[PATCH, i386]: Use ssememalign attribute value to reject insns with misaligned operands

[PATCH, i386]: Use ssememalign attribute value to reject insns with misaligned operands

[Uros Bizjak]

[-- Attachment #1: Type: text/plain, Size: 588 bytes --]

Hello!

Attached patch uses ssememalign attribute to reject insn combinations
where memory operands would be misaligned.

2015-11-12  Uros Bizjak  <ubizjak@gmail.com>

    * config/i386/i386.c (ix86_legitimate_combined_insn): Reject
    combined insn if the alignment of vector mode memory operand
    is less than ssememalign.

testsuite/ChangeLog:

2015-11-12  Uros Bizjak  <ubizjak@gmail.com>

    * gcc.target/i386/sse-1.c (swizzle): Assume that a is
    aligned to 64 bits.

Patch was bootstrapped and regression tested on x86_64-linux-gnu
{,-m32}, committed to mainline SVN.

Uros.

[-- Attachment #2: p.diff.txt --]
[-- Type: text/plain, Size: 1389 bytes --]

Index: config/i386/i386.c
===================================================================
--- config/i386/i386.c	(revision 230213)
+++ config/i386/i386.c	(working copy)
@@ -7236,11 +7236,12 @@ ix86_legitimate_combined_insn (rtx_insn *insn)
 	  /* For pre-AVX disallow unaligned loads/stores where the
 	     instructions don't support it.  */
 	  if (!TARGET_AVX
-	      && VECTOR_MODE_P (GET_MODE (op))
-	      && misaligned_operand (op, GET_MODE (op)))
+	      && VECTOR_MODE_P (mode)
+	      && misaligned_operand (op, mode))
 	    {
-	      int min_align = get_attr_ssememalign (insn);
-	      if (min_align == 0)
+	      unsigned int min_align = get_attr_ssememalign (insn);
+	      if (min_align == 0
+		  || MEM_ALIGN (op) < min_align)
 		return false;
 	    }
 
Index: testsuite/gcc.target/i386/sse-1.c
===================================================================
--- testsuite/gcc.target/i386/sse-1.c	(revision 230213)
+++ testsuite/gcc.target/i386/sse-1.c	(working copy)
@@ -14,8 +14,10 @@ typedef union
 void
 swizzle (const void *a, vector4_t * b, vector4_t * c)
 {
-  b->v = _mm_loadl_pi (b->v, (__m64 *) a);
-  c->v = _mm_loadl_pi (c->v, ((__m64 *) a) + 1);
+  __m64 *t = __builtin_assume_aligned (a, 64);
+
+  b->v = _mm_loadl_pi (b->v, t);
+  c->v = _mm_loadl_pi (c->v, t + 1);
 }
 
 /* While one legal rendering of each statement would be movaps;movlps;movaps,