[PATCH] D94693: Improve KnownBits analyses for SMIN/SMAX DAG nodes.

[Ayman Musa via Phabricator via llvm-commits]

aymanmus created this revision.
aymanmus added reviewers: RKSimon, ArturGainullin, foad.
Herald added subscribers: pengfei, hiraditya.
aymanmus requested review of this revision.
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

In computeKnownBits method for SMIN & SMAX opcodes, an early exit can miss opportunities for valuable information.
The code terminates the function in case no bits were known from the 1st operand, while sine valuable information can be extracted about the result's bits from the 2nd operand alone.

For example:

- smax x, 0  --> MSB of x is always 0.
- smin x, -1  --> MSB of x is always 1.


https://reviews.llvm.org/D94693

Files:
  llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
  llvm/test/CodeGen/X86/avx512-trunc.ll
  llvm/test/CodeGen/X86/known-bits-vector.ll


Index: llvm/test/CodeGen/X86/known-bits-vector.ll
===================================================================
--- llvm/test/CodeGen/X86/known-bits-vector.ll
+++ llvm/test/CodeGen/X86/known-bits-vector.ll
@@ -435,11 +435,7 @@
 ; X32-NEXT:    vpminsd {{\.LCPI.*}}, %xmm0, %xmm0
 ; X32-NEXT:    vpmaxsd {{\.LCPI.*}}, %xmm0, %xmm0
 ; X32-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,3,3]
-; X32-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0],mem[1],xmm0[2],mem[3],xmm0[4],mem[5],xmm0[6],mem[7]
-; X32-NEXT:    vpsrld $16, %xmm0, %xmm0
-; X32-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],mem[1],xmm0[2],mem[3],xmm0[4],mem[5],xmm0[6],mem[7]
-; X32-NEXT:    vsubps {{\.LCPI.*}}, %xmm0, %xmm0
-; X32-NEXT:    vaddps %xmm0, %xmm1, %xmm0
+; X32-NEXT:    vcvtdq2ps %xmm0, %xmm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: knownbits_smax_smin_shuffle_uitofp:
@@ -447,11 +443,7 @@
 ; X64-NEXT:    vpminsd {{.*}}(%rip), %xmm0, %xmm0
 ; X64-NEXT:    vpmaxsd {{.*}}(%rip), %xmm0, %xmm0
 ; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,3,3]
-; X64-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0],mem[1],xmm0[2],mem[3],xmm0[4],mem[5],xmm0[6],mem[7]
-; X64-NEXT:    vpsrld $16, %xmm0, %xmm0
-; X64-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],mem[1],xmm0[2],mem[3],xmm0[4],mem[5],xmm0[6],mem[7]
-; X64-NEXT:    vsubps {{.*}}(%rip), %xmm0, %xmm0
-; X64-NEXT:    vaddps %xmm0, %xmm1, %xmm0
+; X64-NEXT:    vcvtdq2ps %xmm0, %xmm0
 ; X64-NEXT:    retq
   %1 = call <4 x i32> @llvm.x86.sse41.pminsd(<4 x i32> %a0, <4 x i32> <i32 0, i32 -65535, i32 -65535, i32 0>)
   %2 = call <4 x i32> @llvm.x86.sse41.pmaxsd(<4 x i32> %1, <4 x i32> <i32 65535, i32 -1, i32 -1, i32 131071>)
Index: llvm/test/CodeGen/X86/avx512-trunc.ll
===================================================================
--- llvm/test/CodeGen/X86/avx512-trunc.ll
+++ llvm/test/CodeGen/X86/avx512-trunc.ll
@@ -1007,21 +1007,14 @@
 define void @negative_test1_smax_usat_trunc_wb_256_mem(<16 x i16> %i, <16 x i8>* %res) {
 ; KNL-LABEL: negative_test1_smax_usat_trunc_wb_256_mem:
 ; KNL:       ## %bb.0:
-; KNL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; KNL-NEXT:    vpmaxsw %ymm1, %ymm0, %ymm0
-; KNL-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
-; KNL-NEXT:    vpminsw %ymm1, %ymm0, %ymm0
-; KNL-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
+; KNL-NEXT:    vpbroadcastd {{.*#+}} zmm0 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
 ; KNL-NEXT:    vpmovdb %zmm0, (%rdi)
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: negative_test1_smax_usat_trunc_wb_256_mem:
 ; SKX:       ## %bb.0:
-; SKX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; SKX-NEXT:    vpmaxsw %ymm1, %ymm0, %ymm0
-; SKX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
-; SKX-NEXT:    vpminsw %ymm1, %ymm0, %ymm0
+; SKX-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; SKX-NEXT:    vpmovwb %ymm0, (%rdi)
 ; SKX-NEXT:    vzeroupper
 ; SKX-NEXT:    retq
Index: llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
===================================================================
--- llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -3416,8 +3416,8 @@
     }
 
     Known = computeKnownBits(Op.getOperand(0), DemandedElts, Depth + 1);
-    if (Known.isUnknown()) break; // Early-out
     Known2 = computeKnownBits(Op.getOperand(1), DemandedElts, Depth + 1);
+    if (Known.isUnknown() && Known2.isUnknown()) break; // Early-out
     if (IsMax)
       Known = KnownBits::smax(Known, Known2);
     else


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D94693.316662.patch
Type: text/x-patch
Size: 3638 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210114/5f49e11f/attachment.bin>