[x86] fix wrong maskload/store intrinsic definitions in avxintrin.h (follow up of D13776).

[andreadb]

Summary

Hi,

This patch is a follow up of D13776.
According to the Intel documentation, the mask operand of a maskload and maskstore intrinsics is always a vector of packed integer/long integer values.
This patch introduces the following two changes:

1. It fixes the avx maskload/store intrinsic definitions in avxintrin.h;
2. It changes BuiltinsX86.def to match the correct gcc definitions for avx maskload/store (see D13776 for more details).

Please let me know if okay to submit.

Thanks,
Andrea

[delena]

LGTM

[andreadb]

In D13861#270811, @delena wrote:

LGTM

Thanks Elena!
Committed revision250816.

cfe/trunk/include/clang/Basic/BuiltinsX86.def

 TARGET_BUILTIN(__builtin_ia32_vbroadcastf128_ps256, "V8fV4fC*", "", "avx")
 TARGET_BUILTIN(__builtin_ia32_storeupd256, "vd*V4d", "", "avx")
 TARGET_BUILTIN(__builtin_ia32_storeups256, "vf*V8f", "", "avx")
 TARGET_BUILTIN(__builtin_ia32_storedqu256, "vc*V32c", "", "avx")
 TARGET_BUILTIN(__builtin_ia32_lddqu256, "V32ccC*", "", "avx")
 TARGET_BUILTIN(__builtin_ia32_movntdq256, "vV4LLi*V4LLi", "", "avx")
 TARGET_BUILTIN(__builtin_ia32_movntpd256, "vd*V4d", "", "avx")
 TARGET_BUILTIN(__builtin_ia32_movntps256, "vf*V8f", "", "avx")
-TARGET_BUILTIN(__builtin_ia32_maskloadpd, "V2dV2dC*V2d", "", "avx")
-TARGET_BUILTIN(__builtin_ia32_maskloadps, "V4fV4fC*V4f", "", "avx")
-TARGET_BUILTIN(__builtin_ia32_maskloadpd256, "V4dV4dC*V4d", "", "avx")
-TARGET_BUILTIN(__builtin_ia32_maskloadps256, "V8fV8fC*V8f", "", "avx")
-TARGET_BUILTIN(__builtin_ia32_maskstorepd, "vV2d*V2dV2d", "", "avx")
-TARGET_BUILTIN(__builtin_ia32_maskstoreps, "vV4f*V4fV4f", "", "avx")
-TARGET_BUILTIN(__builtin_ia32_maskstorepd256, "vV4d*V4dV4d", "", "avx")
-TARGET_BUILTIN(__builtin_ia32_maskstoreps256, "vV8f*V8fV8f", "", "avx")
+TARGET_BUILTIN(__builtin_ia32_maskloadpd, "V2dV2dC*V2LLi", "", "avx")
+TARGET_BUILTIN(__builtin_ia32_maskloadps, "V4fV4fC*V4i", "", "avx")
+TARGET_BUILTIN(__builtin_ia32_maskloadpd256, "V4dV4dC*V4LLi", "", "avx")
+TARGET_BUILTIN(__builtin_ia32_maskloadps256, "V8fV8fC*V8i", "", "avx")
+TARGET_BUILTIN(__builtin_ia32_maskstorepd, "vV2d*V2LLiV2d", "", "avx")
+TARGET_BUILTIN(__builtin_ia32_maskstoreps, "vV4f*V4iV4f", "", "avx")
+TARGET_BUILTIN(__builtin_ia32_maskstorepd256, "vV4d*V4LLiV4d", "", "avx")
+TARGET_BUILTIN(__builtin_ia32_maskstoreps256, "vV8f*V8iV8f", "", "avx")
 
 // AVX2
 TARGET_BUILTIN(__builtin_ia32_mpsadbw256, "V32cV32cV32cIc", "", "avx2")
 TARGET_BUILTIN(__builtin_ia32_pabsb256, "V32cV32c", "", "avx2")
 TARGET_BUILTIN(__builtin_ia32_pabsw256, "V16sV16s", "", "avx2")
 TARGET_BUILTIN(__builtin_ia32_pabsd256, "V8iV8i", "", "avx2")
 TARGET_BUILTIN(__builtin_ia32_packsswb256, "V32cV16sV16s", "", "avx2")
 TARGET_BUILTIN(__builtin_ia32_packssdw256, "V16sV8iV8i", "", "avx2")

cfe/trunk/lib/Headers/avxintrin.h

 static __inline void __DEFAULT_FN_ATTRS
 _mm256_storeu_si256(__m256i *__p, __m256i __a)
 {
   __builtin_ia32_storedqu256((char *)__p, (__v32qi)__a);
 }
 
 /* Conditional load ops */
 static __inline __m128d __DEFAULT_FN_ATTRS
-_mm_maskload_pd(double const *__p, __m128d __m)
+_mm_maskload_pd(double const *__p, __m128i __m)
 {
-  return (__m128d)__builtin_ia32_maskloadpd((const __v2df *)__p, (__v2df)__m);
+  return (__m128d)__builtin_ia32_maskloadpd((const __v2df *)__p, (__v2di)__m);
 }
 
 static __inline __m256d __DEFAULT_FN_ATTRS
-_mm256_maskload_pd(double const *__p, __m256d __m)
+_mm256_maskload_pd(double const *__p, __m256i __m)
 {
   return (__m256d)__builtin_ia32_maskloadpd256((const __v4df *)__p,
-                                               (__v4df)__m);
+                                               (__v4di)__m);
 }
 
 static __inline __m128 __DEFAULT_FN_ATTRS
-_mm_maskload_ps(float const *__p, __m128 __m)
+_mm_maskload_ps(float const *__p, __m128i __m)
 {
-  return (__m128)__builtin_ia32_maskloadps((const __v4sf *)__p, (__v4sf)__m);
+  return (__m128)__builtin_ia32_maskloadps((const __v4sf *)__p, (__v4si)__m);
 }
 
 static __inline __m256 __DEFAULT_FN_ATTRS
-_mm256_maskload_ps(float const *__p, __m256 __m)
+_mm256_maskload_ps(float const *__p, __m256i __m)
 {
-  return (__m256)__builtin_ia32_maskloadps256((const __v8sf *)__p, (__v8sf)__m);
+  return (__m256)__builtin_ia32_maskloadps256((const __v8sf *)__p, (__v8si)__m);
 }
 
 /* Conditional store ops */
 static __inline void __DEFAULT_FN_ATTRS
-_mm256_maskstore_ps(float *__p, __m256 __m, __m256 __a)
+_mm256_maskstore_ps(float *__p, __m256i __m, __m256 __a)
 {
-  __builtin_ia32_maskstoreps256((__v8sf *)__p, (__v8sf)__m, (__v8sf)__a);
+  __builtin_ia32_maskstoreps256((__v8sf *)__p, (__v8si)__m, (__v8sf)__a);
 }
 
 static __inline void __DEFAULT_FN_ATTRS
-_mm_maskstore_pd(double *__p, __m128d __m, __m128d __a)
+_mm_maskstore_pd(double *__p, __m128i __m, __m128d __a)
 {
-  __builtin_ia32_maskstorepd((__v2df *)__p, (__v2df)__m, (__v2df)__a);
+  __builtin_ia32_maskstorepd((__v2df *)__p, (__v2di)__m, (__v2df)__a);
 }
 
 static __inline void __DEFAULT_FN_ATTRS
-_mm256_maskstore_pd(double *__p, __m256d __m, __m256d __a)
+_mm256_maskstore_pd(double *__p, __m256i __m, __m256d __a)
 {
-  __builtin_ia32_maskstorepd256((__v4df *)__p, (__v4df)__m, (__v4df)__a);
+  __builtin_ia32_maskstorepd256((__v4df *)__p, (__v4di)__m, (__v4df)__a);
 }
 
 static __inline void __DEFAULT_FN_ATTRS
-_mm_maskstore_ps(float *__p, __m128 __m, __m128 __a)
+_mm_maskstore_ps(float *__p, __m128i __m, __m128 __a)
 {
-  __builtin_ia32_maskstoreps((__v4sf *)__p, (__v4sf)__m, (__v4sf)__a);
+  __builtin_ia32_maskstoreps((__v4sf *)__p, (__v4si)__m, (__v4sf)__a);
 }
 
 /* Cacheability support ops */
 static __inline void __DEFAULT_FN_ATTRS
 _mm256_stream_si256(__m256i *__a, __m256i __b)
 {
   __builtin_ia32_movntdq256((__v4di *)__a, (__v4di)__b);
 }

cfe/trunk/test/CodeGen/builtins-x86.c

#endif
   tmp_V8f = __builtin_ia32_vbroadcastf128_ps256(tmp_V4fCp);
   __builtin_ia32_storeupd256(tmp_dp, tmp_V4d);
   __builtin_ia32_storeups256(tmp_fp, tmp_V8f);
   __builtin_ia32_storedqu256(tmp_cp, tmp_V32c);
   tmp_V32c = __builtin_ia32_lddqu256(tmp_cCp);
   __builtin_ia32_movntdq256(tmp_V4LLip, tmp_V4LLi);
   __builtin_ia32_movntpd256(tmp_dp, tmp_V4d);
   __builtin_ia32_movntps256(tmp_fp, tmp_V8f);
-  tmp_V2d = __builtin_ia32_maskloadpd(tmp_V2dCp, tmp_V2d);
-  tmp_V4f = __builtin_ia32_maskloadps(tmp_V4fCp, tmp_V4f);
-  tmp_V4d = __builtin_ia32_maskloadpd256(tmp_V4dCp, tmp_V4d);
-  tmp_V8f = __builtin_ia32_maskloadps256(tmp_V8fCp, tmp_V8f);
-  __builtin_ia32_maskstorepd(tmp_V2dp, tmp_V2d, tmp_V2d);
-  __builtin_ia32_maskstoreps(tmp_V4fp, tmp_V4f, tmp_V4f);
-  __builtin_ia32_maskstorepd256(tmp_V4dp, tmp_V4d, tmp_V4d);
-  __builtin_ia32_maskstoreps256(tmp_V8fp, tmp_V8f, tmp_V8f);
+  tmp_V2d = __builtin_ia32_maskloadpd(tmp_V2dCp, tmp_V2LLi);
+  tmp_V4f = __builtin_ia32_maskloadps(tmp_V4fCp, tmp_V4i);
+  tmp_V4d = __builtin_ia32_maskloadpd256(tmp_V4dCp, tmp_V4LLi);
+  tmp_V8f = __builtin_ia32_maskloadps256(tmp_V8fCp, tmp_V8i);
+  __builtin_ia32_maskstorepd(tmp_V2dp, tmp_V2LLi, tmp_V2d);
+  __builtin_ia32_maskstoreps(tmp_V4fp, tmp_V4i, tmp_V4f);
+  __builtin_ia32_maskstorepd256(tmp_V4dp, tmp_V4LLi, tmp_V4d);
+  __builtin_ia32_maskstoreps256(tmp_V8fp, tmp_V8i, tmp_V8f);
 
 #ifdef USE_3DNOW
   tmp_V8c = __builtin_ia32_pavgusb(tmp_V8c, tmp_V8c);
   tmp_V2i = __builtin_ia32_pf2id(tmp_V2f);
   tmp_V2f = __builtin_ia32_pfacc(tmp_V2f, tmp_V2f);
   tmp_V2f = __builtin_ia32_pfadd(tmp_V2f, tmp_V2f);
   tmp_V2i = __builtin_ia32_pfcmpeq(tmp_V2f, tmp_V2f);
   tmp_V2i = __builtin_ia32_pfcmpge(tmp_V2f, tmp_V2f);