[AMDGPU][CodeGen] use vt in VGPRimm pattern (#131016)

broxigarchen · web-flow · commit cb73271be12a · 2025-03-13T07:58:19.000+07:00
There seems to be a typo error introduced by 2033767 Correct this pattern to use vt.
diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td
@@ -2268,7 +2268,7 @@ foreach pred = [NotHasTrue16BitInsts, UseFakeTrue16Insts] in {
 
   foreach vt = [f16, bf16] in {
     def : GCNPat <
-      (VGPRImm<(f16 fpimm)>:$imm),
+      (VGPRImm<(vt fpimm)>:$imm),
       (V_MOV_B32_e32 (vt (bitcast_fpimm_to_i32 $imm)))
     >;
   }

Original file line number	Diff line number	Diff line change
`@@ -2268,7 +2268,7 @@ foreach pred = [NotHasTrue16BitInsts, UseFakeTrue16Insts] in {`
`2268`	`2268`
`2269`	`2269`	`foreach vt = [f16, bf16] in {`
`2270`	`2270`	`def : GCNPat <`
`2271`		`- (VGPRImm<(f16 fpimm)>:$imm),`
	`2271`	`+ (VGPRImm<(vt fpimm)>:$imm),`
`2272`	`2272`	`(V_MOV_B32_e32 (vt (bitcast_fpimm_to_i32 $imm)))`
`2273`	`2273`	`>;`
`2274`	`2274`	`}`