[RISCV] Use default promotion for i32 CTLZ_ZERO_UNDEF on RV64 with XTHeadBb.

[topperc]

If we don't need to worry about the input being zero, we can shift the input right by 32 and use TH_FF1.

[llvmbot]

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

Author: Craig Topper (topperc)

Changes

If we don't need to worry about the input being zero, we can shift the input right by 32 and use TH_FF1.

---

Full diff: https://github.com/llvm/llvm-project/pull/157955.diff

3 Files Affected:

• (modified) llvm/lib/Target/RISCV/RISCVISelLowering.cpp (+6-2)
• (modified) llvm/test/CodeGen/RISCV/ctlz-cttz-ctpop.ll (+1-2)
• (modified) llvm/test/CodeGen/RISCV/rv64xtheadbb.ll (+2-3)

diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index a68a3c14dc41d..ae9e2fef88673 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -421,8 +421,12 @@ RISCVTargetLowering::RISCVTargetLowering(const TargetMachine &TM,
       (Subtarget.hasVendorXCVbitmanip() && !Subtarget.is64Bit())) {
     // We need the custom lowering to make sure that the resulting sequence
     // for the 32bit case is efficient on 64bit targets.
-    if (Subtarget.is64Bit())
-      setOperationAction({ISD::CTLZ, ISD::CTLZ_ZERO_UNDEF}, MVT::i32, Custom);
+    if (Subtarget.is64Bit()) {
+      setOperationAction(ISD::CTLZ, MVT::i32, Custom);
+      // Use default promotion for XTHeadBb.
+      if (Subtarget.hasStdExtZbb())
+        setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i32, Custom);
+    }
   } else {
     setOperationAction(ISD::CTLZ, XLenVT, Expand);
   }
diff --git a/llvm/test/CodeGen/RISCV/ctlz-cttz-ctpop.ll b/llvm/test/CodeGen/RISCV/ctlz-cttz-ctpop.ll
index 908a12331d1bb..99ed0a355b909 100644
--- a/llvm/test/CodeGen/RISCV/ctlz-cttz-ctpop.ll
+++ b/llvm/test/CodeGen/RISCV/ctlz-cttz-ctpop.ll
@@ -2004,9 +2004,8 @@ define i32 @test_ctlz_i32_zero_undef(i32 %a) nounwind {
 ;
 ; RV64XTHEADBB-LABEL: test_ctlz_i32_zero_undef:
 ; RV64XTHEADBB:       # %bb.0:
-; RV64XTHEADBB-NEXT:    not a0, a0
 ; RV64XTHEADBB-NEXT:    slli a0, a0, 32
-; RV64XTHEADBB-NEXT:    th.ff0 a0, a0
+; RV64XTHEADBB-NEXT:    th.ff1 a0, a0
 ; RV64XTHEADBB-NEXT:    ret
   %tmp = call i32 @llvm.ctlz.i32(i32 %a, i1 true)
   ret i32 %tmp
diff --git a/llvm/test/CodeGen/RISCV/rv64xtheadbb.ll b/llvm/test/CodeGen/RISCV/rv64xtheadbb.ll
index d504d418c150b..f20ebca314b35 100644
--- a/llvm/test/CodeGen/RISCV/rv64xtheadbb.ll
+++ b/llvm/test/CodeGen/RISCV/rv64xtheadbb.ll
@@ -240,10 +240,9 @@ define signext i32 @findLastSet_i32(i32 signext %a) nounwind {
 ;
 ; RV64XTHEADBB-NOB-LABEL: findLastSet_i32:
 ; RV64XTHEADBB-NOB:       # %bb.0:
-; RV64XTHEADBB-NOB-NEXT:    not a1, a0
+; RV64XTHEADBB-NOB-NEXT:    slli a1, a0, 32
 ; RV64XTHEADBB-NOB-NEXT:    snez a0, a0
-; RV64XTHEADBB-NOB-NEXT:    slli a1, a1, 32
-; RV64XTHEADBB-NOB-NEXT:    th.ff0 a1, a1
+; RV64XTHEADBB-NOB-NEXT:    th.ff1 a1, a1
 ; RV64XTHEADBB-NOB-NEXT:    xori a1, a1, 31
 ; RV64XTHEADBB-NOB-NEXT:    addi a0, a0, -1
 ; RV64XTHEADBB-NOB-NEXT:    or a0, a0, a1

[preames]

LGTM

[ptomsich]

Looks like I am late to the show...
LGTM. Thanks for doing this!