[llvm] r305184 - [X86][SSE] Change memop fragment to inherit from vec128load with local alignment controls
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Mon Jun 12 03:01:27 PDT 2017
Author: rksimon
Date: Mon Jun 12 05:01:27 2017
New Revision: 305184
URL: http://llvm.org/viewvc/llvm-project?rev=305184&view=rev
Log:
[X86][SSE] Change memop fragment to inherit from vec128load with local alignment controls
First possible step towards merging SSE/AVX memory folding pattern fragments.
Also allows us to remove the duplicate non-temporal load logic.
Differential Revision: https://reviews.llvm.org/D33902
Modified:
llvm/trunk/lib/Target/X86/X86InstrFragmentsSIMD.td
Modified: llvm/trunk/lib/Target/X86/X86InstrFragmentsSIMD.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86InstrFragmentsSIMD.td?rev=305184&r1=305183&r2=305184&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86InstrFragmentsSIMD.td (original)
+++ llvm/trunk/lib/Target/X86/X86InstrFragmentsSIMD.td Mon Jun 12 05:01:27 2017
@@ -737,19 +737,15 @@ def alignedloadv8f64 : PatFrag<(ops nod
def alignedloadv8i64 : PatFrag<(ops node:$ptr),
(v8i64 (alignedload512 node:$ptr))>;
-// Like 'load', but uses special alignment checks suitable for use in
+// Like 'vec128load', but uses special alignment checks suitable for use in
// memory operands in most SSE instructions, which are required to
// be naturally aligned on some targets but not on others. If the subtarget
// allows unaligned accesses, match any load, though this may require
// setting a feature bit in the processor (on startup, for example).
// Opteron 10h and later implement such a feature.
-// Avoid non-temporal aligned loads on supported targets.
-def memop : PatFrag<(ops node:$ptr), (load node:$ptr), [{
- return (Subtarget->hasSSEUnalignedMem() ||
- cast<LoadSDNode>(N)->getAlignment() >= 16) &&
- (!Subtarget->hasSSE41() ||
- !(cast<LoadSDNode>(N)->getAlignment() >= 16 &&
- cast<LoadSDNode>(N)->isNonTemporal()));
+def memop : PatFrag<(ops node:$ptr), (vec128load node:$ptr), [{
+ return Subtarget->hasSSEUnalignedMem() ||
+ cast<LoadSDNode>(N)->getAlignment() >= 16;
}]>;
// 128-bit memop pattern fragments
More information about the llvm-commits
mailing list