[llvm] 0fec8ac - [SelectionDAG] ComputeNumSignBits - add ISD::ADD vector support

[Simon Pilgrim via llvm-commits]

Author: Simon Pilgrim
Date: 2020-01-23T16:42:12Z
New Revision: 0fec8acdd82a69fc5419b4a9db3c92a86634729d

URL: https://github.com/llvm/llvm-project/commit/0fec8acdd82a69fc5419b4a9db3c92a86634729d
DIFF: https://github.com/llvm/llvm-project/commit/0fec8acdd82a69fc5419b4a9db3c92a86634729d.diff

LOG: [SelectionDAG] ComputeNumSignBits - add ISD::ADD vector support

Add missing handling for (ADD (AND X, 1), -1) uniform vectors

Added: 
    

Modified: 
    llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
    llvm/test/CodeGen/X86/sar_fold64.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index d4fab6535fa4..461d480405aa 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -3760,13 +3760,13 @@ unsigned SelectionDAG::ComputeNumSignBits(SDValue Op, const APInt &DemandedElts,
   case ISD::ADDC:
     // Add can have at most one carry bit.  Thus we know that the output
     // is, at worst, one more bit than the inputs.
-    Tmp = ComputeNumSignBits(Op.getOperand(0), Depth+1);
-    if (Tmp == 1) return 1;  // Early out.
+    Tmp = ComputeNumSignBits(Op.getOperand(0), Depth + 1);
+    if (Tmp == 1) return 1; // Early out.
 
     // Special case decrementing a value (ADD X, -1):
-    if (ConstantSDNode *CRHS = dyn_cast<ConstantSDNode>(Op.getOperand(1)))
+    if (ConstantSDNode *CRHS = isConstOrConstSplat(Op.getOperand(1)))
       if (CRHS->isAllOnesValue()) {
-        KnownBits Known = computeKnownBits(Op.getOperand(0), Depth+1);
+        KnownBits Known = computeKnownBits(Op.getOperand(0), Depth + 1);
 
         // If the input is known to be 0 or 1, the output is 0/-1, which is all
         // sign bits set.
@@ -3779,10 +3779,9 @@ unsigned SelectionDAG::ComputeNumSignBits(SDValue Op, const APInt &DemandedElts,
           return Tmp;
       }
 
-    Tmp2 = ComputeNumSignBits(Op.getOperand(1), Depth+1);
-    if (Tmp2 == 1) return 1;
-    return std::min(Tmp, Tmp2)-1;
-
+    Tmp2 = ComputeNumSignBits(Op.getOperand(1), Depth + 1);
+    if (Tmp2 == 1) return 1; // Early out.
+    return std::min(Tmp, Tmp2) - 1;
   case ISD::SUB:
     Tmp2 = ComputeNumSignBits(Op.getOperand(1), DemandedElts, Depth + 1);
     if (Tmp2 == 1) return 1; // Early out.

diff  --git a/llvm/test/CodeGen/X86/sar_fold64.ll b/llvm/test/CodeGen/X86/sar_fold64.ll
index deed996482db..c53d43292bc7 100644
--- a/llvm/test/CodeGen/X86/sar_fold64.ll
+++ b/llvm/test/CodeGen/X86/sar_fold64.ll
@@ -135,16 +135,7 @@ define <4 x i32> @all_sign_bit_ashr_vec2(<4 x i32> %x) {
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    pand {{.*}}(%rip), %xmm0
 ; SSE-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE-NEXT:    paddd %xmm0, %xmm1
-; SSE-NEXT:    movdqa %xmm1, %xmm0
-; SSE-NEXT:    psrad $5, %xmm0
-; SSE-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
-; SSE-NEXT:    movdqa %xmm1, %xmm2
-; SSE-NEXT:    psrad $31, %xmm2
-; SSE-NEXT:    psrad $1, %xmm1
-; SSE-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
-; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,3]
-; SSE-NEXT:    movaps %xmm1, %xmm0
+; SSE-NEXT:    paddd %xmm1, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: all_sign_bit_ashr_vec2:
@@ -152,12 +143,6 @@ define <4 x i32> @all_sign_bit_ashr_vec2(<4 x i32> %x) {
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrad $31, %xmm0, %xmm1
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5,6,7]
-; AVX1-NEXT:    vpsrad $5, %xmm0, %xmm2
-; AVX1-NEXT:    vpsrad $1, %xmm0, %xmm0
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7]
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: all_sign_bit_ashr_vec2:
@@ -166,7 +151,6 @@ define <4 x i32> @all_sign_bit_ashr_vec2(<4 x i32> %x) {
 ; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsravd {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    retq
   %and = and <4 x i32> %x, <i32 1, i32 1, i32 1 , i32 1>
   %add = add <4 x i32> %and, <i32 -1, i32 -1, i32 -1, i32 -1>