[RISCV][GISel] Use maskedValueIsZero in RISCVInstructionSelector::selectZExtBits.

[topperc]

No description provided.

[llvmbot]

@llvm/pr-subscribers-llvm-globalisel

@llvm/pr-subscribers-backend-risc-v

Author: Craig Topper (topperc)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/115244.diff

3 Files Affected:

• (modified) llvm/lib/Target/RISCV/GISel/RISCVInstructionSelector.cpp (+3-1)
• (modified) llvm/test/CodeGen/RISCV/GlobalISel/rv32zbkb.ll (+61-33)
• (modified) llvm/test/CodeGen/RISCV/GlobalISel/rv64zbkb.ll (+1-2)

diff --git a/llvm/lib/Target/RISCV/GISel/RISCVInstructionSelector.cpp b/llvm/lib/Target/RISCV/GISel/RISCVInstructionSelector.cpp
index bc718b454ffb69..eacb7800261af8 100644
--- a/llvm/lib/Target/RISCV/GISel/RISCVInstructionSelector.cpp
+++ b/llvm/lib/Target/RISCV/GISel/RISCVInstructionSelector.cpp
@@ -261,7 +261,9 @@ RISCVInstructionSelector::selectZExtBits(MachineOperand &Root,
     return {{[=](MachineInstrBuilder &MIB) { MIB.addReg(RegX); }}};
   }
 
-  // TODO: Use computeKnownBits.
+  unsigned Size = MRI->getType(RootReg).getSizeInBits();
+  if (KB->maskedValueIsZero(RootReg, APInt::getBitsSetFrom(Size, Bits)))
+    return {{[=](MachineInstrBuilder &MIB) { MIB.add(Root); }}};
 
   return std::nullopt;
 }
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/rv32zbkb.ll b/llvm/test/CodeGen/RISCV/GlobalISel/rv32zbkb.ll
index da9bffcdb0d582..208d8e441a493d 100644
--- a/llvm/test/CodeGen/RISCV/GlobalISel/rv32zbkb.ll
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/rv32zbkb.ll
@@ -25,11 +25,16 @@ define i32 @pack_i32(i32 %a, i32 %b) nounwind {
 }
 
 define i32 @pack_i32_2(i16 zeroext %a, i16 zeroext %b) nounwind {
-; CHECK-LABEL: pack_i32_2:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    slli a1, a1, 16
-; CHECK-NEXT:    or a0, a1, a0
-; CHECK-NEXT:    ret
+; RV32I-LABEL: pack_i32_2:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    slli a1, a1, 16
+; RV32I-NEXT:    or a0, a1, a0
+; RV32I-NEXT:    ret
+;
+; RV32ZBKB-LABEL: pack_i32_2:
+; RV32ZBKB:       # %bb.0:
+; RV32ZBKB-NEXT:    pack a0, a0, a1
+; RV32ZBKB-NEXT:    ret
   %zexta = zext i16 %a to i32
   %zextb = zext i16 %b to i32
   %shl1 = shl i32 %zextb, 16
@@ -38,12 +43,18 @@ define i32 @pack_i32_2(i16 zeroext %a, i16 zeroext %b) nounwind {
 }
 
 define i32 @pack_i32_3(i16 zeroext %0, i16 zeroext %1, i32 %2) {
-; CHECK-LABEL: pack_i32_3:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    slli a0, a0, 16
-; CHECK-NEXT:    or a0, a0, a1
-; CHECK-NEXT:    add a0, a0, a2
-; CHECK-NEXT:    ret
+; RV32I-LABEL: pack_i32_3:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    slli a0, a0, 16
+; RV32I-NEXT:    or a0, a0, a1
+; RV32I-NEXT:    add a0, a0, a2
+; RV32I-NEXT:    ret
+;
+; RV32ZBKB-LABEL: pack_i32_3:
+; RV32ZBKB:       # %bb.0:
+; RV32ZBKB-NEXT:    pack a0, a1, a0
+; RV32ZBKB-NEXT:    add a0, a0, a2
+; RV32ZBKB-NEXT:    ret
   %4 = zext i16 %0 to i32
   %5 = shl nuw i32 %4, 16
   %6 = zext i16 %1 to i32
@@ -165,9 +176,8 @@ define i64 @packh_i64_2(i64 %a, i64 %b) nounwind {
 ; RV32ZBKB-LABEL: packh_i64_2:
 ; RV32ZBKB:       # %bb.0:
 ; RV32ZBKB-NEXT:    andi a1, a2, 255
-; RV32ZBKB-NEXT:    slli a3, zero, 8
 ; RV32ZBKB-NEXT:    srli a1, a1, 24
-; RV32ZBKB-NEXT:    or a1, a3, a1
+; RV32ZBKB-NEXT:    packh a1, a1, zero
 ; RV32ZBKB-NEXT:    packh a0, a0, a2
 ; RV32ZBKB-NEXT:    ret
   %and = and i64 %a, 255
@@ -179,11 +189,16 @@ define i64 @packh_i64_2(i64 %a, i64 %b) nounwind {
 
 
 define zeroext i16 @packh_i16(i8 zeroext %a, i8 zeroext %b) nounwind {
-; CHECK-LABEL: packh_i16:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    slli a1, a1, 8
-; CHECK-NEXT:    or a0, a1, a0
-; CHECK-NEXT:    ret
+; RV32I-LABEL: packh_i16:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    slli a1, a1, 8
+; RV32I-NEXT:    or a0, a1, a0
+; RV32I-NEXT:    ret
+;
+; RV32ZBKB-LABEL: packh_i16:
+; RV32ZBKB:       # %bb.0:
+; RV32ZBKB-NEXT:    packh a0, a0, a1
+; RV32ZBKB-NEXT:    ret
   %zext = zext i8 %a to i16
   %zext1 = zext i8 %b to i16
   %shl = shl i16 %zext1, 8
@@ -193,13 +208,19 @@ define zeroext i16 @packh_i16(i8 zeroext %a, i8 zeroext %b) nounwind {
 
 
 define zeroext i16 @packh_i16_2(i8 zeroext %0, i8 zeroext %1, i8 zeroext %2) {
-; CHECK-LABEL: packh_i16_2:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    add a0, a1, a0
-; CHECK-NEXT:    andi a0, a0, 255
-; CHECK-NEXT:    slli a0, a0, 8
-; CHECK-NEXT:    or a0, a0, a2
-; CHECK-NEXT:    ret
+; RV32I-LABEL: packh_i16_2:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    add a0, a1, a0
+; RV32I-NEXT:    andi a0, a0, 255
+; RV32I-NEXT:    slli a0, a0, 8
+; RV32I-NEXT:    or a0, a0, a2
+; RV32I-NEXT:    ret
+;
+; RV32ZBKB-LABEL: packh_i16_2:
+; RV32ZBKB:       # %bb.0:
+; RV32ZBKB-NEXT:    add a0, a1, a0
+; RV32ZBKB-NEXT:    packh a0, a2, a0
+; RV32ZBKB-NEXT:    ret
   %4 = add i8 %1, %0
   %5 = zext i8 %4 to i16
   %6 = shl i16 %5, 8
@@ -209,14 +230,21 @@ define zeroext i16 @packh_i16_2(i8 zeroext %0, i8 zeroext %1, i8 zeroext %2) {
 }
 
 define void @packh_i16_3(i8 zeroext %0, i8 zeroext %1, i8 zeroext %2, ptr %p) {
-; CHECK-LABEL: packh_i16_3:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    add a0, a1, a0
-; CHECK-NEXT:    andi a0, a0, 255
-; CHECK-NEXT:    slli a0, a0, 8
-; CHECK-NEXT:    or a0, a0, a2
-; CHECK-NEXT:    sh a0, 0(a3)
-; CHECK-NEXT:    ret
+; RV32I-LABEL: packh_i16_3:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    add a0, a1, a0
+; RV32I-NEXT:    andi a0, a0, 255
+; RV32I-NEXT:    slli a0, a0, 8
+; RV32I-NEXT:    or a0, a0, a2
+; RV32I-NEXT:    sh a0, 0(a3)
+; RV32I-NEXT:    ret
+;
+; RV32ZBKB-LABEL: packh_i16_3:
+; RV32ZBKB:       # %bb.0:
+; RV32ZBKB-NEXT:    add a0, a1, a0
+; RV32ZBKB-NEXT:    packh a0, a2, a0
+; RV32ZBKB-NEXT:    sh a0, 0(a3)
+; RV32ZBKB-NEXT:    ret
   %4 = add i8 %1, %0
   %5 = zext i8 %4 to i16
   %6 = shl i16 %5, 8
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/rv64zbkb.ll b/llvm/test/CodeGen/RISCV/GlobalISel/rv64zbkb.ll
index abeefb3b1c97fd..6b57b179240d70 100644
--- a/llvm/test/CodeGen/RISCV/GlobalISel/rv64zbkb.ll
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/rv64zbkb.ll
@@ -130,8 +130,7 @@ define i64 @pack_i64_3(ptr %0, ptr %1) {
 ; RV64ZBKB:       # %bb.0:
 ; RV64ZBKB-NEXT:    lwu a0, 0(a0)
 ; RV64ZBKB-NEXT:    lwu a1, 0(a1)
-; RV64ZBKB-NEXT:    slli a0, a0, 32
-; RV64ZBKB-NEXT:    or a0, a0, a1
+; RV64ZBKB-NEXT:    pack a0, a1, a0
 ; RV64ZBKB-NEXT:    ret
   %3 = load i32, ptr %0, align 4
   %4 = zext i32 %3 to i64

[arsenm]

scalar size?

[topperc]

It should always be a scalar I think. Unless this gets called before any of the type checks in the pattern. But I made the change.

[arsenm]

I guess it would assert in practice in the APInt constructor