re PR target/85317 (missing constant propagation on _mm(256)_movemask_*)

	PR target/85317
	* config/i386/i386.c (ix86_fold_builtin): Handle
	IX86_BUILTIN_{,P}MOVMSK{PS,PD,B}{,128,256}.

	* gcc.target/i386/pr85317.c: New test.
	* gcc.target/i386/avx2-vpmovmskb-2.c (avx2_test): Add asm volatile
	optimization barrier to avoid optimizing away the expected insn.

From-SVN: r260040

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 74c90d83fbb18749f069340c3900fb2f849434f8..0ff143bf60309bb5684a250d035c3b846b52b910 100644 (file)
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,5 +1,9 @@
 2018-05-08  Jakub Jelinek  <jakub@redhat.com>
 
+       PR target/85317
+       * config/i386/i386.c (ix86_fold_builtin): Handle
+       IX86_BUILTIN_{,P}MOVMSK{PS,PD,B}{,128,256}.
+
        PR target/85480
        * config/i386/sse.md (ssequaterinsnmode): New mode attribute.
        (*<extract_type>_vinsert<shuffletype><extract_suf>_0): New pattern.

diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index 6a2141e48dbd4f44bbd0efc9967d1afa3bf81725..d99ff968c904e6fed82dd6863a93eab6baf4c143 100644 (file)
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -33477,6 +33477,37 @@ ix86_fold_builtin (tree fndecl, int n_args,
            }
          break;
 
+       case IX86_BUILTIN_MOVMSKPS:
+       case IX86_BUILTIN_PMOVMSKB:
+       case IX86_BUILTIN_MOVMSKPD:
+       case IX86_BUILTIN_PMOVMSKB128:
+       case IX86_BUILTIN_MOVMSKPD256:
+       case IX86_BUILTIN_MOVMSKPS256:
+       case IX86_BUILTIN_PMOVMSKB256:
+         gcc_assert (n_args == 1);
+         if (TREE_CODE (args[0]) == VECTOR_CST)
+           {
+             HOST_WIDE_INT res = 0;
+             for (unsigned i = 0; i < VECTOR_CST_NELTS (args[0]); ++i)
+               {
+                 tree e = VECTOR_CST_ELT (args[0], i);
+                 if (TREE_CODE (e) == INTEGER_CST && !TREE_OVERFLOW (e))
+                   {
+                     if (wi::neg_p (wi::to_wide (e)))
+                       res |= HOST_WIDE_INT_1 << i;
+                   }
+                 else if (TREE_CODE (e) == REAL_CST && !TREE_OVERFLOW (e))
+                   {
+                     if (TREE_REAL_CST (e).sign)
+                       res |= HOST_WIDE_INT_1 << i;
+                   }
+                 else
+                   return NULL_TREE;
+               }
+             return build_int_cst (TREE_TYPE (TREE_TYPE (fndecl)), res);
+           }
+         break;
+
        default:
          break;
        }

diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index 6d92f7ee1c2443128689e3eae2fb64b8277a4296..c6d35858e4c009424255940723824264075fd8bd 100644 (file)
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,5 +1,10 @@
 2018-05-08  Jakub Jelinek  <jakub@redhat.com>
 
+       PR target/85317
+       * gcc.target/i386/pr85317.c: New test.
+       * gcc.target/i386/avx2-vpmovmskb-2.c (avx2_test): Add asm volatile
+       optimization barrier to avoid optimizing away the expected insn.
+
        PR target/85480
        * gcc.target/i386/avx512dq-pr85480-1.c: New test.
        * gcc.target/i386/avx512dq-pr85480-2.c: New test.

diff --git a/gcc/testsuite/gcc.target/i386/avx2-vpmovmskb-2.c b/gcc/testsuite/gcc.target/i386/avx2-vpmovmskb-2.c
index e5a9c10e1476912ce71183975016445c4765818a..71e65c937833a6bb14f6d6367aa5422d3043e676 100644 (file)
--- a/gcc/testsuite/gcc.target/i386/avx2-vpmovmskb-2.c
+++ b/gcc/testsuite/gcc.target/i386/avx2-vpmovmskb-2.c
@@ -14,6 +14,7 @@ avx2_test (void)
   s.x = _mm256_set_epi8 (1, 2, 3, 4, 10, 20, 30, 90, -80, -40, -100,
                         15, 98, 25, 98, 7, 1, 2, 3, 4, 10, 20, 30, 90,
                         -80, -40, -100, -15, 98, 25, 98, 7);
+  __asm volatile ("" : "+m" (s) : : "memory");
 
   res = _mm256_movemask_epi8 (s.x);
 

diff --git a/gcc/testsuite/gcc.target/i386/pr85317.c b/gcc/testsuite/gcc.target/i386/pr85317.c
new file mode 100644 (file)
index 0000000..78c5b0c
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr85317.c
@@ -0,0 +1,38 @@
+/* PR target/85317 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx2 -fdump-tree-optimized" } */
+/* { dg-final { scan-tree-dump-not "link_error" "optimized" } } */
+
+#include <x86intrin.h>
+
+extern void link_error (void);
+
+int
+main ()
+{
+  int a = _mm_movemask_pd (_mm_set_pd (-2.0, 2.0));
+  if (a != 2) link_error ();
+  int b = _mm_movemask_pd (_mm_set_pd (0.0, __builtin_copysign (0.0, -4.0)));
+  if (b != 1) link_error ();
+  int c = _mm_movemask_ps (_mm_set_ps (__builtin_copysignf (0.0f, -4.0f), 0.0f,
+                                      -4.0f, 4.0f));
+  if (c != 10) link_error ();
+  int d = _mm_movemask_epi8 (_mm_set_epi8 (-4, 8, -8, -12, 12, 15, 0, -1, -3,
+                                          -128, 127, 126, 120, -120, 0, 5));
+  if (d != 0xb1c4) link_error ();
+  int e = _mm256_movemask_pd (_mm256_set_pd (-4.0, 0.0, 4.0,
+                                            __builtin_copysign (0.0, -4.0)));
+  if (e != 9) link_error ();
+  int f = _mm256_movemask_ps (_mm256_set_ps (-8.0f, -16.0f, 12.0f, 127.0f,
+                                            -4.0f, 0.0f, 4.0f,
+                                            __builtin_copysign (0.0f,
+                                                                -4.0f)));
+  if (f != 0xc9) link_error ();
+  int g = _mm256_movemask_epi8 (_mm256_set_epi8 (-4, 8, -8, -12, 12, 15, 0, -1,
+                                                -3, -128, 127, 126, 120, -120,
+                                                0, 5, 12, 100, -20, -50, -70,
+                                                2, -65, 0, -1, 1, 2, -2, -9,
+                                                -9, 19, -64));
+  if (g != (int) 0xb1c43a9dU) link_error ();
+  return 0;
+}