[GlobalISel] Add CTLZ known bits.

[davemgreen]

Copied from SDAG.

[llvmbot]

@llvm/pr-subscribers-llvm-globalisel
@llvm/pr-subscribers-backend-aarch64

@llvm/pr-subscribers-backend-amdgpu

Author: David Green (davemgreen)

Changes

Copied from SDAG.

---

Full diff: https://github.com/llvm/llvm-project/pull/86436.diff

3 Files Affected:

• (modified) llvm/lib/CodeGen/GlobalISel/GISelKnownBits.cpp (+11)
• (modified) llvm/test/CodeGen/AArch64/setcc_knownbits.ll (+1-3)
• (modified) llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-ctlz-zero-undef.mir (+1-3)

diff --git a/llvm/lib/CodeGen/GlobalISel/GISelKnownBits.cpp b/llvm/lib/CodeGen/GlobalISel/GISelKnownBits.cpp
index 2e2cc9a95bd95c..914022ab7301b1 100644
--- a/llvm/lib/CodeGen/GlobalISel/GISelKnownBits.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/GISelKnownBits.cpp
@@ -589,6 +589,17 @@ void GISelKnownBits::computeKnownBitsImpl(Register R, KnownBits &Known,
     }
     break;
   }
+  case TargetOpcode::G_CTLZ:
+  case TargetOpcode::G_CTLZ_ZERO_UNDEF: {
+    KnownBits SrcOpKnown;
+    computeKnownBitsImpl(MI.getOperand(1).getReg(), SrcOpKnown, DemandedElts,
+                         Depth + 1);
+    // If we have a known 1, its position is our upper bound.
+    unsigned PossibleLZ = SrcOpKnown.countMaxLeadingZeros();
+    unsigned LowBits = llvm::bit_width(PossibleLZ);
+    Known.Zero.setBitsFrom(LowBits);
+    break;
+  }
   }
 
   assert(!Known.hasConflict() && "Bits known to be one AND zero?");
diff --git a/llvm/test/CodeGen/AArch64/setcc_knownbits.ll b/llvm/test/CodeGen/AArch64/setcc_knownbits.ll
index 6a337bf02e09b9..8be63b04d8ce58 100644
--- a/llvm/test/CodeGen/AArch64/setcc_knownbits.ll
+++ b/llvm/test/CodeGen/AArch64/setcc_knownbits.ll
@@ -68,9 +68,7 @@ define i1 @lshr_ctlz_undef_cmpeq_one_i64(i64 %in) {
 ; CHECK-GI-LABEL: lshr_ctlz_undef_cmpeq_one_i64:
 ; CHECK-GI:       // %bb.0:
 ; CHECK-GI-NEXT:    clz x8, x0
-; CHECK-GI-NEXT:    lsr x8, x8, #6
-; CHECK-GI-NEXT:    cmp x8, #1
-; CHECK-GI-NEXT:    cset w0, eq
+; CHECK-GI-NEXT:    lsr w0, w8, #6
 ; CHECK-GI-NEXT:    ret
   %ctlz = call i64 @llvm.ctlz.i64(i64 %in, i1 -1)
   %lshr = lshr i64 %ctlz, 6
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-ctlz-zero-undef.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-ctlz-zero-undef.mir
index e9f8180aae66bb..fed277d7d10d08 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-ctlz-zero-undef.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-ctlz-zero-undef.mir
@@ -64,9 +64,7 @@ body: |
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[CTLZ_ZERO_UNDEF:%[0-9]+]]:_(s32) = G_CTLZ_ZERO_UNDEF [[COPY]](s32)
-    ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
-    ; CHECK-NEXT: [[AND:%[0-9]+]]:_(s32) = G_AND [[CTLZ_ZERO_UNDEF]], [[C]]
-    ; CHECK-NEXT: $vgpr0 = COPY [[AND]](s32)
+    ; CHECK-NEXT: $vgpr0 = COPY [[CTLZ_ZERO_UNDEF]](s32)
     %0:_(s32) = COPY $vgpr0
     %1:_(s16) = G_CTLZ_ZERO_UNDEF %0
     %2:_(s32) = G_ZEXT %1

[github-actions]

✅ With the latest revision this PR passed the Python code formatter.

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.