[PATCH] D89461: [CostModel] remove cost-kind predicate for ctlz/cttz intrinsics in basic TTI implementation

[Sanjay Patel via Phabricator via llvm-commits]

spatel created this revision.
spatel added reviewers: samparker, RKSimon, lebedev.ri, dmgreen, fhahn, craig.topper.
Herald added subscribers: pengfei, mcrosier.
Herald added a project: LLVM.
spatel requested review of this revision.

The cost modeling for intrinsics is a patchwork based on different expectations from the callers, so it's a mess. I'm hoping to untangle this to allow canonicalization to the new min/max intrinsics in IR.
The general goal is to remove the cost-kind restriction here in the basic implementation class. Ie, if some intrinsic has throughput cost of 104, assume that it has the same size, latency, and blended costs. Effectively, an intrinsic with cost N is composed of N simple instructions. If that's not correct, the target should provide a more accurate override.

The x86-64 SSE2 subtarget cost diffs require explanation:

1. The scalar ctlz/cttz are assuming "BSR+XOR+CMOV" or "TEST+BSF+CMOV/BRANCH", so not cheap.
2. The 128-bit SSE vector width versions assume cost of 18 or 26 (no explanation provided in the tables, but this corresponds to a bunch of shift/logic/compare).
3. The 512-bit vectors in the test file are scaled up by a factor of 4 from the legal vector width costs.
4. The plain latency cost-kind is not affected in this patch because that calc is diverted before we get to getIntrinsicInstrCost().


https://reviews.llvm.org/D89461

Files:
  llvm/include/llvm/CodeGen/BasicTTIImpl.h
  llvm/test/Analysis/CostModel/X86/intrinsic-cost-kinds.ll


Index: llvm/test/Analysis/CostModel/X86/intrinsic-cost-kinds.ll
===================================================================
--- llvm/test/Analysis/CostModel/X86/intrinsic-cost-kinds.ll
+++ llvm/test/Analysis/CostModel/X86/intrinsic-cost-kinds.ll
@@ -95,13 +95,13 @@
 ; LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
 ; SIZE-LABEL: 'cttz'
-; SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s = call i32 @llvm.cttz.i32(i32 %a, i1 false)
-; SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %va, i1 false)
+; SIZE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %s = call i32 @llvm.cttz.i32(i32 %a, i1 false)
+; SIZE-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %va, i1 false)
 ; SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
 ; SIZE_LATE-LABEL: 'cttz'
-; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s = call i32 @llvm.cttz.i32(i32 %a, i1 false)
-; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %va, i1 false)
+; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %s = call i32 @llvm.cttz.i32(i32 %a, i1 false)
+; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %va, i1 false)
 ; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %s = call i32 @llvm.cttz.i32(i32 %a, i1 false)
@@ -121,13 +121,13 @@
 ; LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
 ; SIZE-LABEL: 'ctlz'
-; SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s = call i32 @llvm.ctlz.i32(i32 %a, i1 true)
-; SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %va, i1 true)
+; SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %s = call i32 @llvm.ctlz.i32(i32 %a, i1 true)
+; SIZE-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %v = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %va, i1 true)
 ; SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
 ; SIZE_LATE-LABEL: 'ctlz'
-; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s = call i32 @llvm.ctlz.i32(i32 %a, i1 true)
-; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %va, i1 true)
+; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %s = call i32 @llvm.ctlz.i32(i32 %a, i1 true)
+; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %v = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %va, i1 true)
 ; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %s = call i32 @llvm.ctlz.i32(i32 %a, i1 true)
Index: llvm/include/llvm/CodeGen/BasicTTIImpl.h
===================================================================
--- llvm/include/llvm/CodeGen/BasicTTIImpl.h
+++ llvm/include/llvm/CodeGen/BasicTTIImpl.h
@@ -1150,21 +1150,15 @@
       break;
 
     case Intrinsic::cttz:
-      // FIXME: all cost kinds should default to the same thing?
-      if (CostKind != TTI::TCK_RecipThroughput) {
-        if (getTLI()->isCheapToSpeculateCttz())
-          return TargetTransformInfo::TCC_Basic;
-        return BaseT::getIntrinsicInstrCost(ICA, CostKind);
-      }
+      // FIXME: If necessary, this should go in target-specific overrides.
+      if (VF == 1 && RetVF == 1 && getTLI()->isCheapToSpeculateCttz())
+        return TargetTransformInfo::TCC_Basic;
       break;
 
     case Intrinsic::ctlz:
-      // FIXME: all cost kinds should default to the same thing?
-      if (CostKind != TTI::TCK_RecipThroughput) {
-        if (getTLI()->isCheapToSpeculateCtlz())
-          return TargetTransformInfo::TCC_Basic;
-        return BaseT::getIntrinsicInstrCost(ICA, CostKind);
-      }
+      // FIXME: If necessary, this should go in target-specific overrides.
+      if (VF == 1 && RetVF == 1 && getTLI()->isCheapToSpeculateCtlz())
+        return TargetTransformInfo::TCC_Basic;
       break;
 
     case Intrinsic::memcpy:


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D89461.298358.patch
Type: text/x-patch
Size: 4440 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20201015/025d7e49/attachment.bin>