[llvm] r325548 - [X86] Mark XOP vpmac* and vpmadc intrinsics as being commutative so that tablegen will generate patterns with the load in operand 0.

[Craig Topper via llvm-commits]

Author: ctopper
Date: Mon Feb 19 19:58:14 2018
New Revision: 325548

URL: http://llvm.org/viewvc/llvm-project?rev=325548&view=rev
Log:
[X86] Mark XOP vpmac* and vpmadc intrinsics as being commutative so that tablegen will generate patterns with the load in operand 0.

This allows loads to be folded during isel without the peephole pass.

Modified:
    llvm/trunk/include/llvm/IR/IntrinsicsX86.td
    llvm/trunk/test/CodeGen/X86/commute-xop.ll

Modified: llvm/trunk/include/llvm/IR/IntrinsicsX86.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/IR/IntrinsicsX86.td?rev=325548&r1=325547&r2=325548&view=diff
==============================================================================
--- llvm/trunk/include/llvm/IR/IntrinsicsX86.td (original)
+++ llvm/trunk/include/llvm/IR/IntrinsicsX86.td Mon Feb 19 19:58:14 2018
@@ -3056,62 +3056,62 @@ let TargetPrefix = "x86" in {  // All in
               GCCBuiltin<"__builtin_ia32_vpmacsdd">,
               Intrinsic<[llvm_v4i32_ty],
                         [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v4i32_ty],
-                        [IntrNoMem]>;
+                        [IntrNoMem, Commutative]>;
   def int_x86_xop_vpmacsdqh :
               GCCBuiltin<"__builtin_ia32_vpmacsdqh">,
               Intrinsic<[llvm_v2i64_ty],
                         [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v2i64_ty],
-                        [IntrNoMem]>;
+                        [IntrNoMem, Commutative]>;
   def int_x86_xop_vpmacsdql :
               GCCBuiltin<"__builtin_ia32_vpmacsdql">,
               Intrinsic<[llvm_v2i64_ty],
                         [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v2i64_ty],
-                        [IntrNoMem]>;
+                        [IntrNoMem, Commutative]>;
   def int_x86_xop_vpmacssdd :
               GCCBuiltin<"__builtin_ia32_vpmacssdd">,
               Intrinsic<[llvm_v4i32_ty],
                         [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v4i32_ty],
-                        [IntrNoMem]>;
+                        [IntrNoMem, Commutative]>;
   def int_x86_xop_vpmacssdqh :
               GCCBuiltin<"__builtin_ia32_vpmacssdqh">,
               Intrinsic<[llvm_v2i64_ty],
                         [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v2i64_ty],
-                        [IntrNoMem]>;
+                        [IntrNoMem, Commutative]>;
   def int_x86_xop_vpmacssdql :
               GCCBuiltin<"__builtin_ia32_vpmacssdql">,
               Intrinsic<[llvm_v2i64_ty],
                         [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v2i64_ty],
-                        [IntrNoMem]>;
+                        [IntrNoMem, Commutative]>;
   def int_x86_xop_vpmacsswd :
               GCCBuiltin<"__builtin_ia32_vpmacsswd">,
               Intrinsic<[llvm_v4i32_ty],
                         [llvm_v8i16_ty, llvm_v8i16_ty, llvm_v4i32_ty],
-                        [IntrNoMem]>;
+                        [IntrNoMem, Commutative]>;
   def int_x86_xop_vpmacssww :
               GCCBuiltin<"__builtin_ia32_vpmacssww">,
               Intrinsic<[llvm_v8i16_ty],
                         [llvm_v8i16_ty, llvm_v8i16_ty, llvm_v8i16_ty],
-                        [IntrNoMem]>;
+                        [IntrNoMem, Commutative]>;
   def int_x86_xop_vpmacswd :
               GCCBuiltin<"__builtin_ia32_vpmacswd">,
               Intrinsic<[llvm_v4i32_ty],
                         [llvm_v8i16_ty, llvm_v8i16_ty, llvm_v4i32_ty],
-                        [IntrNoMem]>;
+                        [IntrNoMem, Commutative]>;
   def int_x86_xop_vpmacsww :
               GCCBuiltin<"__builtin_ia32_vpmacsww">,
               Intrinsic<[llvm_v8i16_ty],
                         [llvm_v8i16_ty, llvm_v8i16_ty, llvm_v8i16_ty],
-                        [IntrNoMem]>;
+                        [IntrNoMem, Commutative]>;
   def int_x86_xop_vpmadcsswd :
               GCCBuiltin<"__builtin_ia32_vpmadcsswd">,
               Intrinsic<[llvm_v4i32_ty],
                         [llvm_v8i16_ty, llvm_v8i16_ty, llvm_v4i32_ty],
-                        [IntrNoMem]>;
+                        [IntrNoMem, Commutative]>;
   def int_x86_xop_vpmadcswd :
               GCCBuiltin<"__builtin_ia32_vpmadcswd">,
               Intrinsic<[llvm_v4i32_ty],
                         [llvm_v8i16_ty, llvm_v8i16_ty, llvm_v4i32_ty],
-                        [IntrNoMem]>;
+                        [IntrNoMem, Commutative]>;
   def int_x86_xop_vpperm :
               GCCBuiltin<"__builtin_ia32_vpperm">,
               Intrinsic<[llvm_v16i8_ty],

Modified: llvm/trunk/test/CodeGen/X86/commute-xop.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/commute-xop.ll?rev=325548&r1=325547&r2=325548&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/commute-xop.ll (original)
+++ llvm/trunk/test/CodeGen/X86/commute-xop.ll Mon Feb 19 19:58:14 2018
@@ -1,6 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx,+xop | FileCheck %s --check-prefix=X32
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+xop | FileCheck %s --check-prefix=X64
+; RUN: llc < %s -disable-peephole -mtriple=i686-unknown-unknown -mattr=+avx,+xop | FileCheck %s --check-prefix=X32
+; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx,+xop | FileCheck %s --check-prefix=X64
 
 define <16 x i8> @commute_fold_vpcomb(<16 x i8>* %a0, <16 x i8> %a1) {
 ; X32-LABEL: commute_fold_vpcomb: