[SelectionDAG] Add MaskedValueIsZero check to allow folding of zero extended variables we know are safe to extend (#85573)

Add ones for every high bit that will cleared.

This will allow us to evaluate variables that have their bits known to
see if they have no risk of overflow despite the shift amount being
greater than the difference between the two types.

Author: AZero13

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

@@ -13832,11 +13832,20 @@ SDValue DAGCombiner::visitZERO_EXTEND(SDNode *N) {
         if (N0.getOpcode() == ISD::SHL) {
           // If the original shl may be shifting out bits, do not perform this
           // transformation.
-          // TODO: Add MaskedValueIsZero check.
           unsigned KnownZeroBits = ShVal.getValueSizeInBits() -
                                    ShVal.getOperand(0).getValueSizeInBits();
-          if (ShAmtC->getAPIntValue().ugt(KnownZeroBits))
-            return SDValue();
+          if (ShAmtC->getAPIntValue().ugt(KnownZeroBits)) {
+            // If the shift is too large, then see if we can deduce that the
+            // shift is safe anyway.
+            // Create a mask that has ones for the bits being shifted out.
+            APInt ShiftOutMask =
+                APInt::getHighBitsSet(ShVal.getValueSizeInBits(),
+                                      ShAmtC->getAPIntValue().getZExtValue());
+
+            // Check if the bits being shifted out are known to be zero.
+            if (!DAG.MaskedValueIsZero(ShVal, ShiftOutMask))
+              return SDValue();
+          }
         }
 
         // Ensure that the shift amount is wide enough for the shifted value.

llvm/test/CodeGen/X86/dagcombine-shifts.ll

@@ -322,5 +322,132 @@ define void @g(i32 %a) nounwind {
   ret void
 }
 
+define i32 @shift_zext_shl(i8 zeroext %x) {
+; X86-LABEL: shift_zext_shl:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    andl $64, %eax
+; X86-NEXT:    shll $9, %eax
+; X86-NEXT:    retl
+;
+; X64-LABEL: shift_zext_shl:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    andl $64, %eax
+; X64-NEXT:    shll $9, %eax
+; X64-NEXT:    retq
+  %a = and i8 %x, 64
+  %b = zext i8 %a to i16
+  %c = shl i16 %b, 9
+  %d = zext i16 %c to i32
+  ret i32 %d
+}
+
+define i32 @shift_zext_shl2(i8 zeroext %x) {
+; X86-LABEL: shift_zext_shl2:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    andl $64, %eax
+; X86-NEXT:    shll $9, %eax
+; X86-NEXT:    retl
+;
+; X64-LABEL: shift_zext_shl2:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    andl $64, %eax
+; X64-NEXT:    shll $9, %eax
+; X64-NEXT:    retq
+  %a = and i8 %x, 64
+  %b = zext i8 %a to i32
+  %c = shl i32 %b, 9
+  ret i32 %c
+}
+
+define <4 x i32> @shift_zext_shl_vec(<4 x i8> %x) nounwind {
+; X86-LABEL: shift_zext_shl_vec:
+; X86:       # %bb.0:
+; X86-NEXT:    pushl %edi
+; X86-NEXT:    pushl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %edi
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %esi
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    andl $64, %ecx
+; X86-NEXT:    shll $9, %ecx
+; X86-NEXT:    andl $63, %edx
+; X86-NEXT:    shll $8, %edx
+; X86-NEXT:    andl $31, %esi
+; X86-NEXT:    shll $7, %esi
+; X86-NEXT:    andl $23, %edi
+; X86-NEXT:    shll $6, %edi
+; X86-NEXT:    movl %edi, 12(%eax)
+; X86-NEXT:    movl %esi, 8(%eax)
+; X86-NEXT:    movl %edx, 4(%eax)
+; X86-NEXT:    movl %ecx, (%eax)
+; X86-NEXT:    popl %esi
+; X86-NEXT:    popl %edi
+; X86-NEXT:    retl $4
+;
+; X64-LABEL: shift_zext_shl_vec:
+; X64:       # %bb.0:
+; X64-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; X64-NEXT:    pxor %xmm1, %xmm1
+; X64-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]
+; X64-NEXT:    pmullw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; X64-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; X64-NEXT:    retq
+  %a = and <4 x i8> %x, <i8 64, i8 63, i8 31, i8 23>
+  %b = zext <4 x i8> %a to <4 x i16>
+  %c = shl <4 x i16> %b, <i16 9, i16 8, i16 7, i16 6>
+  %d = zext <4 x i16> %c to <4 x i32>
+  ret <4 x i32> %d
+}
+
+define <4 x i32> @shift_zext_shl2_vec(<4 x i8> %x) nounwind {
+; X86-LABEL: shift_zext_shl2_vec:
+; X86:       # %bb.0:
+; X86-NEXT:    pushl %edi
+; X86-NEXT:    pushl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %esi
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %edi
+; X86-NEXT:    andl $23, %edi
+; X86-NEXT:    andl $31, %esi
+; X86-NEXT:    andl $63, %edx
+; X86-NEXT:    andl $64, %ecx
+; X86-NEXT:    shll $9, %ecx
+; X86-NEXT:    shll $8, %edx
+; X86-NEXT:    shll $7, %esi
+; X86-NEXT:    shll $6, %edi
+; X86-NEXT:    movl %edi, 12(%eax)
+; X86-NEXT:    movl %esi, 8(%eax)
+; X86-NEXT:    movl %edx, 4(%eax)
+; X86-NEXT:    movl %ecx, (%eax)
+; X86-NEXT:    popl %esi
+; X86-NEXT:    popl %edi
+; X86-NEXT:    retl $4
+;
+; X64-LABEL: shift_zext_shl2_vec:
+; X64:       # %bb.0:
+; X64-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; X64-NEXT:    pxor %xmm1, %xmm1
+; X64-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]
+; X64-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; X64-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; X64-NEXT:    pmuludq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; X64-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; X64-NEXT:    pmuludq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
+; X64-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; X64-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; X64-NEXT:    retq
+  %a = and <4 x i8> %x, <i8 64, i8 63, i8 31, i8 23>
+  %b = zext <4 x i8> %a to <4 x i32>
+  %c = shl <4 x i32> %b, <i32 9, i32 8, i32 7, i32 6>
+  ret <4 x i32> %c
+}
+
 declare dso_local void @f(i64)