Hi!
As the testcases show, the macros we have for -O0 for intrinsics that require
constant argument(s) should first cast the argument to the type the -O1+
inline uses and afterwards to whatever type e.g. a builtin needs.
The PR reported one which violated this, and I've grepped for all double-casts
and grepped out from that meaningful casts where the __m{128,256,512}{,d,i}
first cast is cast to same sized __v* type and has the same kind of element
type (float, double, integral). These 7 macros were using different casts,
and I've double checked them against the inline function types.
Bootstrapped/regtested on x86_64-linux and i686-linux, committed to trunk
and 9.3 as obvious, queued for backporting to 8.5.
2020-03-05 Jakub Jelinek <ja...@redhat.com>
PR target/94046
* config/i386/avx2intrin.h (_mm_mask_i32gather_ps): Fix first cast of
SRC and MASK arguments to __m128 from __m128d.
(_mm256_mask_i32gather_ps): Fix first cast of MASK argument to __m256
from __m256d.
(_mm_mask_i64gather_ps): Fix first cast of MASK argument to __m128
from __m128d.
* config/i386/xopintrin.h (_mm_permute2_pd): Fix first cast of C
argument to __m128i from __m128d.
(_mm256_permute2_pd): Fix first cast of C argument to __m256i from
__m256d.
(_mm_permute2_ps): Fix first cast of C argument to __m128i from __m128.
(_mm256_permute2_ps): Fix first cast of C argument to __m256i from
__m256.
* g++.target/i386/pr94046-1.C: New test.
* g++.target/i386/pr94046-2.C: New test.
--- gcc/config/i386/avx2intrin.h.jj 2020-01-12 11:54:36.313414917 +0100
+++ gcc/config/i386/avx2intrin.h 2020-03-05 15:22:12.684531786 +0100
@@ -1736,10 +1736,10 @@ _mm256_mask_i64gather_epi32 (__m128i __s
(int)SCALE)
#define _mm_mask_i32gather_ps(SRC, BASE, INDEX, MASK, SCALE) \
- (__m128) __builtin_ia32_gathersiv4sf ((__v4sf)(__m128d)SRC, \
+ (__m128) __builtin_ia32_gathersiv4sf ((__v4sf)(__m128)SRC, \
(float const *)BASE, \
(__v4si)(__m128i)INDEX, \
- (__v4sf)(__m128d)MASK, \
+ (__v4sf)(__m128)MASK, \
(int)SCALE)
#define _mm256_i32gather_ps(BASE, INDEX, SCALE) \
@@ -1754,7 +1754,7 @@ _mm256_mask_i64gather_epi32 (__m128i __s
(__m256) __builtin_ia32_gathersiv8sf ((__v8sf)(__m256)SRC, \
(float const *)BASE, \
(__v8si)(__m256i)INDEX, \
- (__v8sf)(__m256d)MASK, \
+ (__v8sf)(__m256)MASK, \
(int)SCALE)
#define _mm_i64gather_ps(BASE, INDEX, SCALE) \
@@ -1769,7 +1769,7 @@ _mm256_mask_i64gather_epi32 (__m128i __s
(__m128) __builtin_ia32_gatherdiv4sf ((__v4sf)(__m128)SRC, \
(float const *)BASE, \
(__v2di)(__m128i)INDEX, \
- (__v4sf)(__m128d)MASK, \
+ (__v4sf)(__m128)MASK, \
(int)SCALE)
#define _mm256_i64gather_ps(BASE, INDEX, SCALE)
\
--- gcc/config/i386/xopintrin.h.jj 2020-01-12 11:54:36.336414570 +0100
+++ gcc/config/i386/xopintrin.h 2020-03-05 15:40:31.663241763 +0100
@@ -814,25 +814,25 @@ _mm256_permute2_ps (__m256 __X, __m256 _
#define _mm_permute2_pd(X, Y, C, I) \
((__m128d) __builtin_ia32_vpermil2pd ((__v2df)(__m128d)(X), \
(__v2df)(__m128d)(Y), \
- (__v2di)(__m128d)(C), \
+ (__v2di)(__m128i)(C), \
(int)(I)))
#define _mm256_permute2_pd(X, Y, C, I) \
((__m256d) __builtin_ia32_vpermil2pd256 ((__v4df)(__m256d)(X), \
(__v4df)(__m256d)(Y), \
- (__v4di)(__m256d)(C), \
+ (__v4di)(__m256i)(C), \
(int)(I)))
#define _mm_permute2_ps(X, Y, C, I) \
((__m128) __builtin_ia32_vpermil2ps ((__v4sf)(__m128)(X), \
(__v4sf)(__m128)(Y), \
- (__v4si)(__m128)(C), \
+ (__v4si)(__m128i)(C), \
(int)(I)))
#define _mm256_permute2_ps(X, Y, C, I) \
((__m256) __builtin_ia32_vpermil2ps256 ((__v8sf)(__m256)(X), \
(__v8sf)(__m256)(Y), \
- (__v8si)(__m256)(C), \
+ (__v8si)(__m256i)(C), \
(int)(I)))
#endif /* __OPTIMIZE__ */
--- gcc/testsuite/g++.target/i386/pr94046-1.C.jj 2020-03-05
16:08:37.992247115 +0100
+++ gcc/testsuite/g++.target/i386/pr94046-1.C 2020-03-05 16:07:46.908004233
+0100
@@ -0,0 +1,55 @@
+// PR target/94046
+// { dg-do compile }
+// { dg-options "-O2 -mavx2 -mxop" }
+
+#include <x86intrin.h>
+
+#define S(x) struct x { operator __##x (); };
+S (m128)
+S (m128d)
+S (m128i)
+S (m256)
+S (m256d)
+S (m256i)
+
+__m128
+f1 (m128 src, float const *base, m128i idx, m128 mask)
+{
+ return _mm_mask_i32gather_ps (src, base, idx, mask, 2);
+}
+
+__m256
+f2 (m256 src, float const *base, m256i idx, m256 mask)
+{
+ return _mm256_mask_i32gather_ps (src, base, idx, mask, 2);
+}
+
+__m128
+f3 (m128 src, float const *base, m128i idx, m128 mask)
+{
+ return _mm_mask_i64gather_ps (src, base, idx, mask, 2);
+}
+
+__m128d
+f4 (m128d x, m128d y, m128i c)
+{
+ return _mm_permute2_pd (x, y, c, 3);
+}
+
+__m128
+f5 (m128 x, m128 y, m128i c)
+{
+ return _mm_permute2_ps (x, y, c, 3);
+}
+
+__m256d
+f6 (m256d x, m256d y, m256i c)
+{
+ return _mm256_permute2_pd (x, y, c, 3);
+}
+
+__m256
+f7 (m256 x, m256 y, m256i c)
+{
+ return _mm256_permute2_ps (x, y, c, 3);
+}
--- gcc/testsuite/g++.target/i386/pr94046-2.C.jj 2020-03-05
16:08:44.321153311 +0100
+++ gcc/testsuite/g++.target/i386/pr94046-2.C 2020-03-05 16:08:58.667940674
+0100
@@ -0,0 +1,5 @@
+// PR target/94046
+// { dg-do compile }
+// { dg-options "-O0 -mavx2 -mxop" }
+
+#include "pr94046-1.C"
Jakub
