From 91fdab4f89de688d490d02ee8b4267bf5cf1c138 Mon Sep 17 00:00:00 2001
From: Wunkolo <Wunkolo@gmail.com>
Date: Sun, 12 May 2024 21:15:33 -0700
Subject: [PATCH] [a64] Fix `OPCODE_CNTLZ`

8 and 16 bit CNTLZ needs its bit-count fixed to its original element-type
---
 a64_sequences.cc | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/a64_sequences.cc b/a64_sequences.cc
index 9b58a8409..e33e7ef5c 100644
--- a/a64_sequences.cc
+++ b/a64_sequences.cc
@@ -2740,16 +2740,18 @@ EMITTER_OPCODE_TABLE(OPCODE_BYTE_SWAP, BYTE_SWAP_I16, BYTE_SWAP_I32,
 // ============================================================================
 struct CNTLZ_I8 : Sequence<CNTLZ_I8, I<OPCODE_CNTLZ, I8Op, I8Op>> {
   static void Emit(A64Emitter& e, const EmitArgType& i) {
-    // No 8bit lzcnt, so do 16 and sub 8.
+    // No 8bit lzcnt, so do 32 and sub 24.
     e.UXTB(i.dest, i.src1);
     e.CLZ(i.dest, i.dest);
-    e.SUB(i.dest.reg(), i.dest.reg(), 8);
+    e.SUB(i.dest.reg(), i.dest.reg(), 24);
   }
 };
 struct CNTLZ_I16 : Sequence<CNTLZ_I16, I<OPCODE_CNTLZ, I8Op, I16Op>> {
   static void Emit(A64Emitter& e, const EmitArgType& i) {
+    // No 16bit lzcnt, so do 32 and sub 16.
     e.UXTH(i.dest, i.src1);
     e.CLZ(i.dest, i.dest);
+    e.SUB(i.dest.reg(), i.dest.reg(), 16);
   }
 };
 struct CNTLZ_I32 : Sequence<CNTLZ_I32, I<OPCODE_CNTLZ, I8Op, I32Op>> {