[llvm] [X86] X86FixupVectorConstants - load+zero vector constants that can be stored in a truncated form (PR #80428)

Simon Pilgrim via llvm-commits llvm-commits at lists.llvm.org
Fri Feb 2 07:08:20 PST 2024


================
@@ -109,13 +109,13 @@ define <64 x i8> @shuffle_v64i8_79_00_01_02_03_04_05_06_07_08_09_10_11_12_13_14_
 define <64 x i8> @shuffle_v64i8_0zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz(<64 x i8> %a) {
 ; AVX512F-LABEL: shuffle_v64i8_0zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vmovd {{.*#+}} xmm1 = [255,0,0,0]
----------------
RKSimon wrote:

I guess it comes down to whether we want to save those 16-bits in the constant pool in exchange for a shuffle uop

https://github.com/llvm/llvm-project/pull/80428


More information about the llvm-commits mailing list