re PR target/71019 (AVX512BW instructions emitted even without AVX512BW)

	PR target/71019
	* config/i386/sse.md (<sse2_avx2>_packssdw<mask_name>,
	<sse4_1_avx2>_packusdw<mask_name>): Make sure EVEX encoded insn
	is not emitted unless TARGET_AVX512BW.
	(<sse2_avx2>_packuswb<mask_name>, <sse2_avx2>_packsswb<mask_name>):
	Likewise.  For TARGET_AVX512BW, use "=v" constraint instead of "=x"
	for the result operand.

	* gcc.target/i386/avx512vl-pack-1.c: New test.
	* gcc.target/i386/avx512vl-pack-2.c: New test.
	* gcc.target/i386/avx512bw-pack-2.c: New test.

From-SVN: r236163

4 files changed, 281 insertions, 0 deletions

diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index 31f7bc2..ec16edd 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,5 +1,10 @@
 2016-05-12  Jakub Jelinek  <jakub@redhat.com>
 
+	PR target/71019
+	* gcc.target/i386/avx512vl-pack-1.c: New test.
+	* gcc.target/i386/avx512vl-pack-2.c: New test.
+	* gcc.target/i386/avx512bw-pack-2.c: New test.
+
 	* gcc.target/i386/avx512vl-vinsertps-1.c: New test.
 
 	* gcc.target/i386/avx512dq-abs-copysign-1.c: New test.
diff --git a/gcc/testsuite/gcc.target/i386/avx512bw-pack-2.c b/gcc/testsuite/gcc.target/i386/avx512bw-pack-2.c
new file mode 100644
index 0000000..3b9c201
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/avx512bw-pack-2.c
@@ -0,0 +1,100 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-options "-O2 -mavx512vl -mavx512bw" } */
+
+#include <x86intrin.h>
+
+__m128i
+f1 (__m128i a, __m128i b)
+{
+  register __m128i c __asm ("xmm16") = a;
+  asm volatile ("" : "+v" (c));
+  c = _mm_packs_epi16 (c, b);
+  asm volatile ("" : "+v" (c));
+  return c;
+}
+
+/* { dg-final { scan-assembler-times "vpacksswb\[^\n\r\]*xmm16" 1 } } */
+
+__m128i
+f2 (__m128i a, __m128i b)
+{
+  register __m128i c __asm ("xmm16") = a;
+  asm volatile ("" : "+v" (c));
+  c = _mm_packs_epi32 (c, b);
+  asm volatile ("" : "+v" (c));
+  return c;
+}
+
+/* { dg-final { scan-assembler-times "vpackssdw\[^\n\r\]*xmm16" 1 } } */
+
+__m128i
+f3 (__m128i a, __m128i b)
+{
+  register __m128i c __asm ("xmm16") = a;
+  asm volatile ("" : "+v" (c));
+  c = _mm_packus_epi16 (c, b);
+  asm volatile ("" : "+v" (c));
+  return c;
+}
+
+/* { dg-final { scan-assembler-times "vpackuswb\[^\n\r\]*xmm16" 1 } } */
+
+__m128i
+f4 (__m128i a, __m128i b)
+{
+  register __m128i c __asm ("xmm16") = a;
+  asm volatile ("" : "+v" (c));
+  c = _mm_packus_epi32 (c, b);
+  asm volatile ("" : "+v" (c));
+  return c;
+}
+
+/* { dg-final { scan-assembler-times "vpackusdw\[^\n\r\]*xmm16" 1 } } */
+
+__m256i
+f5 (__m256i a, __m256i b)
+{
+  register __m256i c __asm ("xmm16") = a;
+  asm volatile ("" : "+v" (c));
+  c = _mm256_packs_epi16 (c, b);
+  asm volatile ("" : "+v" (c));
+  return c;
+}
+
+/* { dg-final { scan-assembler-times "vpacksswb\[^\n\r\]*ymm16" 1 } } */
+
+__m256i
+f6 (__m256i a, __m256i b)
+{
+  register __m256i c __asm ("xmm16") = a;
+  asm volatile ("" : "+v" (c));
+  c = _mm256_packs_epi32 (c, b);
+  asm volatile ("" : "+v" (c));
+  return c;
+}
+
+/* { dg-final { scan-assembler-times "vpackssdw\[^\n\r\]*ymm16" 1 } } */
+
+__m256i
+f7 (__m256i a, __m256i b)
+{
+  register __m256i c __asm ("xmm16") = a;
+  asm volatile ("" : "+v" (c));
+  c = _mm256_packus_epi16 (c, b);
+  asm volatile ("" : "+v" (c));
+  return c;
+}
+
+/* { dg-final { scan-assembler-times "vpackuswb\[^\n\r\]*ymm16" 1 } } */
+
+__m256i
+f8 (__m256i a, __m256i b)
+{
+  register __m256i c __asm ("xmm16") = a;
+  asm volatile ("" : "+v" (c));
+  c = _mm256_packus_epi32 (c, b);
+  asm volatile ("" : "+v" (c));
+  return c;
+}
+
+/* { dg-final { scan-assembler-times "vpackusdw\[^\n\r\]*ymm16" 1 } } */
diff --git a/gcc/testsuite/gcc.target/i386/avx512vl-pack-1.c b/gcc/testsuite/gcc.target/i386/avx512vl-pack-1.c
new file mode 100644
index 0000000..a589d63
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/avx512vl-pack-1.c
@@ -0,0 +1,68 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl -mno-avx512bw" } */
+
+#include <x86intrin.h>
+
+__m128i
+f1 (__m128i a, __m128i b)
+{
+  return _mm_packs_epi16 (a, b);
+}
+
+/* { dg-final { scan-assembler-times "vpacksswb\[^\n\r\]*xmm\[0-9\]" 1 } } */
+
+__m128i
+f2 (__m128i a, __m128i b)
+{
+  return _mm_packs_epi32 (a, b);
+}
+
+/* { dg-final { scan-assembler-times "vpackssdw\[^\n\r\]*xmm\[0-9\]" 1 } } */
+
+__m128i
+f3 (__m128i a, __m128i b)
+{
+  return _mm_packus_epi16 (a, b);
+}
+
+/* { dg-final { scan-assembler-times "vpackuswb\[^\n\r\]*xmm\[0-9\]" 1 } } */
+
+__m128i
+f4 (__m128i a, __m128i b)
+{
+  return _mm_packus_epi32 (a, b);
+}
+
+/* { dg-final { scan-assembler-times "vpackusdw\[^\n\r\]*xmm\[0-9\]" 1 } } */
+
+__m256i
+f5 (__m256i a, __m256i b)
+{
+  return _mm256_packs_epi16 (a, b);
+}
+
+/* { dg-final { scan-assembler-times "vpacksswb\[^\n\r\]*ymm\[0-9\]" 1 } } */
+
+__m256i
+f6 (__m256i a, __m256i b)
+{
+  return _mm256_packs_epi32 (a, b);
+}
+
+/* { dg-final { scan-assembler-times "vpackssdw\[^\n\r\]*ymm\[0-9\]" 1 } } */
+
+__m256i
+f7 (__m256i a, __m256i b)
+{
+  return _mm256_packus_epi16 (a, b);
+}
+
+/* { dg-final { scan-assembler-times "vpackuswb\[^\n\r\]*ymm\[0-9\]" 1 } } */
+
+__m256i
+f8 (__m256i a, __m256i b)
+{
+  return _mm256_packus_epi32 (a, b);
+}
+
+/* { dg-final { scan-assembler-times "vpackusdw\[^\n\r\]*ymm\[0-9\]" 1 } } */
diff --git a/gcc/testsuite/gcc.target/i386/avx512vl-pack-2.c b/gcc/testsuite/gcc.target/i386/avx512vl-pack-2.c
new file mode 100644
index 0000000..0582006
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/avx512vl-pack-2.c
@@ -0,0 +1,108 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-options "-O2 -mavx512vl -mno-avx512bw" } */
+
+#include <x86intrin.h>
+
+__m128i
+f1 (__m128i a, __m128i b)
+{
+  register __m128i c __asm ("xmm16") = a;
+  asm volatile ("" : "+v" (c));
+  c = _mm_packs_epi16 (c, b);
+  asm volatile ("" : "+v" (c));
+  return c;
+}
+
+/* { dg-final { scan-assembler-times "vpacksswb\[^\n\r\]*xmm\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-not "vpacksswb\[^\n\r\]*xmm16" } } */
+
+__m128i
+f2 (__m128i a, __m128i b)
+{
+  register __m128i c __asm ("xmm16") = a;
+  asm volatile ("" : "+v" (c));
+  c = _mm_packs_epi32 (c, b);
+  asm volatile ("" : "+v" (c));
+  return c;
+}
+
+/* { dg-final { scan-assembler-times "vpackssdw\[^\n\r\]*xmm\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-not "vpackssdw\[^\n\r\]*xmm16" } } */
+
+__m128i
+f3 (__m128i a, __m128i b)
+{
+  register __m128i c __asm ("xmm16") = a;
+  asm volatile ("" : "+v" (c));
+  c = _mm_packus_epi16 (c, b);
+  asm volatile ("" : "+v" (c));
+  return c;
+}
+
+/* { dg-final { scan-assembler-times "vpackuswb\[^\n\r\]*xmm\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-not "vpackuswb\[^\n\r\]*xmm16" } } */
+
+__m128i
+f4 (__m128i a, __m128i b)
+{
+  register __m128i c __asm ("xmm16") = a;
+  asm volatile ("" : "+v" (c));
+  c = _mm_packus_epi32 (c, b);
+  asm volatile ("" : "+v" (c));
+  return c;
+}
+
+/* { dg-final { scan-assembler-times "vpackusdw\[^\n\r\]*xmm\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-not "vpackusdw\[^\n\r\]*xmm16" } } */
+
+__m256i
+f5 (__m256i a, __m256i b)
+{
+  register __m256i c __asm ("xmm16") = a;
+  asm volatile ("" : "+v" (c));
+  c = _mm256_packs_epi16 (c, b);
+  asm volatile ("" : "+v" (c));
+  return c;
+}
+
+/* { dg-final { scan-assembler-times "vpacksswb\[^\n\r\]*ymm\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-not "vpacksswb\[^\n\r\]*ymm16" } } */
+
+__m256i
+f6 (__m256i a, __m256i b)
+{
+  register __m256i c __asm ("xmm16") = a;
+  asm volatile ("" : "+v" (c));
+  c = _mm256_packs_epi32 (c, b);
+  asm volatile ("" : "+v" (c));
+  return c;
+}
+
+/* { dg-final { scan-assembler-times "vpackssdw\[^\n\r\]*ymm\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-not "vpackssdw\[^\n\r\]*ymm16" } } */
+
+__m256i
+f7 (__m256i a, __m256i b)
+{
+  register __m256i c __asm ("xmm16") = a;
+  asm volatile ("" : "+v" (c));
+  c = _mm256_packus_epi16 (c, b);
+  asm volatile ("" : "+v" (c));
+  return c;
+}
+
+/* { dg-final { scan-assembler-times "vpackuswb\[^\n\r\]*ymm\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-not "vpackuswb\[^\n\r\]*ymm16" } } */
+
+__m256i
+f8 (__m256i a, __m256i b)
+{
+  register __m256i c __asm ("xmm16") = a;
+  asm volatile ("" : "+v" (c));
+  c = _mm256_packus_epi32 (c, b);
+  asm volatile ("" : "+v" (c));
+  return c;
+}
+
+/* { dg-final { scan-assembler-times "vpackusdw\[^\n\r\]*ymm\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-not "vpackusdw\[^\n\r\]*ymm16" } } */