i386: Add AVX512 unaligned intrinsics

__m512i _mm512_loadu_epi64( void * sa);
void _mm512_storeu_epi64(void * d, __m512i a);
__m512i _mm512_loadu_epi32( void * sa);
void _mm512_storeu_epi32(void * d, __m512i a);
void _mm256_storeu_epi64(void * d, __m256i a);
void _mm_storeu_epi64(void * d, __m128i a);
void _mm256_storeu_epi32(void * d, __m256i a);
void _mm_storeu_epi32(void * d, __m128i a);

Tested on x86-64.

2019-07-11  Sunil K Pandey  <sunil.k.pandey@intel.com>

gcc/

	PR target/90980
	* config/i386/avx512fintrin.h (_mm512_loadu_epi64): New.
	(_mm512_storeu_epi64): Likewise.
	(_mm512_loadu_epi32): Likewise.
	(_mm512_storeu_epi32): Likewise.
	* config/i386/avx512vlintrin.h (_mm256_storeu_epi64): New.
	(_mm_storeu_epi64): Likewise.
	(_mm256_storeu_epi32): Likewise.
	(_mm_storeu_epi32): Likewise.

gcc/testsuite/

	PR target/90980
	* gcc.target/i386/pr90980-1.c: New test.
	* gcc.target/i386/pr90980-2.c: Likewise.
	* gcc.target/i386/pr90980-3.c: Likewise.

From-SVN: r273416

7 files changed, 129 insertions, 0 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index aa9e05c..235cc53 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,15 @@
+2019-07-11  Sunil K Pandey  <sunil.k.pandey@intel.com>
+
+	PR target/90980
+	* config/i386/avx512fintrin.h (_mm512_loadu_epi64): New.
+	(_mm512_storeu_epi64): Likewise.
+	(_mm512_loadu_epi32): Likewise.
+	(_mm512_storeu_epi32): Likewise.
+	* config/i386/avx512vlintrin.h (_mm256_storeu_epi64): New.
+	(_mm_storeu_epi64): Likewise.
+	(_mm256_storeu_epi32): Likewise.
+	(_mm_storeu_epi32): Likewise.
+
 2019-07-11  Segher Boessenkool  <segher@kernel.crashing.org>
 
 	* config/rs6000/rs6000-logue.c: Add Modula-2 to comment.
diff --git a/gcc/config/i386/avx512fintrin.h b/gcc/config/i386/avx512fintrin.h
index e35eedb..454fd3d 100644
--- a/gcc/config/i386/avx512fintrin.h
+++ b/gcc/config/i386/avx512fintrin.h
@@ -6352,6 +6352,13 @@ _mm_mask_store_sd (double *__P, __mmask8 __U, __m128d __A)
 
 extern __inline __m512i
 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_loadu_epi64 (void const *__P)
+{
+  return *(__m512i_u *) __P;
+}
+
+extern __inline __m512i
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
 _mm512_mask_loadu_epi64 (__m512i __W, __mmask8 __U, void const *__P)
 {
   return (__m512i) __builtin_ia32_loaddqudi512_mask ((const long long *) __P,
@@ -6371,6 +6378,13 @@ _mm512_maskz_loadu_epi64 (__mmask8 __U, void const *__P)
 
 extern __inline void
 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_storeu_epi64 (void *__P, __m512i __A)
+{
+  *(__m512i_u *) __P = (__m512i_u) __A;
+}
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
 _mm512_mask_storeu_epi64 (void *__P, __mmask8 __U, __m512i __A)
 {
   __builtin_ia32_storedqudi512_mask ((long long *) __P, (__v8di) __A,
@@ -6386,6 +6400,13 @@ _mm512_loadu_si512 (void const *__P)
 
 extern __inline __m512i
 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_loadu_epi32 (void const *__P)
+{
+  return *(__m512i_u *) __P;
+}
+
+extern __inline __m512i
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
 _mm512_mask_loadu_epi32 (__m512i __W, __mmask16 __U, void const *__P)
 {
   return (__m512i) __builtin_ia32_loaddqusi512_mask ((const int *) __P,
@@ -6412,6 +6433,13 @@ _mm512_storeu_si512 (void *__P, __m512i __A)
 
 extern __inline void
 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_storeu_epi32 (void *__P, __m512i __A)
+{
+  *(__m512i_u *) __P = (__m512i_u) __A;
+}
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
 _mm512_mask_storeu_epi32 (void *__P, __mmask16 __U, __m512i __A)
 {
   __builtin_ia32_storedqusi512_mask ((int *) __P, (__v16si) __A,
diff --git a/gcc/config/i386/avx512vlintrin.h b/gcc/config/i386/avx512vlintrin.h
index 3eaf817..bd8746d 100644
--- a/gcc/config/i386/avx512vlintrin.h
+++ b/gcc/config/i386/avx512vlintrin.h
@@ -757,6 +757,13 @@ _mm_maskz_loadu_epi64 (__mmask8 __U, void const *__P)
 
 extern __inline void
 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_storeu_epi64 (void *__P, __m256i __A)
+{
+  *(__m256i_u *) __P = (__m256i_u) __A;
+}
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
 _mm256_mask_storeu_epi64 (void *__P, __mmask8 __U, __m256i __A)
 {
   __builtin_ia32_storedqudi256_mask ((long long *) __P,
@@ -766,6 +773,13 @@ _mm256_mask_storeu_epi64 (void *__P, __mmask8 __U, __m256i __A)
 
 extern __inline void
 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_storeu_epi64 (void *__P, __m128i __A)
+{
+  *(__m128i_u *) __P = (__m128i_u) __A;
+}
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
 _mm_mask_storeu_epi64 (void *__P, __mmask8 __U, __m128i __A)
 {
   __builtin_ia32_storedqudi128_mask ((long long *) __P,
@@ -813,6 +827,13 @@ _mm_maskz_loadu_epi32 (__mmask8 __U, void const *__P)
 
 extern __inline void
 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_storeu_epi32 (void *__P, __m256i __A)
+{
+  *(__m256i_u *) __P = (__m256i_u) __A;
+}
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
 _mm256_mask_storeu_epi32 (void *__P, __mmask8 __U, __m256i __A)
 {
   __builtin_ia32_storedqusi256_mask ((int *) __P,
@@ -822,6 +843,13 @@ _mm256_mask_storeu_epi32 (void *__P, __mmask8 __U, __m256i __A)
 
 extern __inline void
 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_storeu_epi32 (void *__P, __m128i __A)
+{
+  *(__m128i_u *) __P = (__m128i_u) __A;
+}
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
 _mm_mask_storeu_epi32 (void *__P, __mmask8 __U, __m128i __A)
 {
   __builtin_ia32_storedqusi128_mask ((int *) __P,
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index f4a6d05..cfc96b0 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,10 @@
+2019-07-11  Sunil K Pandey  <sunil.k.pandey@intel.com>
+
+	PR target/90980
+	* gcc.target/i386/pr90980-1.c: New test.
+	* gcc.target/i386/pr90980-2.c: Likewise.
+	* gcc.target/i386/pr90980-3.c: Likewise.
+
 2019-07-11  Yannick Moy  <moy@adacore.com>
 
 	* gnat.dg/loop_entry1.adb: New testcase.
diff --git a/gcc/testsuite/gcc.target/i386/pr90980-1.c b/gcc/testsuite/gcc.target/i386/pr90980-1.c
new file mode 100644
index 0000000..72a30dc
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr90980-1.c
@@ -0,0 +1,17 @@
+/* { dg-do compile } */
+/* { dg-options "-march=skylake-avx512 -O2" } */
+/* { dg-final { scan-assembler-times "(?:vmovups|vmovdqu)\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\]*\\)(?:\n|\[ \\t\]+#)" 2 } } */
+
+#include <immintrin.h>
+
+int *a;
+long long *b;
+volatile __m128i xx;
+volatile __m128i xx1;
+
+void extern
+avx512vl_test (void)
+{
+  _mm_storeu_epi32 (a, xx);
+  _mm_storeu_epi64 (b, xx1);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr90980-2.c b/gcc/testsuite/gcc.target/i386/pr90980-2.c
new file mode 100644
index 0000000..b1980e6
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr90980-2.c
@@ -0,0 +1,17 @@
+/* { dg-do compile } */
+/* { dg-options "-march=skylake-avx512 -O2" } */
+/* { dg-final { scan-assembler-times "vmovdqu\[0-9\]*\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\]*\\)(?:\n|\[ \\t\]+#)" 2 } } */
+
+#include <immintrin.h>
+
+int *a;
+long long *b;
+volatile __m256i yy;
+volatile __m256i yy1;
+
+void extern
+avx512vl_test (void)
+{
+  _mm256_storeu_epi32 (a, yy);
+  _mm256_storeu_epi64 (b, yy1);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr90980-3.c b/gcc/testsuite/gcc.target/i386/pr90980-3.c
new file mode 100644
index 0000000..d839ee0
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr90980-3.c
@@ -0,0 +1,20 @@
+/* { dg-do compile } */
+/* { dg-options "-march=skylake-avx512 -O2" } */
+/* { dg-final { scan-assembler-times "vmovdqu64\[ \\t\]+\[^\{\n\]*\\)\[^\n\]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 2 } } */
+/* { dg-final { scan-assembler-times "vmovdqu64\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\]*\\)(?:\n|\[ \\t\]+#)" 2 } } */
+
+#include <immintrin.h>
+
+int *a;
+long long *b;
+volatile __m512i zz;
+volatile __m512i zz1;
+
+void extern
+avx512f_test (void)
+{
+  zz = _mm512_loadu_epi32 (a);
+  _mm512_storeu_epi32 (a, zz);
+  zz1 = _mm512_loadu_epi64 (b);
+  _mm512_storeu_epi64 (b, zz1);
+}