[llvm] 70deb7b - [BPF] expand cttz, ctlz for i32, i64 (#73668)

via llvm-commits llvm-commits at lists.llvm.org
Sun Mar 31 19:57:58 PDT 2024


Author: Yingchi Long
Date: 2024-04-01T10:57:54+08:00
New Revision: 70deb7bfe90af91c68454b70683fbe98feaea87d

URL: https://github.com/llvm/llvm-project/commit/70deb7bfe90af91c68454b70683fbe98feaea87d
DIFF: https://github.com/llvm/llvm-project/commit/70deb7bfe90af91c68454b70683fbe98feaea87d.diff

LOG: [BPF] expand cttz, ctlz for i32, i64 (#73668)

Fixes: https://github.com/llvm/llvm-project/issues/62252

Depends on: #73667

Added: 
    llvm/test/CodeGen/BPF/cttz-ctlz.ll

Modified: 
    llvm/lib/Target/BPF/BPFISelLowering.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/BPF/BPFISelLowering.cpp b/llvm/lib/Target/BPF/BPFISelLowering.cpp
index b8ca8ec98c06e5..071fe004806e3e 100644
--- a/llvm/lib/Target/BPF/BPFISelLowering.cpp
+++ b/llvm/lib/Target/BPF/BPFISelLowering.cpp
@@ -113,6 +113,10 @@ BPFTargetLowering::BPFTargetLowering(const TargetMachine &TM,
     setOperationAction(ISD::SRL_PARTS, VT, Expand);
     setOperationAction(ISD::SRA_PARTS, VT, Expand);
     setOperationAction(ISD::CTPOP, VT, Expand);
+    setOperationAction(ISD::CTTZ, VT, Expand);
+    setOperationAction(ISD::CTLZ, VT, Expand);
+    setOperationAction(ISD::CTTZ_ZERO_UNDEF, VT, Expand);
+    setOperationAction(ISD::CTLZ_ZERO_UNDEF, VT, Expand);
 
     setOperationAction(ISD::SETCC, VT, Expand);
     setOperationAction(ISD::SELECT, VT, Expand);
@@ -125,11 +129,6 @@ BPFTargetLowering::BPFTargetLowering(const TargetMachine &TM,
                        STI.getHasJmp32() ? Custom : Promote);
   }
 
-  setOperationAction(ISD::CTTZ, MVT::i64, Custom);
-  setOperationAction(ISD::CTLZ, MVT::i64, Custom);
-  setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i64, Custom);
-  setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i64, Custom);
-
   setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
   if (!STI.hasMovsx()) {
     setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i8, Expand);

diff  --git a/llvm/test/CodeGen/BPF/cttz-ctlz.ll b/llvm/test/CodeGen/BPF/cttz-ctlz.ll
new file mode 100644
index 00000000000000..f42b2e2d10871b
--- /dev/null
+++ b/llvm/test/CodeGen/BPF/cttz-ctlz.ll
@@ -0,0 +1,304 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
+; RUN: llc < %s -march=bpf | FileCheck %s
+
+; test that we can expand CTTZ & CTLZ
+
+declare i32 @llvm.cttz.i32(i32, i1)
+
+define i32 @cttz_i32_zdef(i32 %a) {
+; CHECK-LABEL: cttz_i32_zdef:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 = -r2
+; CHECK-NEXT:    r1 &= r2
+; CHECK-NEXT:    r1 *= 125613361
+; CHECK-NEXT:    r2 = 4160749568 ll
+; CHECK-NEXT:    r1 &= r2
+; CHECK-NEXT:    r1 >>= 27
+; CHECK-NEXT:    r2 = {{\.?LCPI[0-9]+_[0-9]+}} ll
+; CHECK-NEXT:    r2 += r1
+; CHECK-NEXT:    r0 = *(u8 *)(r2 + 0)
+; CHECK-NEXT:    exit
+    %ret = call i32 @llvm.cttz.i32(i32 %a, i1 1)
+    ret i32 %ret
+}
+
+
+define i32 @cttz_i32(i32 %a) {
+; CHECK-LABEL: cttz_i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    r0 = 32
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 <<= 32
+; CHECK-NEXT:    r2 >>= 32
+; CHECK-NEXT:    if r2 == 0 goto LBB1_2
+; CHECK-NEXT:  # %bb.1: # %cond.false
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 = -r2
+; CHECK-NEXT:    r1 &= r2
+; CHECK-NEXT:    r1 *= 125613361
+; CHECK-NEXT:    r2 = 4160749568 ll
+; CHECK-NEXT:    r1 &= r2
+; CHECK-NEXT:    r1 >>= 27
+; CHECK-NEXT:    r2 = {{\.?LCPI[0-9]+_[0-9]+}} ll
+; CHECK-NEXT:    r2 += r1
+; CHECK-NEXT:    r0 = *(u8 *)(r2 + 0)
+; CHECK-NEXT:  LBB1_2: # %cond.end
+; CHECK-NEXT:    exit
+    %ret = call i32 @llvm.cttz.i32(i32 %a, i1 0)
+    ret i32 %ret
+}
+
+declare i64 @llvm.cttz.i64(i64, i1)
+
+define i64 @cttz_i64_zdef(i64 %a) {
+; CHECK-LABEL: cttz_i64_zdef:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 = -r2
+; CHECK-NEXT:    r1 &= r2
+; CHECK-NEXT:    r2 = 151050438420815295 ll
+; CHECK-NEXT:    r1 *= r2
+; CHECK-NEXT:    r1 >>= 58
+; CHECK-NEXT:    r2 = {{\.?LCPI[0-9]+_[0-9]+}} ll
+; CHECK-NEXT:    r2 += r1
+; CHECK-NEXT:    r0 = *(u8 *)(r2 + 0)
+; CHECK-NEXT:    exit
+    %ret = call i64 @llvm.cttz.i64(i64 %a, i1 1)
+    ret i64 %ret
+}
+
+
+define i64 @cttz_i64(i64 %a) {
+; CHECK-LABEL: cttz_i64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    r0 = 64
+; CHECK-NEXT:    if r1 == 0 goto LBB3_2
+; CHECK-NEXT:  # %bb.1: # %cond.false
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 = -r2
+; CHECK-NEXT:    r1 &= r2
+; CHECK-NEXT:    r2 = 151050438420815295 ll
+; CHECK-NEXT:    r1 *= r2
+; CHECK-NEXT:    r1 >>= 58
+; CHECK-NEXT:    r2 = {{\.?LCPI[0-9]+_[0-9]+}} ll
+; CHECK-NEXT:    r2 += r1
+; CHECK-NEXT:    r0 = *(u8 *)(r2 + 0)
+; CHECK-NEXT:  LBB3_2: # %cond.end
+; CHECK-NEXT:    exit
+    %ret = call i64 @llvm.cttz.i64(i64 %a, i1 0)
+    ret i64 %ret
+}
+
+
+declare i32 @llvm.ctlz.i32(i32, i1)
+
+define i32 @ctlz_i32_zdef(i32 %a) {
+; CHECK-LABEL: ctlz_i32_zdef:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    r2 = 4294967294 ll
+; CHECK-NEXT:    r3 = r1
+; CHECK-NEXT:    r3 &= r2
+; CHECK-NEXT:    r3 >>= 1
+; CHECK-NEXT:    r1 |= r3
+; CHECK-NEXT:    r2 = 4294967292 ll
+; CHECK-NEXT:    r3 = r1
+; CHECK-NEXT:    r3 &= r2
+; CHECK-NEXT:    r3 >>= 2
+; CHECK-NEXT:    r1 |= r3
+; CHECK-NEXT:    r2 = 4294967280 ll
+; CHECK-NEXT:    r3 = r1
+; CHECK-NEXT:    r3 &= r2
+; CHECK-NEXT:    r3 >>= 4
+; CHECK-NEXT:    r1 |= r3
+; CHECK-NEXT:    r2 = 4294967040 ll
+; CHECK-NEXT:    r3 = r1
+; CHECK-NEXT:    r3 &= r2
+; CHECK-NEXT:    r3 >>= 8
+; CHECK-NEXT:    r1 |= r3
+; CHECK-NEXT:    r2 = 4294901760 ll
+; CHECK-NEXT:    r3 = r1
+; CHECK-NEXT:    r3 &= r2
+; CHECK-NEXT:    r3 >>= 16
+; CHECK-NEXT:    r1 |= r3
+; CHECK-NEXT:    r1 ^= -1
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 >>= 1
+; CHECK-NEXT:    r2 &= 1431655765
+; CHECK-NEXT:    r1 -= r2
+; CHECK-NEXT:    r0 = r1
+; CHECK-NEXT:    r0 &= 858993459
+; CHECK-NEXT:    r1 >>= 2
+; CHECK-NEXT:    r1 &= 858993459
+; CHECK-NEXT:    r0 += r1
+; CHECK-NEXT:    r1 = r0
+; CHECK-NEXT:    r1 >>= 4
+; CHECK-NEXT:    r0 += r1
+; CHECK-NEXT:    r0 &= 252645135
+; CHECK-NEXT:    r0 *= 16843009
+; CHECK-NEXT:    r1 = 4278190080 ll
+; CHECK-NEXT:    r0 &= r1
+; CHECK-NEXT:    r0 >>= 24
+; CHECK-NEXT:    exit
+    %ret = call i32 @llvm.ctlz.i32(i32 %a, i1 1)
+    ret i32 %ret
+}
+
+
+define i32 @ctlz_i32(i32 %a) {
+; CHECK-LABEL: ctlz_i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    r0 = 32
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 <<= 32
+; CHECK-NEXT:    r2 >>= 32
+; CHECK-NEXT:    if r2 == 0 goto LBB5_2
+; CHECK-NEXT:  # %bb.1: # %cond.false
+; CHECK-NEXT:    r2 = 4294967294 ll
+; CHECK-NEXT:    r3 = r1
+; CHECK-NEXT:    r3 &= r2
+; CHECK-NEXT:    r3 >>= 1
+; CHECK-NEXT:    r1 |= r3
+; CHECK-NEXT:    r2 = 4294967292 ll
+; CHECK-NEXT:    r3 = r1
+; CHECK-NEXT:    r3 &= r2
+; CHECK-NEXT:    r3 >>= 2
+; CHECK-NEXT:    r1 |= r3
+; CHECK-NEXT:    r2 = 4294967280 ll
+; CHECK-NEXT:    r3 = r1
+; CHECK-NEXT:    r3 &= r2
+; CHECK-NEXT:    r3 >>= 4
+; CHECK-NEXT:    r1 |= r3
+; CHECK-NEXT:    r2 = 4294967040 ll
+; CHECK-NEXT:    r3 = r1
+; CHECK-NEXT:    r3 &= r2
+; CHECK-NEXT:    r3 >>= 8
+; CHECK-NEXT:    r1 |= r3
+; CHECK-NEXT:    r2 = 4294901760 ll
+; CHECK-NEXT:    r3 = r1
+; CHECK-NEXT:    r3 &= r2
+; CHECK-NEXT:    r3 >>= 16
+; CHECK-NEXT:    r1 |= r3
+; CHECK-NEXT:    r1 ^= -1
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 >>= 1
+; CHECK-NEXT:    r2 &= 1431655765
+; CHECK-NEXT:    r1 -= r2
+; CHECK-NEXT:    r0 = r1
+; CHECK-NEXT:    r0 &= 858993459
+; CHECK-NEXT:    r1 >>= 2
+; CHECK-NEXT:    r1 &= 858993459
+; CHECK-NEXT:    r0 += r1
+; CHECK-NEXT:    r1 = r0
+; CHECK-NEXT:    r1 >>= 4
+; CHECK-NEXT:    r0 += r1
+; CHECK-NEXT:    r0 &= 252645135
+; CHECK-NEXT:    r0 *= 16843009
+; CHECK-NEXT:    r1 = 4278190080 ll
+; CHECK-NEXT:    r0 &= r1
+; CHECK-NEXT:    r0 >>= 24
+; CHECK-NEXT:  LBB5_2: # %cond.end
+; CHECK-NEXT:    exit
+    %ret = call i32 @llvm.ctlz.i32(i32 %a, i1 0)
+    ret i32 %ret
+}
+
+declare i64 @llvm.ctlz.i64(i64, i1)
+
+define i64 @ctlz_i64_zdef(i64 %a) {
+; CHECK-LABEL: ctlz_i64_zdef:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 >>= 1
+; CHECK-NEXT:    r1 |= r2
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 >>= 2
+; CHECK-NEXT:    r1 |= r2
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 >>= 4
+; CHECK-NEXT:    r1 |= r2
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 >>= 8
+; CHECK-NEXT:    r1 |= r2
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 >>= 16
+; CHECK-NEXT:    r1 |= r2
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 >>= 32
+; CHECK-NEXT:    r1 |= r2
+; CHECK-NEXT:    r1 ^= -1
+; CHECK-NEXT:    r2 = 6148914691236517205 ll
+; CHECK-NEXT:    r3 = r1
+; CHECK-NEXT:    r3 >>= 1
+; CHECK-NEXT:    r3 &= r2
+; CHECK-NEXT:    r1 -= r3
+; CHECK-NEXT:    r2 = 3689348814741910323 ll
+; CHECK-NEXT:    r0 = r1
+; CHECK-NEXT:    r0 &= r2
+; CHECK-NEXT:    r1 >>= 2
+; CHECK-NEXT:    r1 &= r2
+; CHECK-NEXT:    r0 += r1
+; CHECK-NEXT:    r1 = r0
+; CHECK-NEXT:    r1 >>= 4
+; CHECK-NEXT:    r0 += r1
+; CHECK-NEXT:    r1 = 1085102592571150095 ll
+; CHECK-NEXT:    r0 &= r1
+; CHECK-NEXT:    r1 = 72340172838076673 ll
+; CHECK-NEXT:    r0 *= r1
+; CHECK-NEXT:    r0 >>= 56
+; CHECK-NEXT:    exit
+    %ret = call i64 @llvm.ctlz.i64(i64 %a, i1 1)
+    ret i64 %ret
+}
+
+
+define i64 @ctlz_i64(i64 %a) {
+; CHECK-LABEL: ctlz_i64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    r0 = 64
+; CHECK-NEXT:    if r1 == 0 goto LBB7_2
+; CHECK-NEXT:  # %bb.1: # %cond.false
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 >>= 1
+; CHECK-NEXT:    r1 |= r2
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 >>= 2
+; CHECK-NEXT:    r1 |= r2
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 >>= 4
+; CHECK-NEXT:    r1 |= r2
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 >>= 8
+; CHECK-NEXT:    r1 |= r2
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 >>= 16
+; CHECK-NEXT:    r1 |= r2
+; CHECK-NEXT:    r2 = r1
+; CHECK-NEXT:    r2 >>= 32
+; CHECK-NEXT:    r1 |= r2
+; CHECK-NEXT:    r1 ^= -1
+; CHECK-NEXT:    r2 = 6148914691236517205 ll
+; CHECK-NEXT:    r3 = r1
+; CHECK-NEXT:    r3 >>= 1
+; CHECK-NEXT:    r3 &= r2
+; CHECK-NEXT:    r1 -= r3
+; CHECK-NEXT:    r2 = 3689348814741910323 ll
+; CHECK-NEXT:    r0 = r1
+; CHECK-NEXT:    r0 &= r2
+; CHECK-NEXT:    r1 >>= 2
+; CHECK-NEXT:    r1 &= r2
+; CHECK-NEXT:    r0 += r1
+; CHECK-NEXT:    r1 = r0
+; CHECK-NEXT:    r1 >>= 4
+; CHECK-NEXT:    r0 += r1
+; CHECK-NEXT:    r1 = 1085102592571150095 ll
+; CHECK-NEXT:    r0 &= r1
+; CHECK-NEXT:    r1 = 72340172838076673 ll
+; CHECK-NEXT:    r0 *= r1
+; CHECK-NEXT:    r0 >>= 56
+; CHECK-NEXT:  LBB7_2: # %cond.end
+; CHECK-NEXT:    exit
+    %ret = call i64 @llvm.ctlz.i64(i64 %a, i1 0)
+    ret i64 %ret
+}
+


        


More information about the llvm-commits mailing list