[llvm] [RISCV] Fix Lsb > Msb case in (sra (sext_inreg X, _), C) for th.ext (PR #136287)

via llvm-commits llvm-commits at lists.llvm.org
Fri Apr 18 02:13:11 PDT 2025


llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

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

Author: Jim Lin (tclin914)

<details>
<summary>Changes</summary>

According the [spec](https://github.com/XUANTIE-RV/thead-extension-spec/releases/tag/2.3.0), the operation of `th.ext rd, rs1, msb, lsb` is

  reg[rd] := sign_extend(reg[rs1][msb:lsb])

The spec doesn't specify if lsb is greater than msb.

I don't think lsb can be greater than msb. So that If the shift-right amount is greater than msb, we can set lsb equal to msb to extract the bit rs1[msb] and sign-extend it.

---
Full diff: https://github.com/llvm/llvm-project/pull/136287.diff


2 Files Affected:

- (modified) llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp (+3-1) 
- (modified) llvm/test/CodeGen/RISCV/rv32xtheadbb.ll (+32) 


``````````diff
diff --git a/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp b/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp
index 2d07a66ff275e..37b457e8a9b1e 100644
--- a/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp
@@ -653,7 +653,9 @@ bool RISCVDAGToDAGISel::trySignedBitfieldExtract(SDNode *Node) {
       return false;
 
     const unsigned Msb = ExtSize - 1;
-    const unsigned Lsb = RightShAmt;
+    // If the shift-right amount is greater than Msb, it means that extracts
+    // the X[Msb] bit and sign-extend it.
+    const unsigned Lsb = RightShAmt > Msb ? Msb : RightShAmt;
 
     SDNode *TH_EXT = BitfieldExtract(N0, Msb, Lsb, DL, VT);
     ReplaceNode(Node, TH_EXT);
diff --git a/llvm/test/CodeGen/RISCV/rv32xtheadbb.ll b/llvm/test/CodeGen/RISCV/rv32xtheadbb.ll
index 4bb8d6c248caa..0ced0c746a4cf 100644
--- a/llvm/test/CodeGen/RISCV/rv32xtheadbb.ll
+++ b/llvm/test/CodeGen/RISCV/rv32xtheadbb.ll
@@ -401,6 +401,38 @@ define i64 @no_sexth_i64(i64 %a) nounwind {
   ret i64 %shr
 }
 
+define i32 @sext_sext_inreg_sra(i16 %a) nounwind {
+; RV32I-LABEL: sext_sext_inreg_sra:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    slli a0, a0, 16
+; RV32I-NEXT:    srai a0, a0, 26
+; RV32I-NEXT:    ret
+;
+; RV32XTHEADBB-LABEL: sext_sext_inreg_sra:
+; RV32XTHEADBB:       # %bb.0:
+; RV32XTHEADBB-NEXT:    th.ext a0, a0, 15, 10
+; RV32XTHEADBB-NEXT:    ret
+  %sext = sext i16 %a to i32
+  %shr = ashr exact i32 %sext, 10
+  ret i32 %shr
+}
+
+define i32 @sext_sext_inreg_sra_2(i16 %a) nounwind {
+; RV32I-LABEL: sext_sext_inreg_sra_2:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    slli a0, a0, 16
+; RV32I-NEXT:    srai a0, a0, 31
+; RV32I-NEXT:    ret
+;
+; RV32XTHEADBB-LABEL: sext_sext_inreg_sra_2:
+; RV32XTHEADBB:       # %bb.0:
+; RV32XTHEADBB-NEXT:    th.ext a0, a0, 15, 15
+; RV32XTHEADBB-NEXT:    ret
+  %sext = sext i16 %a to i32
+  %shr = ashr exact i32 %sext, 24
+  ret i32 %shr
+}
+
 define i32 @zexth_i32(i32 %a) nounwind {
 ; RV32I-LABEL: zexth_i32:
 ; RV32I:       # %bb.0:

``````````

</details>


https://github.com/llvm/llvm-project/pull/136287


More information about the llvm-commits mailing list