Merge branch 'main' into undor/execute_region_no_inline

Author: undor

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

@@ -422,12 +422,9 @@ 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, MVT::i32, Custom);
-      // Use default promotion for XTHeadBb.
-      if (Subtarget.hasStdExtZbb())
-        setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i32, Custom);
-    }
+    // Use default promotion for i32 without Zbb.
+    if (Subtarget.is64Bit() && Subtarget.hasStdExtZbb())
+      setOperationAction({ISD::CTLZ, ISD::CTLZ_ZERO_UNDEF}, MVT::i32, Custom);
   } else {
     setOperationAction(ISD::CTLZ, XLenVT, Expand);
   }

llvm/lib/Target/RISCV/RISCVInstrInfoXTHead.td

@@ -594,8 +594,6 @@ def : Pat<(i64 (sra (bswap GPR:$rs1), (i64 32))),
           (TH_REVW GPR:$rs1)>;
 def : Pat<(binop_allwusers<srl> (bswap GPR:$rs1), (i64 32)),
           (TH_REVW GPR:$rs1)>;
-def : Pat<(riscv_clzw GPR:$rs1),
-          (TH_FF0 (i64 (SLLI (i64 (XORI GPR:$rs1, -1)), 32)))>;
 } // Predicates = [HasVendorXTHeadBb, IsRV64]
 
 let Predicates = [HasVendorXTHeadBs] in {

llvm/test/CodeGen/RISCV/ctlz-cttz-ctpop.ll

@@ -1375,9 +1375,9 @@ define i32 @test_ctlz_i32(i32 %a) nounwind {
 ;
 ; RV64XTHEADBB-LABEL: test_ctlz_i32:
 ; RV64XTHEADBB:       # %bb.0:
-; RV64XTHEADBB-NEXT:    not a0, a0
-; RV64XTHEADBB-NEXT:    slli a0, a0, 32
-; RV64XTHEADBB-NEXT:    th.ff0 a0, a0
+; RV64XTHEADBB-NEXT:    th.extu a0, a0, 31, 0
+; RV64XTHEADBB-NEXT:    th.ff1 a0, a0
+; RV64XTHEADBB-NEXT:    addi a0, a0, -32
 ; RV64XTHEADBB-NEXT:    ret
   %tmp = call i32 @llvm.ctlz.i32(i32 %a, i1 false)
   ret i32 %tmp

llvm/test/CodeGen/RISCV/rv64xtheadbb.ll

@@ -52,9 +52,9 @@ define signext i32 @ctlz_i32(i32 signext %a) nounwind {
 ;
 ; RV64XTHEADBB-NOB-LABEL: ctlz_i32:
 ; RV64XTHEADBB-NOB:       # %bb.0:
-; RV64XTHEADBB-NOB-NEXT:    not a0, a0
-; RV64XTHEADBB-NOB-NEXT:    slli a0, a0, 32
-; RV64XTHEADBB-NOB-NEXT:    th.ff0 a0, a0
+; RV64XTHEADBB-NOB-NEXT:    th.extu a0, a0, 31, 0
+; RV64XTHEADBB-NOB-NEXT:    th.ff1 a0, a0
+; RV64XTHEADBB-NOB-NEXT:    addi a0, a0, -32
 ; RV64XTHEADBB-NOB-NEXT:    ret
 ;
 ; RV64XTHEADBB-B-LABEL: ctlz_i32:
@@ -112,10 +112,9 @@ define signext i32 @log2_i32(i32 signext %a) nounwind {
 ;
 ; RV64XTHEADBB-NOB-LABEL: log2_i32:
 ; RV64XTHEADBB-NOB:       # %bb.0:
-; RV64XTHEADBB-NOB-NEXT:    not a0, a0
-; RV64XTHEADBB-NOB-NEXT:    slli a0, a0, 32
-; RV64XTHEADBB-NOB-NEXT:    th.ff0 a0, a0
-; RV64XTHEADBB-NOB-NEXT:    li a1, 31
+; RV64XTHEADBB-NOB-NEXT:    th.extu a0, a0, 31, 0
+; RV64XTHEADBB-NOB-NEXT:    th.ff1 a0, a0
+; RV64XTHEADBB-NOB-NEXT:    li a1, 63
 ; RV64XTHEADBB-NOB-NEXT:    sub a0, a1, a0
 ; RV64XTHEADBB-NOB-NEXT:    ret
 ;
@@ -177,10 +176,9 @@ define signext i32 @log2_ceil_i32(i32 signext %a) nounwind {
 ; RV64XTHEADBB-NOB-LABEL: log2_ceil_i32:
 ; RV64XTHEADBB-NOB:       # %bb.0:
 ; RV64XTHEADBB-NOB-NEXT:    addi a0, a0, -1
-; RV64XTHEADBB-NOB-NEXT:    not a0, a0
-; RV64XTHEADBB-NOB-NEXT:    slli a0, a0, 32
-; RV64XTHEADBB-NOB-NEXT:    th.ff0 a0, a0
-; RV64XTHEADBB-NOB-NEXT:    li a1, 32
+; RV64XTHEADBB-NOB-NEXT:    th.extu a0, a0, 31, 0
+; RV64XTHEADBB-NOB-NEXT:    th.ff1 a0, a0
+; RV64XTHEADBB-NOB-NEXT:    li a1, 64
 ; RV64XTHEADBB-NOB-NEXT:    sub a0, a1, a0
 ; RV64XTHEADBB-NOB-NEXT:    ret
 ;
@@ -309,9 +307,8 @@ define i32 @ctlz_lshr_i32(i32 signext %a) {
 ; RV64XTHEADBB-NOB-LABEL: ctlz_lshr_i32:
 ; RV64XTHEADBB-NOB:       # %bb.0:
 ; RV64XTHEADBB-NOB-NEXT:    srliw a0, a0, 1
-; RV64XTHEADBB-NOB-NEXT:    not a0, a0
-; RV64XTHEADBB-NOB-NEXT:    slli a0, a0, 32
-; RV64XTHEADBB-NOB-NEXT:    th.ff0 a0, a0
+; RV64XTHEADBB-NOB-NEXT:    th.ff1 a0, a0
+; RV64XTHEADBB-NOB-NEXT:    addi a0, a0, -32
 ; RV64XTHEADBB-NOB-NEXT:    ret
 ;
 ; RV64XTHEADBB-B-LABEL: ctlz_lshr_i32: