[RISCV] Improve detection of when to skip (and (srl x, c2) c1) -> (srli (slli x,...

We have a special case to skip this transform if c1 is 0xffffffff
and x is sext_inreg in order to use sraiw+zext.w. But we were only
checking that we have a sext_inreg opcode, not how many bits are
being sign extended.

This commit adds a check that it is a sext_inreg from i32 so we know for
sure that an sraiw can be created.

diff --git a/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp b/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp
index a15a6dd..875bad0 100644 (file)
--- a/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp
@@ -774,7 +774,8 @@ void RISCVDAGToDAGISel::Select(SDNode *Node) {
         // (srli (slli x, c3-c2), c3).
         // Skip it in order to select sraiw.
         bool Skip = Subtarget->hasStdExtZba() && C3 == 32 &&
-                    X.getOpcode() == ISD::SIGN_EXTEND_INREG;
+                    X.getOpcode() == ISD::SIGN_EXTEND_INREG &&
+                    cast<VTSDNode>(X.getOperand(1))->getVT() == MVT::i32;
         if (OneUseOrZExtW && !IsANDI && !Skip) {
           SDNode *SLLI = CurDAG->getMachineNode(
               RISCV::SLLI, DL, XLenVT, X,

diff --git a/llvm/test/CodeGen/RISCV/rv64zba.ll b/llvm/test/CodeGen/RISCV/rv64zba.ll
index 2b9c1e2..b0b6c72 100644 (file)
--- a/llvm/test/CodeGen/RISCV/rv64zba.ll
+++ b/llvm/test/CodeGen/RISCV/rv64zba.ll
@@ -2,7 +2,9 @@
 ; RUN: llc -mtriple=riscv64 -mattr=+m -verify-machineinstrs < %s \
 ; RUN:   | FileCheck %s -check-prefix=RV64I
 ; RUN: llc -mtriple=riscv64 -mattr=+m,+zba -verify-machineinstrs < %s \
-; RUN:   | FileCheck %s -check-prefix=RV64ZBA
+; RUN:   | FileCheck %s -check-prefixes=RV64ZBA,RV64ZBANOZBB
+; RUN: llc -mtriple=riscv64 -mattr=+m,+zba,+zbb -verify-machineinstrs < %s \
+; RUN:   | FileCheck %s -check-prefixes=RV64ZBA,RV64ZBAZBB
 
 define i64 @slliuw(i64 %a) nounwind {
 ; RV64I-LABEL: slliuw:
@@ -1155,3 +1157,63 @@ define i64 @addshl64_5_8(i64 %a, i64 %b) {
   %e = add i64 %c, %d
   ret i64 %e
 }
+
+; Make sure we use sext.h+slli+srli for Zba+Zbb.
+; FIXME: The RV64I and Zba only cases can be done with only 3 shifts.
+define zeroext i32 @sext_ashr_zext_i8(i8 %a) nounwind {
+; RV64I-LABEL: sext_ashr_zext_i8:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    slli a0, a0, 56
+; RV64I-NEXT:    srai a0, a0, 56
+; RV64I-NEXT:    slli a0, a0, 23
+; RV64I-NEXT:    srli a0, a0, 32
+; RV64I-NEXT:    ret
+;
+; RV64ZBANOZBB-LABEL: sext_ashr_zext_i8:
+; RV64ZBANOZBB:       # %bb.0:
+; RV64ZBANOZBB-NEXT:    slli a0, a0, 56
+; RV64ZBANOZBB-NEXT:    srai a0, a0, 56
+; RV64ZBANOZBB-NEXT:    slli a0, a0, 23
+; RV64ZBANOZBB-NEXT:    srli a0, a0, 32
+; RV64ZBANOZBB-NEXT:    ret
+;
+; RV64ZBAZBB-LABEL: sext_ashr_zext_i8:
+; RV64ZBAZBB:       # %bb.0:
+; RV64ZBAZBB-NEXT:    sext.b a0, a0
+; RV64ZBAZBB-NEXT:    slli a0, a0, 23
+; RV64ZBAZBB-NEXT:    srli a0, a0, 32
+; RV64ZBAZBB-NEXT:    ret
+  %ext = sext i8 %a to i32
+  %1 = ashr i32 %ext, 9
+  ret i32 %1
+}
+
+; Make sure we use sext.h+slli+srli for Zba+Zbb.
+; FIXME: The RV64I and Zba only cases can be done with only 3 shifts.
+define zeroext i32 @sext_ashr_zext_i16(i16 %a) nounwind {
+; RV64I-LABEL: sext_ashr_zext_i16:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    slli a0, a0, 48
+; RV64I-NEXT:    srai a0, a0, 48
+; RV64I-NEXT:    slli a0, a0, 23
+; RV64I-NEXT:    srli a0, a0, 32
+; RV64I-NEXT:    ret
+;
+; RV64ZBANOZBB-LABEL: sext_ashr_zext_i16:
+; RV64ZBANOZBB:       # %bb.0:
+; RV64ZBANOZBB-NEXT:    slli a0, a0, 48
+; RV64ZBANOZBB-NEXT:    srai a0, a0, 48
+; RV64ZBANOZBB-NEXT:    slli a0, a0, 23
+; RV64ZBANOZBB-NEXT:    srli a0, a0, 32
+; RV64ZBANOZBB-NEXT:    ret
+;
+; RV64ZBAZBB-LABEL: sext_ashr_zext_i16:
+; RV64ZBAZBB:       # %bb.0:
+; RV64ZBAZBB-NEXT:    sext.h a0, a0
+; RV64ZBAZBB-NEXT:    slli a0, a0, 23
+; RV64ZBAZBB-NEXT:    srli a0, a0, 32
+; RV64ZBAZBB-NEXT:    ret
+  %ext = sext i16 %a to i32
+  %1 = ashr i32 %ext, 9
+  ret i32 %1
+}