i386: Fix blend vector permutation for 8-byte modes

8-byte modes should be processed only for TARGET_MMX_WITH_SSE. Handle
V2SFmode and fix V2HImode handling. The resulting BLEND instructions
are always faster than MOVSS/MOVSD, so prioritize them w.r.t MOVSS/MOVSD
for TARGET_SSE4_1.

gcc/ChangeLog:

	* config/i386/i386-expand.cc (expand_vec_perm_blend):
	Handle 8-byte modes only with TARGET_MMX_WITH_SSE. Handle V2SFmode
	and fix V2HImode handling.
	(expand_vec_perm_1): Try to emit BLEND instruction
	before MOVSS/MOVSD.
	* config/i386/mmx.md (*mmx_blendps): New insn pattern.

gcc/testsuite/ChangeLog:

	* gcc.target/i386/merge-1.c (dg-options): Use -mno-sse4.
	* gcc.target/i386/sse2-mmx-21.c (dg-options): Ditto.
	* gcc.target/i386/sse-movss-4.c (dg-options):
	Use -mno-sse4.  Simplify scan-assembler-not strings.
	* gcc.target/i386/sse2-movsd-3.c (dg-options): Ditto.
	* gcc.target/i386/sse2-mmx-movss-1.c: New test.

7 files changed, 53 insertions, 18 deletions

diff --git a/gcc/config/i386/i386-expand.cc b/gcc/config/i386/i386-expand.cc
index e89abf2..1545d43 100644
--- a/gcc/config/i386/i386-expand.cc
+++ b/gcc/config/i386/i386-expand.cc
@@ -19007,9 +19007,10 @@ expand_vec_perm_blend (struct expand_vec_perm_d *d)
     ;
   else if (TARGET_AVX && (vmode == V4DFmode || vmode == V8SFmode))
     ;
-  else if (TARGET_SSE4_1 && (GET_MODE_SIZE (vmode) == 16
-			     || GET_MODE_SIZE (vmode) == 8
-			     || GET_MODE_SIZE (vmode) == 4))
+  else if (TARGET_SSE4_1
+	   && (GET_MODE_SIZE (vmode) == 16
+	       || (TARGET_MMX_WITH_SSE && GET_MODE_SIZE (vmode) == 8)
+	       || GET_MODE_SIZE (vmode) == 4))
     ;
   else
     return false;
@@ -19042,6 +19043,8 @@ expand_vec_perm_blend (struct expand_vec_perm_d *d)
     case E_V8SFmode:
     case E_V2DFmode:
     case E_V4SFmode:
+    case E_V2SFmode:
+    case E_V2HImode:
     case E_V4HImode:
     case E_V8HImode:
     case E_V8SImode:
@@ -19897,11 +19900,15 @@ expand_vec_perm_1 (struct expand_vec_perm_d *d)
 	}
     }
 
+  /* Try the SSE4.1 blend variable merge instructions.  */
+  if (expand_vec_perm_blend (d))
+    return true;
+
   /* Try movss/movsd instructions.  */
   if (expand_vec_perm_movs (d))
     return true;
 
-  /* Finally, try the fully general two operand permute.  */
+  /* Try the fully general two operand permute.  */
   if (expand_vselect_vconcat (d->target, d->op0, d->op1, d->perm, nelt,
 			      d->testing_p))
     return true;
@@ -19924,10 +19931,6 @@ expand_vec_perm_1 (struct expand_vec_perm_d *d)
 	return true;
     }
 
-  /* Try the SSE4.1 blend variable merge instructions.  */
-  if (expand_vec_perm_blend (d))
-    return true;
-
   /* Try one of the AVX vpermil variable permutations.  */
   if (expand_vec_perm_vpermil (d))
     return true;
diff --git a/gcc/config/i386/mmx.md b/gcc/config/i386/mmx.md
index f9c6611..18dae03 100644
--- a/gcc/config/i386/mmx.md
+++ b/gcc/config/i386/mmx.md
@@ -1154,6 +1154,25 @@
   DONE;
 })
 
+(define_insn "*mmx_blendps"
+  [(set (match_operand:V2SF 0 "register_operand" "=Yr,*x,x")
+	(vec_merge:V2SF
+	  (match_operand:V2SF 2 "register_operand" "Yr,*x,x")
+	  (match_operand:V2SF 1 "register_operand" "0,0,x")
+	  (match_operand:SI 3 "const_0_to_3_operand")))]
+  "TARGET_SSE4_1 && TARGET_MMX_WITH_SSE"
+  "@
+   blendps\t{%3, %2, %0|%0, %2, %3}
+   blendps\t{%3, %2, %0|%0, %2, %3}
+   vblendps\t{%3, %2, %1, %0|%0, %1, %2, %3}"
+  [(set_attr "isa" "noavx,noavx,avx")
+   (set_attr "type" "ssemov")
+   (set_attr "length_immediate" "1")
+   (set_attr "prefix_data16" "1,1,*")
+   (set_attr "prefix_extra" "1")
+   (set_attr "prefix" "orig,orig,vex")
+   (set_attr "mode" "V4SF")])
+
 (define_insn "mmx_blendvps"
   [(set (match_operand:V2SF 0 "register_operand" "=Yr,*x,x")
 	(unspec:V2SF
diff --git a/gcc/testsuite/gcc.target/i386/merge-1.c b/gcc/testsuite/gcc.target/i386/merge-1.c
index d525685..b018eb1 100644
--- a/gcc/testsuite/gcc.target/i386/merge-1.c
+++ b/gcc/testsuite/gcc.target/i386/merge-1.c
@@ -1,5 +1,5 @@
 /* { dg-do compile } */
-/* { dg-options "-O1 -msse2" } */
+/* { dg-options "-O1 -msse2 -mno-sse4" } */
 
 #include <x86intrin.h>
 
diff --git a/gcc/testsuite/gcc.target/i386/sse-movss-4.c b/gcc/testsuite/gcc.target/i386/sse-movss-4.c
index ec3019c..d8a8a03 100644
--- a/gcc/testsuite/gcc.target/i386/sse-movss-4.c
+++ b/gcc/testsuite/gcc.target/i386/sse-movss-4.c
@@ -1,5 +1,5 @@
 /* { dg-do compile } */
-/* { dg-options "-O2 -msse" } */
+/* { dg-options "-O2 -msse -mno-sse4" } */
 
 typedef unsigned int v4si __attribute__((vector_size(16)));
 typedef float v4sf __attribute__((vector_size(16)));
@@ -7,7 +7,7 @@ typedef float v4sf __attribute__((vector_size(16)));
 v4si foo(v4si x,v4si y) { return (v4si){y[0],x[1],x[2],x[3]}; }
 v4sf bar(v4sf x,v4sf y) { return (v4sf){y[0],x[1],x[2],x[3]}; }
 
-/* { dg-final { scan-assembler-times "\tv?movss\t" 2 } } */
+/* { dg-final { scan-assembler-times "\tmovss\t" 2 } } */
 /* { dg-final { scan-assembler-not "movaps" } } */
 /* { dg-final { scan-assembler-not "shufps" } } */
-/* { dg-final { scan-assembler-not "vpblendw" } } */
+/* { dg-final { scan-assembler-not "pblendw" } } */
diff --git a/gcc/testsuite/gcc.target/i386/sse2-mmx-21.c b/gcc/testsuite/gcc.target/i386/sse2-mmx-21.c
index 8f5341e..7f8098a 100644
--- a/gcc/testsuite/gcc.target/i386/sse2-mmx-21.c
+++ b/gcc/testsuite/gcc.target/i386/sse2-mmx-21.c
@@ -1,5 +1,5 @@
 /* { dg-do compile { target { ! ia32 } } } */
-/* { dg-options "-O2 -msse2 -mno-mmx" } */
+/* { dg-options "-O2 -msse2 -mno-mmx -mno-sse4" } */
 /* { dg-final { scan-assembler-times "pshufd" 1 } } */
 /* { dg-final { scan-assembler-times "movd" 1 } } */
 /* { dg-final { scan-assembler-not "%mm" } } */
diff --git a/gcc/testsuite/gcc.target/i386/sse2-mmx-movss-1.c b/gcc/testsuite/gcc.target/i386/sse2-mmx-movss-1.c
new file mode 100644
index 0000000..bb79628
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/sse2-mmx-movss-1.c
@@ -0,0 +1,13 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-options "-O2 -msse2 -mno-sse4" } */
+
+typedef unsigned int v2si __attribute__((vector_size(8)));
+typedef float v2sf __attribute__((vector_size(8)));
+
+v2si foo(v2si x,v2si y) { return (v2si){y[0],x[1]}; }
+v2sf bar(v2sf x,v2sf y) { return (v2sf){y[0],x[1]}; }
+
+/* { dg-final { scan-assembler-times "\tmovss\t" 2 } } */
+/* { dg-final { scan-assembler-not "movaps" } } */
+/* { dg-final { scan-assembler-not "shufps" } } */
+/* { dg-final { scan-assembler-not "pblendw" } } */
diff --git a/gcc/testsuite/gcc.target/i386/sse2-movsd-3.c b/gcc/testsuite/gcc.target/i386/sse2-movsd-3.c
index fadbe2b..edd4a44 100644
--- a/gcc/testsuite/gcc.target/i386/sse2-movsd-3.c
+++ b/gcc/testsuite/gcc.target/i386/sse2-movsd-3.c
@@ -1,5 +1,5 @@
 /* { dg-do compile } */
-/* { dg-options "-O2 -msse2" } */
+/* { dg-options "-O2 -msse2 -mno-sse4" } */
 
 typedef unsigned long long v2di __attribute__((vector_size(16)));
 typedef double v2df __attribute__((vector_size(16)));
@@ -7,9 +7,9 @@ typedef double v2df __attribute__((vector_size(16)));
 v2di foo(v2di x,v2di y) { return (v2di){y[0],x[1]}; }
 v2df bar(v2df x,v2df y) { return (v2df){y[0],x[1]}; }
 
-/* { dg-final { scan-assembler-times "\tv?movsd\t" 2 } } */
-/* { dg-final { scan-assembler-not "v?shufpd" } } */
+/* { dg-final { scan-assembler-times "\tmovsd\t" 2 } } */
+/* { dg-final { scan-assembler-not "shufpd" } } */
 /* { dg-final { scan-assembler-not "movdqa" } } */
 /* { dg-final { scan-assembler-not "pshufd" } } */
-/* { dg-final { scan-assembler-not "v?punpckldq" } } */
-/* { dg-final { scan-assembler-not "v?movq" } } */
+/* { dg-final { scan-assembler-not "punpckldq" } } */
+/* { dg-final { scan-assembler-not "movq" } } */