This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

fix new min/max reductions on x86


Adds the new patterns required to allow these tests to be vectorized.
Tested on amd64-linux.


r~


        * config/i386/sse.md (vec_shl_<SSEMODEI>, vec_shr_<SSEMODEI>): New.
        (smaxv16qi3, umaxv8hi3, sminv16qi3, uminv8hi3): New.

        * gcc.dg/vect/vect-reduc-1short.c: Remove XFAIL.
        * gcc.dg/vect/vect-reduc-2char.c: Remove XFAIL.

Index: config/i386/sse.md
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/i386/sse.md,v
retrieving revision 1.18
diff -u -p -d -r1.18 sse.md
--- config/i386/sse.md	25 Jun 2005 01:21:24 -0000	1.18
+++ config/i386/sse.md	28 Jun 2005 06:11:20 -0000
@@ -2705,6 +2705,18 @@
   [(set_attr "type" "sseishft")
    (set_attr "mode" "TI")])
 
+(define_expand "vec_shl_<mode>"
+  [(set (match_operand:SSEMODEI 0 "register_operand" "")
+        (ashift:TI (match_operand:SSEMODEI 1 "register_operand" "")
+		   (match_operand:SI 2 "general_operand" "")))]
+  "TARGET_SSE2"
+{
+  if (!const_0_to_255_mul_8_operand (operands[2], SImode))
+    FAIL;
+  operands[0] = gen_lowpart (TImode, operands[0]);
+  operands[1] = gen_lowpart (TImode, operands[1]);
+})
+
 (define_insn "sse2_lshrti3"
   [(set (match_operand:TI 0 "register_operand" "=x")
  	(lshiftrt:TI (match_operand:TI 1 "register_operand" "0")
@@ -2717,6 +2729,33 @@
   [(set_attr "type" "sseishft")
    (set_attr "mode" "TI")])
 
+(define_expand "vec_shr_<mode>"
+  [(set (match_operand:SSEMODEI 0 "register_operand" "")
+        (lshiftrt:TI (match_operand:SSEMODEI 1 "register_operand" "")
+		     (match_operand:SI 2 "general_operand" "")))]
+  "TARGET_SSE2"
+{
+  if (!const_0_to_255_mul_8_operand (operands[2], SImode))
+    FAIL;
+  operands[0] = gen_lowpart (TImode, operands[0]);
+  operands[1] = gen_lowpart (TImode, operands[1]);
+})
+
+(define_expand "smaxv16qi3"
+  [(set (match_operand:V16QI 0 "register_operand" "")
+	(smax:V16QI (match_operand:V16QI 1 "register_operand" "")
+		    (match_operand:V16QI 2 "register_operand" "")))]
+  "TARGET_SSE2"
+{
+  bool ok;
+  operands[3] = gen_rtx_GT (VOIDmode, operands[1], operands[2]);
+  operands[4] = operands[1];
+  operands[5] = operands[2];
+  ok = ix86_expand_int_vcond (operands, false);
+  gcc_assert (ok);
+  DONE;
+})
+
 (define_expand "umaxv16qi3"
   [(set (match_operand:V16QI 0 "register_operand" "")
 	(umax:V16QI (match_operand:V16QI 1 "nonimmediate_operand" "")
@@ -2749,6 +2788,42 @@
   [(set_attr "type" "sseiadd")
    (set_attr "mode" "TI")])
 
+(define_expand "umaxv8hi3"
+  [(set (match_operand:V8HI 0 "register_operand" "")
+	(umax:V8HI (match_operand:V8HI 1 "register_operand" "")
+		   (match_operand:V8HI 2 "register_operand" "")))]
+  "TARGET_SSE2"
+{
+  rtx t1, t2;
+  bool ok;
+
+  t1 = gen_reg_rtx (V8HImode);
+  emit_insn (gen_sse2_ussubv8hi3 (t1, operands[2], operands[1]));
+  t2 = force_reg (V8HImode, CONST0_RTX (V8HImode));
+
+  operands[3] = gen_rtx_EQ (VOIDmode, t1, t2);
+  operands[4] = t1;
+  operands[5] = t2;
+  ok = ix86_expand_int_vcond (operands, false);
+  gcc_assert (ok);
+  DONE;
+})
+
+(define_expand "sminv16qi3"
+  [(set (match_operand:V16QI 0 "register_operand" "")
+	(smin:V16QI (match_operand:V16QI 1 "register_operand" "")
+		    (match_operand:V16QI 2 "register_operand" "")))]
+  "TARGET_SSE2"
+{
+  bool ok;
+  operands[3] = gen_rtx_GT (VOIDmode, operands[1], operands[2]);
+  operands[4] = operands[2];
+  operands[5] = operands[1];
+  ok = ix86_expand_int_vcond (operands, false);
+  gcc_assert (ok);
+  DONE;
+})
+
 (define_expand "uminv16qi3"
   [(set (match_operand:V16QI 0 "register_operand" "")
 	(umin:V16QI (match_operand:V16QI 1 "nonimmediate_operand" "")
@@ -2781,6 +2856,27 @@
   [(set_attr "type" "sseiadd")
    (set_attr "mode" "TI")])
 
+(define_expand "uminv8hi3"
+  [(set (match_operand:V8HI 0 "register_operand" "")
+	(umin:V8HI (match_operand:V8HI 1 "register_operand" "")
+		   (match_operand:V8HI 2 "register_operand" "")))]
+  "TARGET_SSE2"
+{
+  rtx t1, t2;
+  bool ok;
+
+  t1 = gen_reg_rtx (V8HImode);
+  emit_insn (gen_sse2_ussubv8hi3 (t1, operands[1], operands[2]));
+  t2 = force_reg (V8HImode, CONST0_RTX (V8HImode));
+
+  operands[3] = gen_rtx_EQ (VOIDmode, t1, t2);
+  operands[4] = t1;
+  operands[5] = t2;
+  ok = ix86_expand_int_vcond (operands, false);
+  gcc_assert (ok);
+  DONE;
+})
+
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;;
 ;; Parallel integral comparisons
Index: testsuite/gcc.dg/vect/vect-reduc-1short.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/testsuite/gcc.dg/vect/vect-reduc-1short.c,v
retrieving revision 1.1
diff -u -p -d -r1.1 vect-reduc-1short.c
--- testsuite/gcc.dg/vect/vect-reduc-1short.c	21 Jun 2005 09:01:59 -0000	1.1
+++ testsuite/gcc.dg/vect/vect-reduc-1short.c	28 Jun 2005 06:12:00 -0000
@@ -47,5 +47,5 @@ int main (void)
   return 0;
 }
 
-/* { dg-final { scan-tree-dump-times "vectorized 3 loops" 1 "vect" { xfail i?86-*-* x86_64-*-* } } } */
+/* { dg-final { scan-tree-dump-times "vectorized 3 loops" 1 "vect" } } */
 /* { dg-final { cleanup-tree-dump "vect" } } */
Index: testsuite/gcc.dg/vect/vect-reduc-2char.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/testsuite/gcc.dg/vect/vect-reduc-2char.c,v
retrieving revision 1.1
diff -u -p -d -r1.1 vect-reduc-2char.c
--- testsuite/gcc.dg/vect/vect-reduc-2char.c	21 Jun 2005 09:01:59 -0000	1.1
+++ testsuite/gcc.dg/vect/vect-reduc-2char.c	28 Jun 2005 06:12:00 -0000
@@ -47,5 +47,5 @@ int main (void)
   return 0 ;
 }
 
-/* { dg-final { scan-tree-dump-times "vectorized 3 loops" 1 "vect" { xfail i?86-*-* x86_64-*-* } } } */
+/* { dg-final { scan-tree-dump-times "vectorized 3 loops" 1 "vect" } } */
 /* { dg-final { cleanup-tree-dump "vect" } } */


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]