[llvm] [DAG] visitIS_FPCLASS - fold to constant when result is fully determined by KnownFPClass (PR #193737)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-risc-v

Author: Xinlong Chen (Xinlong-Chen)

<details>
<summary>Changes</summary>

This PR teaches `DAGCombiner::visitIS_FPCLASS` to fold directly to constant `false`/`true` based on the source's `KnownFPClass`, instead of only narrowing the test mask.

Prep work to help with https://github.com/llvm/llvm-project/pull/193672

---
Full diff: https://github.com/llvm/llvm-project/pull/193737.diff


1 Files Affected:

- (modified) llvm/test/CodeGen/RISCV/combine-is_fpclass.ll (+13) 


``````````diff
diff --git a/llvm/test/CodeGen/RISCV/combine-is_fpclass.ll b/llvm/test/CodeGen/RISCV/combine-is_fpclass.ll
index 71621b8837d6b..d325f9563f6c3 100644
--- a/llvm/test/CodeGen/RISCV/combine-is_fpclass.ll
+++ b/llvm/test/CodeGen/RISCV/combine-is_fpclass.ll
@@ -186,6 +186,19 @@ define i1 @extract_scalable_fabs_isneg(<vscale x 4 x float> %a0) nounwind {
   ret i1 %res
 }
 
+define i1 @fabs_is_nonneg_or_nan(float %x) nounwind {
+; CHECK-LABEL: fabs_is_nonneg_or_nan:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fabs.s fa5, fa0
+; CHECK-NEXT:    fclass.s a0, fa5
+; CHECK-NEXT:    andi a0, a0, 1008
+; CHECK-NEXT:    snez a0, a0
+; CHECK-NEXT:    ret
+  %a = call float @llvm.fabs.f32(float %x)
+  %res = call i1 @llvm.is.fpclass.f32(float %a, i32 963)  ; 0x3C3 = any positive | nan
+  ret i1 %res
+}
+
 define <vscale x 4 x i1> @splat_constant_is_pos_normal() {
 ; CHECK-LABEL: splat_constant_is_pos_normal:
 ; CHECK:       # %bb.0:

``````````

</details>


https://github.com/llvm/llvm-project/pull/193737