diff options
| author | Craig Topper <craig.topper@intel.com> | 2018-02-20 03:58:14 +0000 |
|---|---|---|
| committer | Craig Topper <craig.topper@intel.com> | 2018-02-20 03:58:14 +0000 |
| commit | 41f64c3204e90b8d82ea06804f954e6624635ae1 (patch) | |
| tree | b893d6f9d9cd666dc6e4a8fbb3b7b1fd99d84858 | |
| parent | a05ed17316f326315587d09e3a782e215d686a23 (diff) | |
| download | bcm5719-llvm-41f64c3204e90b8d82ea06804f954e6624635ae1.tar.gz bcm5719-llvm-41f64c3204e90b8d82ea06804f954e6624635ae1.zip | |
[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.
llvm-svn: 325548
| -rw-r--r-- | llvm/include/llvm/IR/IntrinsicsX86.td | 24 | ||||
| -rw-r--r-- | llvm/test/CodeGen/X86/commute-xop.ll | 4 |
2 files changed, 14 insertions, 14 deletions
diff --git a/llvm/include/llvm/IR/IntrinsicsX86.td b/llvm/include/llvm/IR/IntrinsicsX86.td index a4e7b03a002..ad3f2f1f121 100644 --- a/llvm/include/llvm/IR/IntrinsicsX86.td +++ b/llvm/include/llvm/IR/IntrinsicsX86.td @@ -3056,62 +3056,62 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". 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], diff --git a/llvm/test/CodeGen/X86/commute-xop.ll b/llvm/test/CodeGen/X86/commute-xop.ll index 3dfb24db7fb..606ff126dad 100644 --- a/llvm/test/CodeGen/X86/commute-xop.ll +++ b/llvm/test/CodeGen/X86/commute-xop.ll @@ -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: |
