[llvm] 30e9cdd - [DAG] computeKnownBits - add ISD::AVGCEILU handling

[Simon Pilgrim via llvm-commits]

Author: Simon Pilgrim
Date: 2022-02-16T13:00:15Z
New Revision: 30e9cdd1aa2e60c2ce9a69fe444a2b8d7a3341fe

URL: https://github.com/llvm/llvm-project/commit/30e9cdd1aa2e60c2ce9a69fe444a2b8d7a3341fe
DIFF: https://github.com/llvm/llvm-project/commit/30e9cdd1aa2e60c2ce9a69fe444a2b8d7a3341fe.diff

LOG: [DAG] computeKnownBits - add ISD::AVGCEILU handling

Expand the ISD::AVGCEILU to determine the known bits of the result.

First part of PR53622

Differential Revision: https://reviews.llvm.org/D119629

Added: 
    

Modified: 
    llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
    llvm/test/CodeGen/X86/combine-pavg.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index b3a929050d7cd..32d961d9bf044 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -3129,6 +3129,16 @@ KnownBits SelectionDAG::computeKnownBits(SDValue Op, const APInt &DemandedElts,
     Known = KnownBits::udiv(Known, Known2);
     break;
   }
+  case ISD::AVGCEILU: {
+    Known = computeKnownBits(Op.getOperand(0), DemandedElts, Depth + 1);
+    Known2 = computeKnownBits(Op.getOperand(1), DemandedElts, Depth + 1);
+    Known = Known.zext(BitWidth + 1);
+    Known2 = Known2.zext(BitWidth + 1);
+    KnownBits One = KnownBits::makeConstant(APInt(1, 1));
+    Known = KnownBits::computeForAddCarry(Known, Known2, One);
+    Known = Known.extractBits(BitWidth, 1);
+    break;
+  }
   case ISD::SELECT:
   case ISD::VSELECT:
     Known = computeKnownBits(Op.getOperand(2), DemandedElts, Depth+1);

diff  --git a/llvm/test/CodeGen/X86/combine-pavg.ll b/llvm/test/CodeGen/X86/combine-pavg.ll
index bfeebb473c74b..493a1af9a591c 100644
--- a/llvm/test/CodeGen/X86/combine-pavg.ll
+++ b/llvm/test/CodeGen/X86/combine-pavg.ll
@@ -1,7 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=SSE
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX
 
 declare <16 x i8> @llvm.x86.sse2.pavg.b(<16 x i8>, <16 x i8>) nounwind readnone
 declare <8 x i16> @llvm.x86.sse2.pavg.w(<8 x i16>, <8 x i16>) nounwind readnone
@@ -21,7 +21,6 @@ define <16 x i8> @combine_pavgb_self(<16 x i8> %a0) {
   ret <16 x i8> %1
 }
 
-; TODO: Failure to remove masks as we know the upper bits are zero
 define <16 x i8> @combine_pavgw_knownbits(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> %a2, <8 x i16> %a3) {
 ; SSE-LABEL: combine_pavgw_knownbits:
 ; SSE:       # %bb.0:
@@ -32,41 +31,20 @@ define <16 x i8> @combine_pavgw_knownbits(<8 x i16> %a0, <8 x i16> %a1, <8 x i16
 ; SSE-NEXT:    pand %xmm4, %xmm2
 ; SSE-NEXT:    pand %xmm4, %xmm3
 ; SSE-NEXT:    pavgw %xmm2, %xmm3
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]
-; SSE-NEXT:    pand %xmm1, %xmm3
-; SSE-NEXT:    pand %xmm1, %xmm0
 ; SSE-NEXT:    packuswb %xmm3, %xmm0
 ; SSE-NEXT:    retq
 ;
-; AVX1-LABEL: combine_pavgw_knownbits:
-; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [31,31,31,31,31,31,31,31]
-; AVX1-NEXT:    vpand %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm4, %xmm1, %xmm1
-; AVX1-NEXT:    vpavgw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm4, %xmm2, %xmm1
-; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm2
-; AVX1-NEXT:    vpavgw %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]
-; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: combine_pavgw_knownbits:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [31,31,31,31,31,31,31,31]
-; AVX2-NEXT:    vpand %xmm4, %xmm0, %xmm0
-; AVX2-NEXT:    vpand %xmm4, %xmm1, %xmm1
-; AVX2-NEXT:    vpavgw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpand %xmm4, %xmm2, %xmm1
-; AVX2-NEXT:    vpand %xmm4, %xmm3, %xmm2
-; AVX2-NEXT:    vpavgw %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
-; AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpand %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    retq
+; AVX-LABEL: combine_pavgw_knownbits:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm4 = [31,31,31,31,31,31,31,31]
+; AVX-NEXT:    vpand %xmm4, %xmm0, %xmm0
+; AVX-NEXT:    vpand %xmm4, %xmm1, %xmm1
+; AVX-NEXT:    vpavgw %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpand %xmm4, %xmm2, %xmm1
+; AVX-NEXT:    vpand %xmm4, %xmm3, %xmm2
+; AVX-NEXT:    vpavgw %xmm2, %xmm1, %xmm1
+; AVX-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    retq
   %m0 = and <8 x i16> %a0, <i16 31, i16 31, i16 31, i16 31, i16 31, i16 31, i16 31, i16 31>
   %m1 = and <8 x i16> %a1, <i16 31, i16 31, i16 31, i16 31, i16 31, i16 31, i16 31, i16 31>
   %m2 = and <8 x i16> %a2, <i16 31, i16 31, i16 31, i16 31, i16 31, i16 31, i16 31, i16 31>