[llvm] [AArch64] Improve index selection for histograms (PR #111150)
Graham Hunter via llvm-commits
llvm-commits at lists.llvm.org
Tue Oct 8 07:42:07 PDT 2024
================
@@ -267,5 +267,79 @@ define void @histogram_i16_8_lane(ptr %base, <vscale x 8 x i32> %indices, i16 %i
ret void
}
+define void @histogram_i32_zextend(ptr %base, <vscale x 4 x i32> %indices, <vscale x 4 x i1> %mask) #0 {
----------------
huntergr-arm wrote:
Yes, that's pretty much what we want -- if it's possible to use a 32b index type, we'd prefer to do so because we can process twice as much data that way. My idea was to reuse the existing logic for gathers and scatters (which does the same thing for the same reasons), and I believe handles more cases than just extends.
But I'll take incremental progress with just supporting extends to begin with :)
https://github.com/llvm/llvm-project/pull/111150
More information about the llvm-commits
mailing list