Lines Matching refs:SLM

3 … -mtriple=x86_64-unknown -mcpu=slm -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefixes=SLM
48 ; SLM-LABEL: @smul_v8i64(
49 ; SLM-NEXT:    [[TMP1:%.*]] = load <2 x i64>, <2 x i64>* bitcast ([8 x i64]* @a64 to <2 x i64>*), a…
50 ; SLM-NEXT:    [[TMP2:%.*]] = load <2 x i64>, <2 x i64>* bitcast ([8 x i64]* @b64 to <2 x i64>*), a…
51 ; SLM-NEXT:    [[TMP3:%.*]] = call <2 x i64> @llvm.smul.fix.v2i64(<2 x i64> [[TMP1]], <2 x i64> [[T…
52 ; SLM-NEXT:    store <2 x i64> [[TMP3]], <2 x i64>* bitcast ([8 x i64]* @c64 to <2 x i64>*), align 8
53 ; SLM-NEXT:    [[TMP4:%.*]] = load <2 x i64>, <2 x i64>* bitcast (i64* getelementptr inbounds ([8 x…
54 ; SLM-NEXT:    [[TMP5:%.*]] = load <2 x i64>, <2 x i64>* bitcast (i64* getelementptr inbounds ([8 x…
55 ; SLM-NEXT:    [[TMP6:%.*]] = call <2 x i64> @llvm.smul.fix.v2i64(<2 x i64> [[TMP4]], <2 x i64> [[T…
56 ; SLM-NEXT:    store <2 x i64> [[TMP6]], <2 x i64>* bitcast (i64* getelementptr inbounds ([8 x i64]…
57 ; SLM-NEXT:    [[TMP7:%.*]] = load <2 x i64>, <2 x i64>* bitcast (i64* getelementptr inbounds ([8 x…
58 ; SLM-NEXT:    [[TMP8:%.*]] = load <2 x i64>, <2 x i64>* bitcast (i64* getelementptr inbounds ([8 x…
59 ; SLM-NEXT:    [[TMP9:%.*]] = call <2 x i64> @llvm.smul.fix.v2i64(<2 x i64> [[TMP7]], <2 x i64> [[T…
60 ; SLM-NEXT:    store <2 x i64> [[TMP9]], <2 x i64>* bitcast (i64* getelementptr inbounds ([8 x i64]…
61 ; SLM-NEXT:    [[TMP10:%.*]] = load <2 x i64>, <2 x i64>* bitcast (i64* getelementptr inbounds ([8 …
62 ; SLM-NEXT:    [[TMP11:%.*]] = load <2 x i64>, <2 x i64>* bitcast (i64* getelementptr inbounds ([8 …
63 ; SLM-NEXT:    [[TMP12:%.*]] = call <2 x i64> @llvm.smul.fix.v2i64(<2 x i64> [[TMP10]], <2 x i64> […
64 ; SLM-NEXT:    store <2 x i64> [[TMP12]], <2 x i64>* bitcast (i64* getelementptr inbounds ([8 x i64…
65 ; SLM-NEXT:    ret void
170 ; SLM-LABEL: @smul_v16i32(
171 ; SLM-NEXT:    [[A0:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
172 ; SLM-NEXT:    [[A1:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
173 ; SLM-NEXT:    [[A2:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
174 ; SLM-NEXT:    [[A3:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
175 ; SLM-NEXT:    [[A4:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
176 ; SLM-NEXT:    [[A5:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
177 ; SLM-NEXT:    [[A6:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
178 ; SLM-NEXT:    [[A7:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
179 ; SLM-NEXT:    [[A8:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
180 ; SLM-NEXT:    [[A9:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
181 ; SLM-NEXT:    [[A10:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i…
182 ; SLM-NEXT:    [[A11:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i…
183 ; SLM-NEXT:    [[A12:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i…
184 ; SLM-NEXT:    [[A13:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i…
185 ; SLM-NEXT:    [[A14:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i…
186 ; SLM-NEXT:    [[A15:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i…
187 ; SLM-NEXT:    [[B0:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
188 ; SLM-NEXT:    [[B1:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
189 ; SLM-NEXT:    [[B2:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
190 ; SLM-NEXT:    [[B3:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
191 ; SLM-NEXT:    [[B4:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
192 ; SLM-NEXT:    [[B5:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
193 ; SLM-NEXT:    [[B6:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
194 ; SLM-NEXT:    [[B7:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
195 ; SLM-NEXT:    [[B8:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
196 ; SLM-NEXT:    [[B9:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
197 ; SLM-NEXT:    [[B10:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i…
198 ; SLM-NEXT:    [[B11:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i…
199 ; SLM-NEXT:    [[B12:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i…
200 ; SLM-NEXT:    [[B13:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i…
201 ; SLM-NEXT:    [[B14:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i…
202 ; SLM-NEXT:    [[B15:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i…
203 ; SLM-NEXT:    [[R0:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A0]], i32 [[B0]], i32 3)
204 ; SLM-NEXT:    [[R1:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A1]], i32 [[B1]], i32 3)
205 ; SLM-NEXT:    [[R2:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A2]], i32 [[B2]], i32 3)
206 ; SLM-NEXT:    [[R3:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A3]], i32 [[B3]], i32 3)
207 ; SLM-NEXT:    [[R4:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A4]], i32 [[B4]], i32 3)
208 ; SLM-NEXT:    [[R5:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A5]], i32 [[B5]], i32 3)
209 ; SLM-NEXT:    [[R6:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A6]], i32 [[B6]], i32 3)
210 ; SLM-NEXT:    [[R7:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A7]], i32 [[B7]], i32 3)
211 ; SLM-NEXT:    [[R8:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A8]], i32 [[B8]], i32 3)
212 ; SLM-NEXT:    [[R9:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A9]], i32 [[B9]], i32 3)
213 ; SLM-NEXT:    [[R10:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A10]], i32 [[B10]], i32 3)
214 ; SLM-NEXT:    [[R11:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A11]], i32 [[B11]], i32 3)
215 ; SLM-NEXT:    [[R12:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A12]], i32 [[B12]], i32 3)
216 ; SLM-NEXT:    [[R13:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A13]], i32 [[B13]], i32 3)
217 ; SLM-NEXT:    [[R14:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A14]], i32 [[B14]], i32 3)
218 ; SLM-NEXT:    [[R15:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A15]], i32 [[B15]], i32 3)
219 ; SLM-NEXT:    store i32 [[R0]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
220 ; SLM-NEXT:    store i32 [[R1]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
221 ; SLM-NEXT:    store i32 [[R2]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
222 ; SLM-NEXT:    store i32 [[R3]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
223 ; SLM-NEXT:    store i32 [[R4]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
224 ; SLM-NEXT:    store i32 [[R5]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
225 ; SLM-NEXT:    store i32 [[R6]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
226 ; SLM-NEXT:    store i32 [[R7]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
227 ; SLM-NEXT:    store i32 [[R8]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
228 ; SLM-NEXT:    store i32 [[R9]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
229 ; SLM-NEXT:    store i32 [[R10]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0,…
230 ; SLM-NEXT:    store i32 [[R11]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0,…
231 ; SLM-NEXT:    store i32 [[R12]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0,…
232 ; SLM-NEXT:    store i32 [[R13]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0,…
233 ; SLM-NEXT:    store i32 [[R14]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0,…
234 ; SLM-NEXT:    store i32 [[R15]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0,…
235 ; SLM-NEXT:    ret void
342 ; SLM-LABEL: @smul_v32i16(
343 ; SLM-NEXT:    [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* bitcast ([32 x i16]* @a16 to <8 x i16>*), …
344 ; SLM-NEXT:    [[TMP2:%.*]] = load <8 x i16>, <8 x i16>* bitcast ([32 x i16]* @b16 to <8 x i16>*), …
345 ; SLM-NEXT:    [[TMP3:%.*]] = call <8 x i16> @llvm.smul.fix.v8i16(<8 x i16> [[TMP1]], <8 x i16> [[T…
346 ; SLM-NEXT:    store <8 x i16> [[TMP3]], <8 x i16>* bitcast ([32 x i16]* @c16 to <8 x i16>*), align…
347 ; SLM-NEXT:    [[TMP4:%.*]] = load <8 x i16>, <8 x i16>* bitcast (i16* getelementptr inbounds ([32 …
348 ; SLM-NEXT:    [[TMP5:%.*]] = load <8 x i16>, <8 x i16>* bitcast (i16* getelementptr inbounds ([32 …
349 ; SLM-NEXT:    [[TMP6:%.*]] = call <8 x i16> @llvm.smul.fix.v8i16(<8 x i16> [[TMP4]], <8 x i16> [[T…
350 ; SLM-NEXT:    store <8 x i16> [[TMP6]], <8 x i16>* bitcast (i16* getelementptr inbounds ([32 x i16…
351 ; SLM-NEXT:    [[TMP7:%.*]] = load <8 x i16>, <8 x i16>* bitcast (i16* getelementptr inbounds ([32 …
352 ; SLM-NEXT:    [[TMP8:%.*]] = load <8 x i16>, <8 x i16>* bitcast (i16* getelementptr inbounds ([32 …
353 ; SLM-NEXT:    [[TMP9:%.*]] = call <8 x i16> @llvm.smul.fix.v8i16(<8 x i16> [[TMP7]], <8 x i16> [[T…
354 ; SLM-NEXT:    store <8 x i16> [[TMP9]], <8 x i16>* bitcast (i16* getelementptr inbounds ([32 x i16…
355 ; SLM-NEXT:    [[TMP10:%.*]] = load <8 x i16>, <8 x i16>* bitcast (i16* getelementptr inbounds ([32…
356 ; SLM-NEXT:    [[TMP11:%.*]] = load <8 x i16>, <8 x i16>* bitcast (i16* getelementptr inbounds ([32…
357 ; SLM-NEXT:    [[TMP12:%.*]] = call <8 x i16> @llvm.smul.fix.v8i16(<8 x i16> [[TMP10]], <8 x i16> […
358 ; SLM-NEXT:    store <8 x i16> [[TMP12]], <8 x i16>* bitcast (i16* getelementptr inbounds ([32 x i1…
359 ; SLM-NEXT:    ret void
530 ; SLM-LABEL: @smul_v64i8(
531 ; SLM-NEXT:    [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* bitcast ([64 x i8]* @a8 to <16 x i8>*), al…
532 ; SLM-NEXT:    [[TMP2:%.*]] = load <16 x i8>, <16 x i8>* bitcast ([64 x i8]* @b8 to <16 x i8>*), al…
533 ; SLM-NEXT:    [[TMP3:%.*]] = call <16 x i8> @llvm.smul.fix.v16i8(<16 x i8> [[TMP1]], <16 x i8> [[T…
534 ; SLM-NEXT:    store <16 x i8> [[TMP3]], <16 x i8>* bitcast ([64 x i8]* @c8 to <16 x i8>*), align 1
535 ; SLM-NEXT:    [[TMP4:%.*]] = load <16 x i8>, <16 x i8>* bitcast (i8* getelementptr inbounds ([64 x…
536 ; SLM-NEXT:    [[TMP5:%.*]] = load <16 x i8>, <16 x i8>* bitcast (i8* getelementptr inbounds ([64 x…
537 ; SLM-NEXT:    [[TMP6:%.*]] = call <16 x i8> @llvm.smul.fix.v16i8(<16 x i8> [[TMP4]], <16 x i8> [[T…
538 ; SLM-NEXT:    store <16 x i8> [[TMP6]], <16 x i8>* bitcast (i8* getelementptr inbounds ([64 x i8],…
539 ; SLM-NEXT:    [[TMP7:%.*]] = load <16 x i8>, <16 x i8>* bitcast (i8* getelementptr inbounds ([64 x…
540 ; SLM-NEXT:    [[TMP8:%.*]] = load <16 x i8>, <16 x i8>* bitcast (i8* getelementptr inbounds ([64 x…
541 ; SLM-NEXT:    [[TMP9:%.*]] = call <16 x i8> @llvm.smul.fix.v16i8(<16 x i8> [[TMP7]], <16 x i8> [[T…
542 ; SLM-NEXT:    [[TMP10:%.*]] = load <16 x i8>, <16 x i8>* bitcast (i8* getelementptr inbounds ([64 …
543 ; SLM-NEXT:    [[TMP11:%.*]] = load <16 x i8>, <16 x i8>* bitcast (i8* getelementptr inbounds ([64 …
544 ; SLM-NEXT:    [[TMP12:%.*]] = call <16 x i8> @llvm.smul.fix.v16i8(<16 x i8> [[TMP10]], <16 x i8> […
545 ; SLM-NEXT:    store <16 x i8> [[TMP9]], <16 x i8>* bitcast (i8* getelementptr inbounds ([64 x i8],…
546 ; SLM-NEXT:    store <16 x i8> [[TMP12]], <16 x i8>* bitcast (i8* getelementptr inbounds ([64 x i8]…
547 ; SLM-NEXT:    ret void
851 ; SLM-LABEL: @umul_v8i64(
852 ; SLM-NEXT:    [[TMP1:%.*]] = load <2 x i64>, <2 x i64>* bitcast ([8 x i64]* @a64 to <2 x i64>*), a…
853 ; SLM-NEXT:    [[TMP2:%.*]] = load <2 x i64>, <2 x i64>* bitcast ([8 x i64]* @b64 to <2 x i64>*), a…
854 ; SLM-NEXT:    [[TMP3:%.*]] = call <2 x i64> @llvm.umul.fix.v2i64(<2 x i64> [[TMP1]], <2 x i64> [[T…
855 ; SLM-NEXT:    store <2 x i64> [[TMP3]], <2 x i64>* bitcast ([8 x i64]* @c64 to <2 x i64>*), align 8
856 ; SLM-NEXT:    [[TMP4:%.*]] = load <2 x i64>, <2 x i64>* bitcast (i64* getelementptr inbounds ([8 x…
857 ; SLM-NEXT:    [[TMP5:%.*]] = load <2 x i64>, <2 x i64>* bitcast (i64* getelementptr inbounds ([8 x…
858 ; SLM-NEXT:    [[TMP6:%.*]] = call <2 x i64> @llvm.umul.fix.v2i64(<2 x i64> [[TMP4]], <2 x i64> [[T…
859 ; SLM-NEXT:    store <2 x i64> [[TMP6]], <2 x i64>* bitcast (i64* getelementptr inbounds ([8 x i64]…
860 ; SLM-NEXT:    [[TMP7:%.*]] = load <2 x i64>, <2 x i64>* bitcast (i64* getelementptr inbounds ([8 x…
861 ; SLM-NEXT:    [[TMP8:%.*]] = load <2 x i64>, <2 x i64>* bitcast (i64* getelementptr inbounds ([8 x…
862 ; SLM-NEXT:    [[TMP9:%.*]] = call <2 x i64> @llvm.umul.fix.v2i64(<2 x i64> [[TMP7]], <2 x i64> [[T…
863 ; SLM-NEXT:    store <2 x i64> [[TMP9]], <2 x i64>* bitcast (i64* getelementptr inbounds ([8 x i64]…
864 ; SLM-NEXT:    [[TMP10:%.*]] = load <2 x i64>, <2 x i64>* bitcast (i64* getelementptr inbounds ([8 …
865 ; SLM-NEXT:    [[TMP11:%.*]] = load <2 x i64>, <2 x i64>* bitcast (i64* getelementptr inbounds ([8 …
866 ; SLM-NEXT:    [[TMP12:%.*]] = call <2 x i64> @llvm.umul.fix.v2i64(<2 x i64> [[TMP10]], <2 x i64> […
867 ; SLM-NEXT:    store <2 x i64> [[TMP12]], <2 x i64>* bitcast (i64* getelementptr inbounds ([8 x i64…
868 ; SLM-NEXT:    ret void
973 ; SLM-LABEL: @umul_v16i32(
974 ; SLM-NEXT:    [[A0:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
975 ; SLM-NEXT:    [[A1:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
976 ; SLM-NEXT:    [[A2:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
977 ; SLM-NEXT:    [[A3:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
978 ; SLM-NEXT:    [[A4:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
979 ; SLM-NEXT:    [[A5:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
980 ; SLM-NEXT:    [[A6:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
981 ; SLM-NEXT:    [[A7:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
982 ; SLM-NEXT:    [[A8:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
983 ; SLM-NEXT:    [[A9:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i3…
984 ; SLM-NEXT:    [[A10:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i…
985 ; SLM-NEXT:    [[A11:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i…
986 ; SLM-NEXT:    [[A12:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i…
987 ; SLM-NEXT:    [[A13:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i…
988 ; SLM-NEXT:    [[A14:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i…
989 ; SLM-NEXT:    [[A15:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i…
990 ; SLM-NEXT:    [[B0:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
991 ; SLM-NEXT:    [[B1:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
992 ; SLM-NEXT:    [[B2:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
993 ; SLM-NEXT:    [[B3:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
994 ; SLM-NEXT:    [[B4:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
995 ; SLM-NEXT:    [[B5:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
996 ; SLM-NEXT:    [[B6:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
997 ; SLM-NEXT:    [[B7:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
998 ; SLM-NEXT:    [[B8:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
999 ; SLM-NEXT:    [[B9:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i3…
1000 ; SLM-NEXT:    [[B10:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i…
1001 ; SLM-NEXT:    [[B11:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i…
1002 ; SLM-NEXT:    [[B12:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i…
1003 ; SLM-NEXT:    [[B13:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i…
1004 ; SLM-NEXT:    [[B14:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i…
1005 ; SLM-NEXT:    [[B15:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i…
1006 ; SLM-NEXT:    [[R0:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A0]], i32 [[B0]], i32 3)
1007 ; SLM-NEXT:    [[R1:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A1]], i32 [[B1]], i32 3)
1008 ; SLM-NEXT:    [[R2:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A2]], i32 [[B2]], i32 3)
1009 ; SLM-NEXT:    [[R3:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A3]], i32 [[B3]], i32 3)
1010 ; SLM-NEXT:    [[R4:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A4]], i32 [[B4]], i32 3)
1011 ; SLM-NEXT:    [[R5:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A5]], i32 [[B5]], i32 3)
1012 ; SLM-NEXT:    [[R6:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A6]], i32 [[B6]], i32 3)
1013 ; SLM-NEXT:    [[R7:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A7]], i32 [[B7]], i32 3)
1014 ; SLM-NEXT:    [[R8:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A8]], i32 [[B8]], i32 3)
1015 ; SLM-NEXT:    [[R9:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A9]], i32 [[B9]], i32 3)
1016 ; SLM-NEXT:    [[R10:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A10]], i32 [[B10]], i32 3)
1017 ; SLM-NEXT:    [[R11:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A11]], i32 [[B11]], i32 3)
1018 ; SLM-NEXT:    [[R12:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A12]], i32 [[B12]], i32 3)
1019 ; SLM-NEXT:    [[R13:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A13]], i32 [[B13]], i32 3)
1020 ; SLM-NEXT:    [[R14:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A14]], i32 [[B14]], i32 3)
1021 ; SLM-NEXT:    [[R15:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A15]], i32 [[B15]], i32 3)
1022 ; SLM-NEXT:    store i32 [[R0]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
1023 ; SLM-NEXT:    store i32 [[R1]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
1024 ; SLM-NEXT:    store i32 [[R2]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
1025 ; SLM-NEXT:    store i32 [[R3]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
1026 ; SLM-NEXT:    store i32 [[R4]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
1027 ; SLM-NEXT:    store i32 [[R5]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
1028 ; SLM-NEXT:    store i32 [[R6]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
1029 ; SLM-NEXT:    store i32 [[R7]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
1030 ; SLM-NEXT:    store i32 [[R8]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
1031 ; SLM-NEXT:    store i32 [[R9]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, …
1032 ; SLM-NEXT:    store i32 [[R10]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0,…
1033 ; SLM-NEXT:    store i32 [[R11]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0,…
1034 ; SLM-NEXT:    store i32 [[R12]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0,…
1035 ; SLM-NEXT:    store i32 [[R13]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0,…
1036 ; SLM-NEXT:    store i32 [[R14]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0,…
1037 ; SLM-NEXT:    store i32 [[R15]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0,…
1038 ; SLM-NEXT:    ret void
1145 ; SLM-LABEL: @umul_v32i16(
1146 ; SLM-NEXT:    [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* bitcast ([32 x i16]* @a16 to <8 x i16>*), …
1147 ; SLM-NEXT:    [[TMP2:%.*]] = load <8 x i16>, <8 x i16>* bitcast ([32 x i16]* @b16 to <8 x i16>*), …
1148 ; SLM-NEXT:    [[TMP3:%.*]] = call <8 x i16> @llvm.umul.fix.v8i16(<8 x i16> [[TMP1]], <8 x i16> [[T…
1149 ; SLM-NEXT:    store <8 x i16> [[TMP3]], <8 x i16>* bitcast ([32 x i16]* @c16 to <8 x i16>*), align…
1150 ; SLM-NEXT:    [[TMP4:%.*]] = load <8 x i16>, <8 x i16>* bitcast (i16* getelementptr inbounds ([32 …
1151 ; SLM-NEXT:    [[TMP5:%.*]] = load <8 x i16>, <8 x i16>* bitcast (i16* getelementptr inbounds ([32 …
1152 ; SLM-NEXT:    [[TMP6:%.*]] = call <8 x i16> @llvm.umul.fix.v8i16(<8 x i16> [[TMP4]], <8 x i16> [[T…
1153 ; SLM-NEXT:    store <8 x i16> [[TMP6]], <8 x i16>* bitcast (i16* getelementptr inbounds ([32 x i16…
1154 ; SLM-NEXT:    [[TMP7:%.*]] = load <8 x i16>, <8 x i16>* bitcast (i16* getelementptr inbounds ([32 …
1155 ; SLM-NEXT:    [[TMP8:%.*]] = load <8 x i16>, <8 x i16>* bitcast (i16* getelementptr inbounds ([32 …
1156 ; SLM-NEXT:    [[TMP9:%.*]] = call <8 x i16> @llvm.umul.fix.v8i16(<8 x i16> [[TMP7]], <8 x i16> [[T…
1157 ; SLM-NEXT:    store <8 x i16> [[TMP9]], <8 x i16>* bitcast (i16* getelementptr inbounds ([32 x i16…
1158 ; SLM-NEXT:    [[TMP10:%.*]] = load <8 x i16>, <8 x i16>* bitcast (i16* getelementptr inbounds ([32…
1159 ; SLM-NEXT:    [[TMP11:%.*]] = load <8 x i16>, <8 x i16>* bitcast (i16* getelementptr inbounds ([32…
1160 ; SLM-NEXT:    [[TMP12:%.*]] = call <8 x i16> @llvm.umul.fix.v8i16(<8 x i16> [[TMP10]], <8 x i16> […
1161 ; SLM-NEXT:    store <8 x i16> [[TMP12]], <8 x i16>* bitcast (i16* getelementptr inbounds ([32 x i1…
1162 ; SLM-NEXT:    ret void
1333 ; SLM-LABEL: @umul_v64i8(
1334 ; SLM-NEXT:    [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* bitcast ([64 x i8]* @a8 to <16 x i8>*), al…
1335 ; SLM-NEXT:    [[TMP2:%.*]] = load <16 x i8>, <16 x i8>* bitcast ([64 x i8]* @b8 to <16 x i8>*), al…
1336 ; SLM-NEXT:    [[TMP3:%.*]] = call <16 x i8> @llvm.umul.fix.v16i8(<16 x i8> [[TMP1]], <16 x i8> [[T…
1337 ; SLM-NEXT:    store <16 x i8> [[TMP3]], <16 x i8>* bitcast ([64 x i8]* @c8 to <16 x i8>*), align 1
1338 ; SLM-NEXT:    [[TMP4:%.*]] = load <16 x i8>, <16 x i8>* bitcast (i8* getelementptr inbounds ([64 x…
1339 ; SLM-NEXT:    [[TMP5:%.*]] = load <16 x i8>, <16 x i8>* bitcast (i8* getelementptr inbounds ([64 x…
1340 ; SLM-NEXT:    [[TMP6:%.*]] = call <16 x i8> @llvm.umul.fix.v16i8(<16 x i8> [[TMP4]], <16 x i8> [[T…
1341 ; SLM-NEXT:    store <16 x i8> [[TMP6]], <16 x i8>* bitcast (i8* getelementptr inbounds ([64 x i8],…
1342 ; SLM-NEXT:    [[TMP7:%.*]] = load <16 x i8>, <16 x i8>* bitcast (i8* getelementptr inbounds ([64 x…
1343 ; SLM-NEXT:    [[TMP8:%.*]] = load <16 x i8>, <16 x i8>* bitcast (i8* getelementptr inbounds ([64 x…
1344 ; SLM-NEXT:    [[TMP9:%.*]] = call <16 x i8> @llvm.umul.fix.v16i8(<16 x i8> [[TMP7]], <16 x i8> [[T…
1345 ; SLM-NEXT:    [[TMP10:%.*]] = load <16 x i8>, <16 x i8>* bitcast (i8* getelementptr inbounds ([64 …
1346 ; SLM-NEXT:    [[TMP11:%.*]] = load <16 x i8>, <16 x i8>* bitcast (i8* getelementptr inbounds ([64 …
1347 ; SLM-NEXT:    [[TMP12:%.*]] = call <16 x i8> @llvm.umul.fix.v16i8(<16 x i8> [[TMP10]], <16 x i8> […
1348 ; SLM-NEXT:    store <16 x i8> [[TMP9]], <16 x i8>* bitcast (i8* getelementptr inbounds ([64 x i8],…
1349 ; SLM-NEXT:    store <16 x i8> [[TMP12]], <16 x i8>* bitcast (i8* getelementptr inbounds ([64 x i8]…
1350 ; SLM-NEXT:    ret void