i386: Introduce mulv2si3 instruction

For SSE2 targets the expander unpacks input elements into the correct
position in the V4SI vector and emits PMULUDQ instruction.  The output
elements are then shuffled back to their positions in the V2SI vector.

For SSE4 targets PMULLD instruction is emitted directly.

gcc/ChangeLog:

	* config/i386/mmx.md (mulv2si3): New expander.
	(*mulv2si3): New insn pattern.

gcc/testsuite/ChangeLog:

	* gcc.target/i386/sse2-mmx-mult-vec.c: New test.

2 files changed, 76 insertions, 0 deletions

diff --git a/gcc/config/i386/mmx.md b/gcc/config/i386/mmx.md
index 872ddbc..6dd203f 100644
--- a/gcc/config/i386/mmx.md
+++ b/gcc/config/i386/mmx.md
@@ -2092,6 +2092,55 @@
    (set_attr "type" "sseadd")
    (set_attr "mode" "TI")])
 
+(define_expand "mulv2si3"
+  [(set (match_operand:V2SI 0 "register_operand")
+	(mult:V2SI
+	  (match_operand:V2SI 1 "register_operand")
+	  (match_operand:V2SI 2 "register_operand")))]
+  "TARGET_MMX_WITH_SSE"
+{
+  if (!TARGET_SSE4_1)
+    {
+      rtx op1 = lowpart_subreg (V4SImode, force_reg (V2SImode, operands[1]),
+				V2SImode);
+      rtx op2 = lowpart_subreg (V4SImode, force_reg (V2SImode, operands[2]),
+				V2SImode);
+
+      rtx tmp1 = gen_reg_rtx (V4SImode);
+      emit_insn (gen_vec_interleave_lowv4si (tmp1, op1, op1));
+      rtx tmp2 = gen_reg_rtx (V4SImode);
+      emit_insn (gen_vec_interleave_lowv4si (tmp2, op2, op2));
+
+      rtx res = gen_reg_rtx (V2DImode);
+      emit_insn (gen_vec_widen_umult_even_v4si (res, tmp1, tmp2));
+
+      rtx op0 = gen_reg_rtx (V4SImode);
+      emit_insn (gen_sse2_pshufd_1 (op0, gen_lowpart (V4SImode, res),
+				    const0_rtx, const2_rtx,
+				    const0_rtx, const2_rtx));
+
+      emit_move_insn (operands[0], lowpart_subreg (V2SImode, op0, V4SImode));
+      DONE;
+    }
+})
+
+(define_insn "*mulv2si3"
+  [(set (match_operand:V2SI 0 "register_operand" "=Yr,*x,v")
+	(mult:V2SI
+	  (match_operand:V2SI 1 "register_operand" "%0,0,v")
+	  (match_operand:V2SI 2 "register_operand" "Yr,*x,v")))]
+  "TARGET_SSE4_1 && TARGET_MMX_WITH_SSE"
+  "@
+   pmulld\t{%2, %0|%0, %2}
+   pmulld\t{%2, %0|%0, %2}
+   vpmulld\t{%2, %1, %0|%0, %1, %2}"
+  [(set_attr "isa" "noavx,noavx,avx")
+   (set_attr "type" "sseimul")
+   (set_attr "prefix_extra" "1")
+   (set_attr "prefix" "orig,orig,vex")
+   (set_attr "btver2_decode" "vector")
+   (set_attr "mode" "TI")])
+
 (define_expand "mmx_mulv4hi3"
   [(set (match_operand:V4HI 0 "register_operand")
         (mult:V4HI (match_operand:V4HI 1 "register_mmxmem_operand")
diff --git a/gcc/testsuite/gcc.target/i386/sse2-mmx-mult-vec.c b/gcc/testsuite/gcc.target/i386/sse2-mmx-mult-vec.c
new file mode 100644
index 0000000..cdc9a7b
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/sse2-mmx-mult-vec.c
@@ -0,0 +1,27 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -ftree-vectorize -msse2" } */
+/* { dg-require-effective-target sse2 } */
+
+#include "sse2-check.h"
+
+#define N 2
+
+int a[N] = {-287807, 604344};
+int b[N] = {474362, 874120};
+int r[N];
+
+int rc[N] = {914249338, -11800128};
+
+static void
+sse2_test (void)
+{
+  int i;
+
+  for (i = 0; i < N; i++)
+    r[i] = a[i] * b[i];
+
+  /* check results:  */
+  for (i = 0; i < N; i++)
+    if (r[i] != rc[i])
+      abort ();
+}