[llvm] r223996 - [X86] Add back AVX2 VR256 PMOVX patterns.
Ahmed Bougacha
ahmed.bougacha at gmail.com
Wed Dec 10 20:32:17 PST 2014
Author: ab
Date: Wed Dec 10 22:32:17 2014
New Revision: 223996
URL: http://llvm.org/viewvc/llvm-project?rev=223996&view=rev
Log:
[X86] Add back AVX2 VR256 PMOVX patterns.
We can't reach those from zext, but other parts of the backend (the shuffle
lowering) generate 256-bit VZEXT nodes.
Fixes PR21876.
Modified:
llvm/trunk/lib/Target/X86/X86InstrSSE.td
Modified: llvm/trunk/lib/Target/X86/X86InstrSSE.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86InstrSSE.td?rev=223996&r1=223995&r2=223996&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86InstrSSE.td (original)
+++ llvm/trunk/lib/Target/X86/X86InstrSSE.td Wed Dec 10 22:32:17 2014
@@ -6177,6 +6177,22 @@ multiclass SS41I_pmovx_avx2_patterns<str
def : Pat<(v4i64 (ExtOp (v4i32 VR128:$src))),
(!cast<I>(OpcPrefix#DQYrr) VR128:$src)>;
+ // On AVX2, we also support 256bit inputs.
+ def : Pat<(v16i16 (ExtOp (v32i8 VR256:$src))),
+ (!cast<I>(OpcPrefix#BWYrr) (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+ def : Pat<(v8i32 (ExtOp (v32i8 VR256:$src))),
+ (!cast<I>(OpcPrefix#BDYrr) (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+ def : Pat<(v4i64 (ExtOp (v32i8 VR256:$src))),
+ (!cast<I>(OpcPrefix#BQYrr) (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+
+ def : Pat<(v8i32 (ExtOp (v16i16 VR256:$src))),
+ (!cast<I>(OpcPrefix#WDYrr) (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+ def : Pat<(v4i64 (ExtOp (v16i16 VR256:$src))),
+ (!cast<I>(OpcPrefix#WQYrr) (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+
+ def : Pat<(v4i64 (ExtOp (v8i32 VR256:$src))),
+ (!cast<I>(OpcPrefix#DQYrr) (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+
// AVX2 Register-Memory patterns
def : Pat<(v16i16 (ExtOp (bc_v16i8 (loadv2i64 addr:$src)))),
(!cast<I>(OpcPrefix#BWYrm) addr:$src)>;
More information about the llvm-commits
mailing list