[PATCH] D114765: [X86][FP16] Only generate approximate rsqrt when Reciprocal is true for half type

[Phoebe Wang via Phabricator via llvm-commits]

pengfei created this revision.
pengfei added reviewers: craig.topper, LuoYuanke, RKSimon, spatel, LiuChen3.
Herald added a subscriber: hiraditya.
pengfei requested review of this revision.
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

We have reasonable fast sqrt and accurate rsqrt for half type due to the
limited fractions. So neither do we need multi steps refinement for
rsqrt nor replace sqrt by rsqrt.
This fixes a correctness issue when `RefinementSteps` = 0.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D114765

Files:
  llvm/lib/Target/X86/X86ISelLowering.cpp
  llvm/test/CodeGen/X86/avx512fp16vl-intrinsics.ll


Index: llvm/test/CodeGen/X86/avx512fp16vl-intrinsics.ll
===================================================================
--- llvm/test/CodeGen/X86/avx512fp16vl-intrinsics.ll
+++ llvm/test/CodeGen/X86/avx512fp16vl-intrinsics.ll
@@ -969,6 +969,15 @@
   ret <8 x half> %2
 }
 
+define <8 x half> @test_sqrt_ph_128_fast2(<8 x half> %a0, <8 x half> %a1) {
+; CHECK-LABEL: test_sqrt_ph_128_fast2:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsqrtph %xmm0, %xmm0
+; CHECK-NEXT:    retq
+  %1 = call fast <8 x half> @llvm.sqrt.v8f16(<8 x half> %a0)
+  ret <8 x half> %1
+}
+
 define <8 x half> @test_mask_sqrt_ph_128(<8 x half> %a0, <8 x half> %passthru, i8 %mask) {
 ; CHECK-LABEL: test_mask_sqrt_ph_128:
 ; CHECK:       # %bb.0:
Index: llvm/lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- llvm/lib/Target/X86/X86ISelLowering.cpp
+++ llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -23232,7 +23232,7 @@
   }
 
   if (VT.getScalarType() == MVT::f16 && isTypeLegal(VT) &&
-      Subtarget.hasFP16()) {
+      Subtarget.hasFP16() && Reciprocal) {
     if (RefinementSteps == ReciprocalEstimate::Unspecified)
       RefinementSteps = 0;
 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D114765.390543.patch
Type: text/x-patch
Size: 1184 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20211130/cf295b88/attachment.bin>