, int64_t vnum)
157157
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
158158
// CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
159159
// CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 3
160-
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
160+
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
161161
// CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
162162
// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 4 x i16> @llvm.masked.load.nxv4i16.p0(ptr [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i16> zeroinitializer)
163163
// CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 4 x i16> [[TMP4]] to <vscale x 4 x i32>
@@ -168,7 +168,7 @@ svint64_t test_svld1sh_vnum_s64(svbool_t pg, const int16_t *base, int64_t vnum)
168168
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
169169
// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
170170
// CPP-CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 3
171-
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
171+
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
172172
// CPP-CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
173173
// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 4 x i16> @llvm.masked.load.nxv4i16.p0(ptr [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i16> zeroinitializer)
174174
// CPP-CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 4 x i16> [[TMP4]] to <vscale x 4 x i32>
@@ -184,7 +184,7 @@ svuint32_t test_svld1sh_vnum_u32(svbool_t pg, const int16_t *base, int64_t vnum)
184184
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
185185
// CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
186186
// CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 2
187-
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
187+
// CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
188188
// CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
189189
// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 2 x i16> @llvm.masked.load.nxv2i16.p0(ptr [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i16> zeroinitializer)
190190
// CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 2 x i16> [[TMP4]] to <vscale x 2 x i64>
@@ -195,7 +195,7 @@ svuint32_t test_svld1sh_vnum_u32(svbool_t pg, const int16_t *base, int64_t vnum)
195195
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
196196
// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
197197
// CPP-CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 2
198-
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[TMP2]], [[VNUM:%.*]]
198+
// CPP-CHECK-NEXT: [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
199199
// CPP-CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
200200
// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 2 x i16> @llvm.masked.load.nxv2i16.p0(ptr [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i16> zeroinitializer)
201201
// CPP-CHECK-NEXT: [[TMP5:%.*]] = sext <vscale x 2 x i16> [[TMP4]] to <vscale x 2 x i64>
@@ -206,7 +206,7 @@ svuint64_t test_svld1sh_vnum_u64(svbool_t pg, const int16_t *base, int64_t vnum)
206206
return svld1sh_vnum_u64(pg, base, vnum);
207207
}
208208

209-
#ifndef __ARM_FEATURE_SME
209+
#ifndef __ARM_FEATURE_SME
210210

211211
// CHECK-LABEL: @test_svld1sh_gather_u32base_s32(
212212
// CHECK-NEXT: entry:

0 commit comments

Comments
 (0)