From 6f2bee667ad9244be65193892961d5d1eed24551 Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Mon, 8 Aug 2022 17:27:05 +0100
Subject: [PATCH] [DAG] canCreateUndefOrPoison - add freeze(bswap(x)) ->
 bswap(freeze(x)) and freeze(bitreverse(x)) -> bitreverse(freeze(x)) support

Both are guaranteed not to create undef/poison
---
 llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp |   4 +-
 llvm/test/CodeGen/X86/freeze-unary.ll          | 195 +------------------------
 2 files changed, 11 insertions(+), 188 deletions(-)

diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index 5406b02..cecc59d 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -4560,10 +4560,12 @@ bool SelectionDAG::canCreateUndefOrPoison(SDValue Op, const APInt &DemandedElts,
 
   unsigned Opcode = Op.getOpcode();
   switch (Opcode) {
+  case ISD::FREEZE:
+  case ISD::BSWAP:
+  case ISD::BITREVERSE:
   case ISD::SIGN_EXTEND:
   case ISD::ZERO_EXTEND:
   case ISD::BITCAST:
-  case ISD::FREEZE:
     return false;
 
   default:
diff --git a/llvm/test/CodeGen/X86/freeze-unary.ll b/llvm/test/CodeGen/X86/freeze-unary.ll
index 96d5010..f4a8ad2 100644
--- a/llvm/test/CodeGen/X86/freeze-unary.ll
+++ b/llvm/test/CodeGen/X86/freeze-unary.ll
@@ -1,6 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=i686-- -mattr=+sse2 | FileCheck %s --check-prefixes=X86
-; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse4.2 | FileCheck %s --check-prefixes=X64
+; RUN: llc < %s -mtriple=i686-- -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,X86
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse4.2 | FileCheck %s --check-prefixes=CHECK,X64
 
 define i32 @freeze_sext(i8 %a0) nounwind {
 ; X86-LABEL: freeze_sext:
@@ -74,15 +74,11 @@ define i32 @freeze_bswap(i32 %a0) nounwind {
 ; X86-LABEL: freeze_bswap:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    bswapl %eax
-; X86-NEXT:    bswapl %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: freeze_bswap:
 ; X64:       # %bb.0:
 ; X64-NEXT:    movl %edi, %eax
-; X64-NEXT:    bswapl %eax
-; X64-NEXT:    bswapl %eax
 ; X64-NEXT:    retq
   %x = call i32 @llvm.bswap.i32(i32 %a0)
   %y = freeze i32 %x
@@ -92,33 +88,9 @@ define i32 @freeze_bswap(i32 %a0) nounwind {
 declare i32 @llvm.bswap.i32(i32)
 
 define <4 x i32> @freeze_bswap_vec(<4 x i32> %a0) nounwind {
-; X86-LABEL: freeze_bswap_vec:
-; X86:       # %bb.0:
-; X86-NEXT:    pxor %xmm1, %xmm1
-; X86-NEXT:    movdqa %xmm0, %xmm2
-; X86-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15]
-; X86-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[3,2,1,0,4,5,6,7]
-; X86-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,6,5,4]
-; X86-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; X86-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,2,1,0,4,5,6,7]
-; X86-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,5,4]
-; X86-NEXT:    packuswb %xmm2, %xmm0
-; X86-NEXT:    movdqa %xmm0, %xmm2
-; X86-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15]
-; X86-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[3,2,1,0,4,5,6,7]
-; X86-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,6,5,4]
-; X86-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; X86-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,2,1,0,4,5,6,7]
-; X86-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,5,4]
-; X86-NEXT:    packuswb %xmm2, %xmm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: freeze_bswap_vec:
-; X64:       # %bb.0:
-; X64-NEXT:    movdqa {{.*#+}} xmm1 = [3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12]
-; X64-NEXT:    pshufb %xmm1, %xmm0
-; X64-NEXT:    pshufb %xmm1, %xmm0
-; X64-NEXT:    retq
+; CHECK-LABEL: freeze_bswap_vec:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    ret{{[l|q]}}
   %x = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a0)
   %y = freeze <4 x i32> %x
   %z = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %y)
@@ -130,79 +102,11 @@ define i32 @freeze_bitreverse(i32 %a0) nounwind {
 ; X86-LABEL: freeze_bitreverse:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    bswapl %eax
-; X86-NEXT:    movl %eax, %ecx
-; X86-NEXT:    andl $252645135, %ecx # imm = 0xF0F0F0F
-; X86-NEXT:    shll $4, %ecx
-; X86-NEXT:    shrl $4, %eax
-; X86-NEXT:    andl $252645135, %eax # imm = 0xF0F0F0F
-; X86-NEXT:    orl %ecx, %eax
-; X86-NEXT:    movl %eax, %ecx
-; X86-NEXT:    andl $858993459, %ecx # imm = 0x33333333
-; X86-NEXT:    shrl $2, %eax
-; X86-NEXT:    andl $858993459, %eax # imm = 0x33333333
-; X86-NEXT:    leal (%eax,%ecx,4), %eax
-; X86-NEXT:    movl %eax, %ecx
-; X86-NEXT:    andl $1431655765, %ecx # imm = 0x55555555
-; X86-NEXT:    shrl %eax
-; X86-NEXT:    andl $1431655765, %eax # imm = 0x55555555
-; X86-NEXT:    leal (%eax,%ecx,2), %eax
-; X86-NEXT:    bswapl %eax
-; X86-NEXT:    movl %eax, %ecx
-; X86-NEXT:    andl $252645135, %ecx # imm = 0xF0F0F0F
-; X86-NEXT:    shll $4, %ecx
-; X86-NEXT:    shrl $4, %eax
-; X86-NEXT:    andl $252645135, %eax # imm = 0xF0F0F0F
-; X86-NEXT:    orl %ecx, %eax
-; X86-NEXT:    movl %eax, %ecx
-; X86-NEXT:    andl $858993459, %ecx # imm = 0x33333333
-; X86-NEXT:    shrl $2, %eax
-; X86-NEXT:    andl $858993459, %eax # imm = 0x33333333
-; X86-NEXT:    leal (%eax,%ecx,4), %eax
-; X86-NEXT:    movl %eax, %ecx
-; X86-NEXT:    andl $1431655765, %ecx # imm = 0x55555555
-; X86-NEXT:    shrl %eax
-; X86-NEXT:    andl $1431655765, %eax # imm = 0x55555555
-; X86-NEXT:    leal (%eax,%ecx,2), %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: freeze_bitreverse:
 ; X64:       # %bb.0:
-; X64-NEXT:    # kill: def $edi killed $edi def $rdi
-; X64-NEXT:    bswapl %edi
-; X64-NEXT:    movl %edi, %eax
-; X64-NEXT:    andl $252645135, %eax # imm = 0xF0F0F0F
-; X64-NEXT:    shll $4, %eax
-; X64-NEXT:    shrl $4, %edi
-; X64-NEXT:    andl $252645135, %edi # imm = 0xF0F0F0F
-; X64-NEXT:    orl %eax, %edi
 ; X64-NEXT:    movl %edi, %eax
-; X64-NEXT:    andl $858993459, %eax # imm = 0x33333333
-; X64-NEXT:    shrl $2, %edi
-; X64-NEXT:    andl $858993459, %edi # imm = 0x33333333
-; X64-NEXT:    leal (%rdi,%rax,4), %eax
-; X64-NEXT:    movl %eax, %ecx
-; X64-NEXT:    andl $1431655765, %ecx # imm = 0x55555555
-; X64-NEXT:    shrl %eax
-; X64-NEXT:    andl $1431655765, %eax # imm = 0x55555555
-; X64-NEXT:    leal (%rax,%rcx,2), %eax
-; X64-NEXT:    bswapl %eax
-; X64-NEXT:    movl %eax, %ecx
-; X64-NEXT:    andl $252645135, %ecx # imm = 0xF0F0F0F
-; X64-NEXT:    shll $4, %ecx
-; X64-NEXT:    shrl $4, %eax
-; X64-NEXT:    andl $252645135, %eax # imm = 0xF0F0F0F
-; X64-NEXT:    orl %ecx, %eax
-; X64-NEXT:    movl %eax, %ecx
-; X64-NEXT:    andl $858993459, %ecx # imm = 0x33333333
-; X64-NEXT:    shrl $2, %eax
-; X64-NEXT:    andl $858993459, %eax # imm = 0x33333333
-; X64-NEXT:    leal (%rax,%rcx,4), %eax
-; X64-NEXT:    movl %eax, %ecx
-; X64-NEXT:    andl $1431655765, %ecx # imm = 0x55555555
-; X64-NEXT:    shrl %eax
-; X64-NEXT:    andl $1431655765, %eax # imm = 0x55555555
-; X64-NEXT:    leal (%rax,%rcx,2), %eax
 ; X64-NEXT:    retq
   %x = call i32 @llvm.bitreverse.i32(i32 %a0)
   %y = freeze i32 %x
@@ -212,92 +116,9 @@ define i32 @freeze_bitreverse(i32 %a0) nounwind {
 declare i32 @llvm.bitreverse.i32(i32)
 
 define <4 x i32> @freeze_bitreverse_vec(<4 x i32> %a0) nounwind {
-; X86-LABEL: freeze_bitreverse_vec:
-; X86:       # %bb.0:
-; X86-NEXT:    pxor %xmm1, %xmm1
-; X86-NEXT:    movdqa %xmm0, %xmm2
-; X86-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15]
-; X86-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[3,2,1,0,4,5,6,7]
-; X86-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,6,5,4]
-; X86-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; X86-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,2,1,0,4,5,6,7]
-; X86-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,5,4]
-; X86-NEXT:    packuswb %xmm2, %xmm0
-; X86-NEXT:    movdqa %xmm0, %xmm3
-; X86-NEXT:    psrlw $4, %xmm3
-; X86-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X86-NEXT:    pand %xmm2, %xmm3
-; X86-NEXT:    pand %xmm2, %xmm0
-; X86-NEXT:    psllw $4, %xmm0
-; X86-NEXT:    por %xmm3, %xmm0
-; X86-NEXT:    movdqa %xmm0, %xmm4
-; X86-NEXT:    psrlw $2, %xmm4
-; X86-NEXT:    movdqa {{.*#+}} xmm3 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; X86-NEXT:    pand %xmm3, %xmm4
-; X86-NEXT:    pand %xmm3, %xmm0
-; X86-NEXT:    psllw $2, %xmm0
-; X86-NEXT:    por %xmm4, %xmm0
-; X86-NEXT:    movdqa %xmm0, %xmm5
-; X86-NEXT:    psrlw $1, %xmm5
-; X86-NEXT:    movdqa {{.*#+}} xmm4 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
-; X86-NEXT:    pand %xmm4, %xmm5
-; X86-NEXT:    pand %xmm4, %xmm0
-; X86-NEXT:    paddb %xmm0, %xmm0
-; X86-NEXT:    por %xmm5, %xmm0
-; X86-NEXT:    movdqa %xmm0, %xmm5
-; X86-NEXT:    punpckhbw {{.*#+}} xmm5 = xmm5[8],xmm1[8],xmm5[9],xmm1[9],xmm5[10],xmm1[10],xmm5[11],xmm1[11],xmm5[12],xmm1[12],xmm5[13],xmm1[13],xmm5[14],xmm1[14],xmm5[15],xmm1[15]
-; X86-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[3,2,1,0,4,5,6,7]
-; X86-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,7,6,5,4]
-; X86-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; X86-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,2,1,0,4,5,6,7]
-; X86-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,5,4]
-; X86-NEXT:    packuswb %xmm5, %xmm0
-; X86-NEXT:    movdqa %xmm0, %xmm1
-; X86-NEXT:    psrlw $4, %xmm1
-; X86-NEXT:    pand %xmm2, %xmm1
-; X86-NEXT:    pand %xmm2, %xmm0
-; X86-NEXT:    psllw $4, %xmm0
-; X86-NEXT:    por %xmm1, %xmm0
-; X86-NEXT:    movdqa %xmm0, %xmm1
-; X86-NEXT:    psrlw $2, %xmm1
-; X86-NEXT:    pand %xmm3, %xmm1
-; X86-NEXT:    pand %xmm3, %xmm0
-; X86-NEXT:    psllw $2, %xmm0
-; X86-NEXT:    por %xmm1, %xmm0
-; X86-NEXT:    movdqa %xmm0, %xmm1
-; X86-NEXT:    psrlw $1, %xmm1
-; X86-NEXT:    pand %xmm4, %xmm1
-; X86-NEXT:    pand %xmm4, %xmm0
-; X86-NEXT:    paddb %xmm0, %xmm0
-; X86-NEXT:    por %xmm1, %xmm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: freeze_bitreverse_vec:
-; X64:       # %bb.0:
-; X64-NEXT:    movdqa {{.*#+}} xmm2 = [3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12]
-; X64-NEXT:    pshufb %xmm2, %xmm0
-; X64-NEXT:    movdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X64-NEXT:    movdqa %xmm0, %xmm1
-; X64-NEXT:    pand %xmm3, %xmm1
-; X64-NEXT:    movdqa {{.*#+}} xmm4 = [0,128,64,192,32,160,96,224,16,144,80,208,48,176,112,240]
-; X64-NEXT:    movdqa %xmm4, %xmm5
-; X64-NEXT:    pshufb %xmm1, %xmm5
-; X64-NEXT:    psrlw $4, %xmm0
-; X64-NEXT:    pand %xmm3, %xmm0
-; X64-NEXT:    movdqa {{.*#+}} xmm1 = [0,8,4,12,2,10,6,14,1,9,5,13,3,11,7,15]
-; X64-NEXT:    movdqa %xmm1, %xmm6
-; X64-NEXT:    pshufb %xmm0, %xmm6
-; X64-NEXT:    por %xmm5, %xmm6
-; X64-NEXT:    pshufb %xmm2, %xmm6
-; X64-NEXT:    movdqa %xmm6, %xmm0
-; X64-NEXT:    pand %xmm3, %xmm0
-; X64-NEXT:    pshufb %xmm0, %xmm4
-; X64-NEXT:    psrlw $4, %xmm6
-; X64-NEXT:    pand %xmm3, %xmm6
-; X64-NEXT:    pshufb %xmm6, %xmm1
-; X64-NEXT:    por %xmm4, %xmm1
-; X64-NEXT:    movdqa %xmm1, %xmm0
-; X64-NEXT:    retq
+; CHECK-LABEL: freeze_bitreverse_vec:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    ret{{[l|q]}}
   %x = call <4 x i32> @llvm.bitreverse.v4i32(<4 x i32> %a0)
   %y = freeze <4 x i32> %x
   %z = call <4 x i32> @llvm.bitreverse.v4i32(<4 x i32> %y)
-- 
2.7.4