[llvm] AMDGPU: Enable selection of strict_fp16_to_fp (PR #175834)

[Matt Arsenault via llvm-commits]

https://github.com/arsenm created https://github.com/llvm/llvm-project/pull/175834

This avoids regressions when softPromoteHalfType is switched on.

>From fe0a9a6a2fe31536b9405b2ac7da13cfa9526314 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Tue, 13 Jan 2026 21:52:44 +0100
Subject: [PATCH] AMDGPU: Enable selection of strict_fp16_to_fp

This avoids regressions when softPromoteHalfType is switched on.
---
 llvm/lib/Target/AMDGPU/SIInstructions.td | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td
index 22564b8347110..e06bc912113a8 100644
--- a/llvm/lib/Target/AMDGPU/SIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SIInstructions.td
@@ -1357,22 +1357,22 @@ multiclass f16_to_fp_Pats<Instruction cvt_f16_f32_inst_e64, Instruction cvt_f32_
   >;
 
   def : GCNPat <
-    (f32 (f16_to_fp (and_oneuse i32:$src0, 0x7fff))),
+    (f32 (any_f16_to_fp (and_oneuse i32:$src0, 0x7fff))),
     (cvt_f32_f16_inst_e64 SRCMODS.ABS, $src0)
   >;
 
   def : GCNPat <
-    (f32 (f16_to_fp (i32 (srl_oneuse (and_oneuse i32:$src0, 0x7fff0000), (i32 16))))),
+    (f32 (any_f16_to_fp (i32 (srl_oneuse (and_oneuse i32:$src0, 0x7fff0000), (i32 16))))),
     (cvt_f32_f16_inst_e64 SRCMODS.ABS, (i32 (V_LSHRREV_B32_e64 (i32 16), i32:$src0)))
   >;
 
   def : GCNPat <
-    (f32 (f16_to_fp (or_oneuse i32:$src0, 0x8000))),
+    (f32 (any_f16_to_fp (or_oneuse i32:$src0, 0x8000))),
     (cvt_f32_f16_inst_e64 SRCMODS.NEG_ABS, $src0)
   >;
 
   def : GCNPat <
-    (f32 (f16_to_fp (xor_oneuse i32:$src0, 0x8000))),
+    (f32 (any_f16_to_fp (xor_oneuse i32:$src0, 0x8000))),
     (cvt_f32_f16_inst_e64 SRCMODS.NEG, $src0)
   >;