From b1983bab68a9b1acdd3a4e49ab07d1b6a3b74ce2 Mon Sep 17 00:00:00 2001
From: Wunkolo <Wunkolo@gmail.com>
Date: Thu, 19 Mar 2026 14:49:39 -0700
Subject: [PATCH] [x64] Add `GFNI` optimization for single-byte vector splats

Use the `vgf2p8affineqb` instruction to splat byte-values across a whole
register without having to touch memory.
---
 src/xenia/cpu/backend/x64/x64_emitter.cc | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/src/xenia/cpu/backend/x64/x64_emitter.cc b/src/xenia/cpu/backend/x64/x64_emitter.cc
index c7fd92e98..5da8f6adc 100644
--- a/src/xenia/cpu/backend/x64/x64_emitter.cc
+++ b/src/xenia/cpu/backend/x64/x64_emitter.cc
@@ -1313,6 +1313,12 @@ void X64Emitter::LoadConstantXmm(Xbyak::Xmm dest, const vec128_t& v) {
       }
 
       if (all_equal_bytes) {
+        if (IsFeatureEnabled(kX64EmitGFNI)) {
+          vpxor(dest, dest);
+          vgf2p8affineqb(dest, dest, dest, firstbyte);
+          return;
+        }
+
         void* bval = FindByteConstantOffset(firstbyte);
 
         if (bval) {