[LV][VPlan] Add fast flags for selectRecipe

[LiqinWeng]

Change the inheritance of class VPWidenSelectRecipe to class VPRecipeWithIRFlags, which allows recipe of the select to pass the fastmath flags.The patch of #119847 will add the fastmath flag to for recipe

[llvmbot]

@llvm/pr-subscribers-vectorizers

@llvm/pr-subscribers-llvm-transforms

Author: LiqinWeng (LiqinWeng)

Changes

The patch of #119847 will add the fastmath flag to for recipe

---

Full diff: https://github.com/llvm/llvm-project/pull/121023.diff

1 Files Affected:

• (modified) llvm/lib/Transforms/Vectorize/VPlan.h (+2-3)

diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index 6486c6745a6800..4f457e9876358b 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -1813,11 +1813,10 @@ class VPHistogramRecipe : public VPRecipeBase {
 };
 
 /// A recipe for widening select instructions.
-struct VPWidenSelectRecipe : public VPSingleDefRecipe {
+struct VPWidenSelectRecipe : public VPRecipeWithIRFlags {
   template <typename IterT>
   VPWidenSelectRecipe(SelectInst &I, iterator_range<IterT> Operands)
-      : VPSingleDefRecipe(VPDef::VPWidenSelectSC, Operands, &I,
-                          I.getDebugLoc()) {}
+      : VPRecipeWithIRFlags(VPDef::VPWidenSelectSC, Operands, I) {}
 
   ~VPWidenSelectRecipe() override = default;
 

[LiqinWeng]

@Mel-Chen :）

[LiqinWeng]

Any problems with this patch? @Mel-Chen @fhahn

[fhahn]

Could you add a test case with a select with fast-math flags?

[LiqinWeng]

Could you add a test case with a select with fast-math flags?

done, pls give me LG:)

[Mel-Chen]

Don't we need to call setFlags() in execute()?

[LiqinWeng]

Don't we need to call setFlags() in execute()?
Need, fixed and added tests~

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[fhahn]

can you move this to vplan-printing.ll? RISCV triple and other flags shouldn't be needed for the printing test (if they re needed, the test needs to be in the RISCV subdirectory otherwise it will fail when RISCV is not built

[LiqinWeng]

sorry, fixed ~

[fhahn]

Is RISCV triple really needed? Better to use -force-vector-width; if RISCV is really needed, it should be moved to the RISCV subdir.

(Also doesn't need REQUIRES: asserts?)

[LiqinWeng]

thanks,fixed:)

[LiqinWeng]

If there is no problem, can you give me LG:)

[fhahn]

LGTM, thanks

[Mel-Chen]

LG