Re: [PATCH, rs6000] (v2) Gimple folding for vec_madd()

[Will Schmidt]

Hi,

[PATCH, rs6000] (v2) Gimple folding for vec_madd()
    
Add support for gimple folding of the vec_madd() (vector multiply-add)
intrinsics.   Per earlier feedback and education, this now includes
the addition of a "define_expand fmav8hi4" in altivec.md.
    
Testcase coverage is provided by the existing tests as
 gcc.target/powerpc/fold-vec-madd-*.c
    
Sniff-tests passed. Regtests will be kicked off shortly. OK for trunk?
(pending successful test results, of course:-) )
    
Thanks,
-Will
    
[gcc]
    
2017-10-26  Will Schmidt <will_schm...@vnet.ibm.com>
    
        * config/rs6000/rs6000.c: (rs6000_gimple_fold_builtin) Add support for
          gimple folding of vec_madd() intrinsics.
        * config/rs6000/altivec.md: Add define_expand fmav8hi4

diff --git a/gcc/config/rs6000/altivec.md b/gcc/config/rs6000/altivec.md
index 6ea529b..36e6ddd 100644
--- a/gcc/config/rs6000/altivec.md
+++ b/gcc/config/rs6000/altivec.md
@@ -943,10 +943,22 @@
                   (match_operand:V8HI 3 "register_operand" "v")))]
   "TARGET_ALTIVEC"
   "vmladduhm %0,%1,%2,%3"
   [(set_attr "type" "veccomplex")])
 
+(define_expand "fmav8hi4"
+  [(use (match_operand:V8HI 0 "register_operand" ""))
+   (use (match_operand:V8HI 1 "register_operand" ""))
+   (use (match_operand:V8HI 2 "register_operand" ""))
+   (use (match_operand:V8HI 3 "register_operand" ""))]
+   "TARGET_ALTIVEC"
+{
+  emit_insn (gen_altivec_vmladduhm (operands[0], operands[1],
+                                   operands[2], operands[3]));
+  DONE;
+})
+
 (define_expand "altivec_vmrghb"
   [(use (match_operand:V16QI 0 "register_operand" ""))
    (use (match_operand:V16QI 1 "register_operand" ""))
    (use (match_operand:V16QI 2 "register_operand" ""))]
   "TARGET_ALTIVEC"
diff --git a/gcc/config/rs6000/rs6000.c b/gcc/config/rs6000/rs6000.c
index 4837e14..1cd4278 100644
--- a/gcc/config/rs6000/rs6000.c
+++ b/gcc/config/rs6000/rs6000.c
@@ -16606,10 +16606,25 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi)
                                           build_int_cst (arg2_type, 0)), arg0);
         gimple_set_location (g, loc);
         gsi_replace (gsi, g, true);
         return true;
       }
+    /* vec_madd (Float) */
+    case ALTIVEC_BUILTIN_VMADDFP:
+    case VSX_BUILTIN_XVMADDDP:
+    case ALTIVEC_BUILTIN_VMLADDUHM:
+      {
+       arg0 = gimple_call_arg (stmt, 0);
+       arg1 = gimple_call_arg (stmt, 1);
+       tree arg2 = gimple_call_arg (stmt, 2);
+       lhs = gimple_call_lhs (stmt);
+       gimple *g = gimple_build_assign (lhs, FMA_EXPR , arg0, arg1, arg2);
+       gimple_set_location (g, gimple_location (stmt));
+       gsi_replace (gsi, g, true);
+       return true;
+      }
+
     default:
        if (TARGET_DEBUG_BUILTIN)
           fprintf (stderr, "gimple builtin intrinsic not matched:%d %s %s\n",
                    fn_code, fn_name1, fn_name2);
       break;