[llvm] 5990199 - [RISCV] Add CZERO_EQZ/CZERO_NEZ to ComputeNumSignBitsForTargetNode.

[Craig Topper via llvm-commits]

Author: Craig Topper
Date: 2023-07-24T07:43:02-07:00
New Revision: 5990199e2c1c1d99367437ddaef0bbadfa11aa85

URL: https://github.com/llvm/llvm-project/commit/5990199e2c1c1d99367437ddaef0bbadfa11aa85
DIFF: https://github.com/llvm/llvm-project/commit/5990199e2c1c1d99367437ddaef0bbadfa11aa85.diff

LOG: [RISCV] Add CZERO_EQZ/CZERO_NEZ to ComputeNumSignBitsForTargetNode.

Reviewed By: wangpc

Differential Revision: https://reviews.llvm.org/D156082

Added: 
    

Modified: 
    llvm/lib/Target/RISCV/RISCVISelLowering.cpp
    llvm/test/CodeGen/RISCV/condops.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 898f65bfe74f01..def0eac9ded180 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -13821,6 +13821,11 @@ unsigned RISCVTargetLowering::ComputeNumSignBitsForTargetNode(
         DAG.ComputeNumSignBits(Op.getOperand(4), DemandedElts, Depth + 1);
     return std::min(Tmp, Tmp2);
   }
+  case RISCVISD::CZERO_EQZ:
+  case RISCVISD::CZERO_NEZ:
+    // Output is either all zero or operand 0. We can propagate sign bit count
+    // from operand 0.
+    return DAG.ComputeNumSignBits(Op.getOperand(0), DemandedElts, Depth + 1);
   case RISCVISD::ABSW: {
     // We expand this at isel to negw+max. The result will have 33 sign bits
     // if the input has at least 33 sign bits.

diff  --git a/llvm/test/CodeGen/RISCV/condops.ll b/llvm/test/CodeGen/RISCV/condops.ll
index 210e1fd0734e8c..19018a3e461b46 100644
--- a/llvm/test/CodeGen/RISCV/condops.ll
+++ b/llvm/test/CodeGen/RISCV/condops.ll
@@ -3467,15 +3467,14 @@ define signext i16 @numsignbits(i16 signext %0, i16 signext %1, i16 signext %2,
 ; RV64XVENTANACONDOPS-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64XVENTANACONDOPS-NEXT:    sd s0, 0(sp) # 8-byte Folded Spill
 ; RV64XVENTANACONDOPS-NEXT:    vt.maskcn a3, a3, a0
-; RV64XVENTANACONDOPS-NEXT:    vt.maskc a0, a2, a0
-; RV64XVENTANACONDOPS-NEXT:    or a0, a0, a3
-; RV64XVENTANACONDOPS-NEXT:    slli s0, a0, 48
+; RV64XVENTANACONDOPS-NEXT:    vt.maskc s0, a2, a0
+; RV64XVENTANACONDOPS-NEXT:    or s0, s0, a3
 ; RV64XVENTANACONDOPS-NEXT:    beqz a1, .LBB58_2
 ; RV64XVENTANACONDOPS-NEXT:  # %bb.1:
-; RV64XVENTANACONDOPS-NEXT:    srai a0, s0, 48
+; RV64XVENTANACONDOPS-NEXT:    mv a0, s0
 ; RV64XVENTANACONDOPS-NEXT:    call bat at plt
 ; RV64XVENTANACONDOPS-NEXT:  .LBB58_2:
-; RV64XVENTANACONDOPS-NEXT:    srai a0, s0, 48
+; RV64XVENTANACONDOPS-NEXT:    mv a0, s0
 ; RV64XVENTANACONDOPS-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64XVENTANACONDOPS-NEXT:    ld s0, 0(sp) # 8-byte Folded Reload
 ; RV64XVENTANACONDOPS-NEXT:    addi sp, sp, 16
@@ -3505,15 +3504,14 @@ define signext i16 @numsignbits(i16 signext %0, i16 signext %1, i16 signext %2,
 ; RV32ZICOND-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32ZICOND-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
 ; RV32ZICOND-NEXT:    czero.nez a3, a3, a0
-; RV32ZICOND-NEXT:    czero.eqz a0, a2, a0
-; RV32ZICOND-NEXT:    or a0, a0, a3
-; RV32ZICOND-NEXT:    slli s0, a0, 16
+; RV32ZICOND-NEXT:    czero.eqz s0, a2, a0
+; RV32ZICOND-NEXT:    or s0, s0, a3
 ; RV32ZICOND-NEXT:    beqz a1, .LBB58_2
 ; RV32ZICOND-NEXT:  # %bb.1:
-; RV32ZICOND-NEXT:    srai a0, s0, 16
+; RV32ZICOND-NEXT:    mv a0, s0
 ; RV32ZICOND-NEXT:    call bat at plt
 ; RV32ZICOND-NEXT:  .LBB58_2:
-; RV32ZICOND-NEXT:    srai a0, s0, 16
+; RV32ZICOND-NEXT:    mv a0, s0
 ; RV32ZICOND-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32ZICOND-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
 ; RV32ZICOND-NEXT:    addi sp, sp, 16
@@ -3525,15 +3523,14 @@ define signext i16 @numsignbits(i16 signext %0, i16 signext %1, i16 signext %2,
 ; RV64ZICOND-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64ZICOND-NEXT:    sd s0, 0(sp) # 8-byte Folded Spill
 ; RV64ZICOND-NEXT:    czero.nez a3, a3, a0
-; RV64ZICOND-NEXT:    czero.eqz a0, a2, a0
-; RV64ZICOND-NEXT:    or a0, a0, a3
-; RV64ZICOND-NEXT:    slli s0, a0, 48
+; RV64ZICOND-NEXT:    czero.eqz s0, a2, a0
+; RV64ZICOND-NEXT:    or s0, s0, a3
 ; RV64ZICOND-NEXT:    beqz a1, .LBB58_2
 ; RV64ZICOND-NEXT:  # %bb.1:
-; RV64ZICOND-NEXT:    srai a0, s0, 48
+; RV64ZICOND-NEXT:    mv a0, s0
 ; RV64ZICOND-NEXT:    call bat at plt
 ; RV64ZICOND-NEXT:  .LBB58_2:
-; RV64ZICOND-NEXT:    srai a0, s0, 48
+; RV64ZICOND-NEXT:    mv a0, s0
 ; RV64ZICOND-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64ZICOND-NEXT:    ld s0, 0(sp) # 8-byte Folded Reload
 ; RV64ZICOND-NEXT:    addi sp, sp, 16