[llvm] [RISCV] Support codegen for some scalar P extension instructions. (PR #164359)
via llvm-commits
llvm-commits at lists.llvm.org
Mon Oct 20 22:49:59 PDT 2025
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-risc-v
Author: Craig Topper (topperc)
<details>
<summary>Changes</summary>
This includes sext.b, sext.h, min/max, rev8, clz(w), and abs(w).
Test cases copied from rv32zbb.ll and rv64zbb.ll and pruned to what was needed for P. Eventually we should merge these back into a single test file, but I wanted to keep P separated while it is experimental.
---
Patch is 54.06 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/164359.diff
7 Files Affected:
- (modified) llvm/lib/Target/RISCV/RISCVFeatures.td (+2-1)
- (modified) llvm/lib/Target/RISCV/RISCVISelLowering.cpp (+18-5)
- (modified) llvm/lib/Target/RISCV/RISCVInstrInfoP.td (+14)
- (modified) llvm/lib/Target/RISCV/RISCVInstrInfoZb.td (+12-6)
- (modified) llvm/lib/Target/RISCV/RISCVSubtarget.h (+2-2)
- (added) llvm/test/CodeGen/RISCV/rv32p.ll (+709)
- (added) llvm/test/CodeGen/RISCV/rv64p.ll (+963)
``````````diff
diff --git a/llvm/lib/Target/RISCV/RISCVFeatures.td b/llvm/lib/Target/RISCV/RISCVFeatures.td
index 19992e667d192..e2bfbeb41db9c 100644
--- a/llvm/lib/Target/RISCV/RISCVFeatures.td
+++ b/llvm/lib/Target/RISCV/RISCVFeatures.td
@@ -1123,7 +1123,8 @@ def HasStdExtZbkbOrP
"'Base P' (Packed-SIMD)">;
def HasStdExtZbbOrZbkbOrP
- : Predicate<"Subtarget->HasStdExtZbbOrZbkb()|| Subtarget->hasStdExtP()">,
+ : Predicate<"Subtarget->hasStdExtZbb() || Subtarget->hasStdExtZbkb() || "
+ "Subtarget->hasStdExtP()">,
AssemblerPredicate<(any_of FeatureStdExtZbb, FeatureStdExtZbkb, FeatureStdExtP),
"'Zbb' (Basic Bit-Manipulation) or "
"'Zbkb' (Bitmanip instructions for Cryptography) or "
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 0a53ba935b1a7..877695f2ad8de 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -318,8 +318,9 @@ RISCVTargetLowering::RISCVTargetLowering(const TargetMachine &TM,
setOperationAction(ISD::EH_DWARF_CFA, MVT::i32, Custom);
- if (!Subtarget.hasStdExtZbb() && !Subtarget.hasVendorXTHeadBb() &&
- !Subtarget.hasVendorXqcibm() && !Subtarget.hasVendorXAndesPerf() &&
+ if (!Subtarget.hasStdExtZbb() && !Subtarget.hasStdExtP() &&
+ !Subtarget.hasVendorXTHeadBb() && !Subtarget.hasVendorXqcibm() &&
+ !Subtarget.hasVendorXAndesPerf() &&
!(Subtarget.hasVendorXCValu() && !Subtarget.is64Bit()))
setOperationAction(ISD::SIGN_EXTEND_INREG, {MVT::i8, MVT::i16}, Expand);
@@ -392,7 +393,7 @@ RISCVTargetLowering::RISCVTargetLowering(const TargetMachine &TM,
setOperationAction(ISD::BITREVERSE, MVT::i8, Custom);
}
- if (Subtarget.hasStdExtZbb() ||
+ if (Subtarget.hasStdExtZbb() || Subtarget.hasStdExtP() ||
(Subtarget.hasVendorXCValu() && !Subtarget.is64Bit())) {
setOperationAction({ISD::SMIN, ISD::SMAX, ISD::UMIN, ISD::UMAX}, XLenVT,
Legal);
@@ -419,14 +420,18 @@ RISCVTargetLowering::RISCVTargetLowering(const TargetMachine &TM,
// We need the custom lowering to make sure that the resulting sequence
// for the 32bit case is efficient on 64bit targets.
// Use default promotion for i32 without Zbb.
- if (Subtarget.is64Bit() && Subtarget.hasStdExtZbb())
+ if (Subtarget.is64Bit() &&
+ (Subtarget.hasStdExtZbb() || Subtarget.hasStdExtP()))
setOperationAction({ISD::CTLZ, ISD::CTLZ_ZERO_UNDEF}, MVT::i32, Custom);
} else {
setOperationAction(ISD::CTLZ, XLenVT, Expand);
}
- if (Subtarget.hasVendorXCValu() && !Subtarget.is64Bit()) {
+ if (Subtarget.hasStdExtP() ||
+ (Subtarget.hasVendorXCValu() && !Subtarget.is64Bit())) {
setOperationAction(ISD::ABS, XLenVT, Legal);
+ if (Subtarget.is64Bit())
+ setOperationAction(ISD::ABS, MVT::i32, Custom);
} else if (Subtarget.hasShortForwardBranchOpt()) {
// We can use PseudoCCSUB to implement ABS.
setOperationAction(ISD::ABS, XLenVT, Legal);
@@ -14791,6 +14796,14 @@ void RISCVTargetLowering::ReplaceNodeResults(SDNode *N,
assert(N->getValueType(0) == MVT::i32 && Subtarget.is64Bit() &&
"Unexpected custom legalisation");
+ if (Subtarget.hasStdExtP()) {
+ SDValue Src =
+ DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(0));
+ SDValue Abs = DAG.getNode(RISCVISD::ABSW_P, DL, MVT::i64, Src);
+ Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Abs));
+ return;
+ }
+
if (Subtarget.hasStdExtZbb()) {
// Emit a special ABSW node that will be expanded to NEGW+MAX at isel.
// This allows us to remember that the result is sign extended. Expanding
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoP.td b/llvm/lib/Target/RISCV/RISCVInstrInfoP.td
index 7d8a9192d9847..56044d2be4f27 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoP.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoP.td
@@ -1455,3 +1455,17 @@ let Predicates = [HasStdExtP, IsRV32] in {
def PMAXU_DW : RVPPairBinaryExchanged_rr<0b1111, 0b01, "pmaxu.dw">;
def PMAXU_DB : RVPPairBinaryExchanged_rr<0b1111, 0b10, "pmaxu.db">;
} // Predicates = [HasStdExtP, IsRV32]
+
+//===----------------------------------------------------------------------===//
+// Codegen patterns
+//===----------------------------------------------------------------------===//
+
+// TODO: Rename RISCVISD::ABSW to RISCVISD::NEGW_MAX or something so this can
+// be RISCVISD::ABSW.
+def riscv_absw_p : RVSDNode<"ABSW_P", SDTIntUnaryOp>;
+
+let Predicates = [HasStdExtP] in
+def : PatGpr<abs, ABS>;
+
+let Predicates = [HasStdExtP, IsRV64] in
+def : PatGpr<riscv_absw_p, ABSW>;
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoZb.td b/llvm/lib/Target/RISCV/RISCVInstrInfoZb.td
index 62b7bcd67283a..b895857529208 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoZb.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoZb.td
@@ -599,14 +599,20 @@ def : PatGpr<riscv_zip, ZIP_RV32, i32>;
def : PatGpr<riscv_unzip, UNZIP_RV32, i32>;
} // Predicates = [HasStdExtZbkb, IsRV32]
-let Predicates = [HasStdExtZbb] in {
+let Predicates = [HasStdExtZbbOrP] in {
def : PatGpr<ctlz, CLZ>;
+}
+
+let Predicates = [HasStdExtZbb] in {
def : PatGpr<cttz, CTZ>;
def : PatGpr<ctpop, CPOP>;
} // Predicates = [HasStdExtZbb]
-let Predicates = [HasStdExtZbb, IsRV64] in {
+let Predicates = [HasStdExtZbbOrP, IsRV64] in {
def : PatGpr<riscv_clzw, CLZW>;
+}
+
+let Predicates = [HasStdExtZbb, IsRV64] in {
def : PatGpr<riscv_ctzw, CTZW>;
def : Pat<(i64 (ctpop (i64 (zexti32 (i64 GPR:$rs1))))), (CPOPW GPR:$rs1)>;
@@ -614,22 +620,22 @@ def : Pat<(i64 (riscv_absw GPR:$rs1)),
(MAX GPR:$rs1, (XLenVT (SUBW (XLenVT X0), GPR:$rs1)))>;
} // Predicates = [HasStdExtZbb, IsRV64]
-let Predicates = [HasStdExtZbb] in {
+let Predicates = [HasStdExtZbbOrP] in {
def : Pat<(XLenVT (sext_inreg GPR:$rs1, i8)), (SEXT_B GPR:$rs1)>;
def : Pat<(XLenVT (sext_inreg GPR:$rs1, i16)), (SEXT_H GPR:$rs1)>;
} // Predicates = [HasStdExtZbb]
-let Predicates = [HasStdExtZbb] in {
+let Predicates = [HasStdExtZbbOrP] in {
def : PatGprGpr<smin, MIN>;
def : PatGprGpr<smax, MAX>;
def : PatGprGpr<umin, MINU>;
def : PatGprGpr<umax, MAXU>;
} // Predicates = [HasStdExtZbb]
-let Predicates = [HasStdExtZbbOrZbkb, IsRV32] in
+let Predicates = [HasStdExtZbbOrZbkbOrP, IsRV32] in
def : PatGpr<bswap, REV8_RV32, i32>;
-let Predicates = [HasStdExtZbbOrZbkb, IsRV64] in
+let Predicates = [HasStdExtZbbOrZbkbOrP, IsRV64] in
def : PatGpr<bswap, REV8_RV64, i64>;
let Predicates = [HasStdExtZbkb] in {
diff --git a/llvm/lib/Target/RISCV/RISCVSubtarget.h b/llvm/lib/Target/RISCV/RISCVSubtarget.h
index 334db4bfb75d9..4b4fc8f0d8e76 100644
--- a/llvm/lib/Target/RISCV/RISCVSubtarget.h
+++ b/llvm/lib/Target/RISCV/RISCVSubtarget.h
@@ -187,7 +187,7 @@ class RISCVSubtarget : public RISCVGenSubtargetInfo {
}
bool hasCLZLike() const {
- return HasStdExtZbb || HasVendorXTHeadBb ||
+ return HasStdExtZbb || HasStdExtP || HasVendorXTHeadBb ||
(HasVendorXCVbitmanip && !IsRV64);
}
bool hasCTZLike() const {
@@ -197,7 +197,7 @@ class RISCVSubtarget : public RISCVGenSubtargetInfo {
return HasStdExtZbb || (HasVendorXCVbitmanip && !IsRV64);
}
bool hasREV8Like() const {
- return HasStdExtZbb || HasStdExtZbkb || HasVendorXTHeadBb;
+ return HasStdExtZbb || HasStdExtZbkb || HasStdExtP || HasVendorXTHeadBb;
}
bool hasBEXTILike() const { return HasStdExtZbs || HasVendorXTHeadBs; }
diff --git a/llvm/test/CodeGen/RISCV/rv32p.ll b/llvm/test/CodeGen/RISCV/rv32p.ll
new file mode 100644
index 0000000000000..4eee880a398a9
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/rv32p.ll
@@ -0,0 +1,709 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+experimental-p -verify-machineinstrs < %s \
+; RUN: | FileCheck %s
+
+declare i32 @llvm.ctlz.i32(i32, i1)
+
+define i32 @ctlz_i32(i32 %a) nounwind {
+; CHECK-LABEL: ctlz_i32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: clz a0, a0
+; CHECK-NEXT: ret
+ %1 = call i32 @llvm.ctlz.i32(i32 %a, i1 false)
+ ret i32 %1
+}
+
+declare i64 @llvm.ctlz.i64(i64, i1)
+
+define i64 @ctlz_i64(i64 %a) nounwind {
+; CHECK-LABEL: ctlz_i64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: bnez a1, .LBB1_2
+; CHECK-NEXT: # %bb.1:
+; CHECK-NEXT: clz a0, a0
+; CHECK-NEXT: addi a0, a0, 32
+; CHECK-NEXT: li a1, 0
+; CHECK-NEXT: ret
+; CHECK-NEXT: .LBB1_2:
+; CHECK-NEXT: clz a0, a1
+; CHECK-NEXT: li a1, 0
+; CHECK-NEXT: ret
+ %1 = call i64 @llvm.ctlz.i64(i64 %a, i1 false)
+ ret i64 %1
+}
+
+declare i32 @llvm.cttz.i32(i32, i1)
+
+define i32 @cttz_i32(i32 %a) nounwind {
+; CHECK-LABEL: cttz_i32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: beqz a0, .LBB2_2
+; CHECK-NEXT: # %bb.1: # %cond.false
+; CHECK-NEXT: addi a1, a0, -1
+; CHECK-NEXT: not a0, a0
+; CHECK-NEXT: and a0, a0, a1
+; CHECK-NEXT: clz a0, a0
+; CHECK-NEXT: li a1, 32
+; CHECK-NEXT: sub a0, a1, a0
+; CHECK-NEXT: ret
+; CHECK-NEXT: .LBB2_2:
+; CHECK-NEXT: li a0, 32
+; CHECK-NEXT: ret
+ %1 = call i32 @llvm.cttz.i32(i32 %a, i1 false)
+ ret i32 %1
+}
+
+declare i64 @llvm.cttz.i64(i64, i1)
+
+define i64 @cttz_i64(i64 %a) nounwind {
+; CHECK-LABEL: cttz_i64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: or a2, a0, a1
+; CHECK-NEXT: beqz a2, .LBB3_3
+; CHECK-NEXT: # %bb.1: # %cond.false
+; CHECK-NEXT: bnez a0, .LBB3_4
+; CHECK-NEXT: # %bb.2: # %cond.false
+; CHECK-NEXT: addi a0, a1, -1
+; CHECK-NEXT: not a1, a1
+; CHECK-NEXT: and a0, a1, a0
+; CHECK-NEXT: clz a0, a0
+; CHECK-NEXT: li a1, 64
+; CHECK-NEXT: j .LBB3_5
+; CHECK-NEXT: .LBB3_3:
+; CHECK-NEXT: li a1, 0
+; CHECK-NEXT: li a0, 64
+; CHECK-NEXT: ret
+; CHECK-NEXT: .LBB3_4:
+; CHECK-NEXT: addi a1, a0, -1
+; CHECK-NEXT: not a0, a0
+; CHECK-NEXT: and a0, a0, a1
+; CHECK-NEXT: clz a0, a0
+; CHECK-NEXT: li a1, 32
+; CHECK-NEXT: .LBB3_5: # %cond.false
+; CHECK-NEXT: sub a0, a1, a0
+; CHECK-NEXT: li a1, 0
+; CHECK-NEXT: ret
+ %1 = call i64 @llvm.cttz.i64(i64 %a, i1 false)
+ ret i64 %1
+}
+
+define i32 @sextb_i32(i32 %a) nounwind {
+; CHECK-LABEL: sextb_i32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: sext.b a0, a0
+; CHECK-NEXT: ret
+ %shl = shl i32 %a, 24
+ %shr = ashr exact i32 %shl, 24
+ ret i32 %shr
+}
+
+define i64 @sextb_i64(i64 %a) nounwind {
+; CHECK-LABEL: sextb_i64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: sext.b a0, a0
+; CHECK-NEXT: srai a1, a0, 31
+; CHECK-NEXT: ret
+ %shl = shl i64 %a, 56
+ %shr = ashr exact i64 %shl, 56
+ ret i64 %shr
+}
+
+define i32 @sexth_i32(i32 %a) nounwind {
+; CHECK-LABEL: sexth_i32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: sext.h a0, a0
+; CHECK-NEXT: ret
+ %shl = shl i32 %a, 16
+ %shr = ashr exact i32 %shl, 16
+ ret i32 %shr
+}
+
+define i64 @sexth_i64(i64 %a) nounwind {
+; CHECK-LABEL: sexth_i64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: sext.h a0, a0
+; CHECK-NEXT: srai a1, a0, 31
+; CHECK-NEXT: ret
+ %shl = shl i64 %a, 48
+ %shr = ashr exact i64 %shl, 48
+ ret i64 %shr
+}
+
+define i32 @min_i32(i32 %a, i32 %b) nounwind {
+; CHECK-LABEL: min_i32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: min a0, a0, a1
+; CHECK-NEXT: ret
+ %cmp = icmp slt i32 %a, %b
+ %cond = select i1 %cmp, i32 %a, i32 %b
+ ret i32 %cond
+}
+
+; As we are not matching directly i64 code patterns on RV32 some i64 patterns
+; don't have yet any matching bit manipulation instructions on RV32.
+; This test is presented here in case future expansions of the Bitmanip
+; extensions introduce instructions suitable for this pattern.
+
+define i64 @min_i64(i64 %a, i64 %b) nounwind {
+; CHECK-LABEL: min_i64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: beq a1, a3, .LBB9_2
+; CHECK-NEXT: # %bb.1:
+; CHECK-NEXT: slt a4, a1, a3
+; CHECK-NEXT: beqz a4, .LBB9_3
+; CHECK-NEXT: j .LBB9_4
+; CHECK-NEXT: .LBB9_2:
+; CHECK-NEXT: sltu a4, a0, a2
+; CHECK-NEXT: bnez a4, .LBB9_4
+; CHECK-NEXT: .LBB9_3:
+; CHECK-NEXT: mv a0, a2
+; CHECK-NEXT: mv a1, a3
+; CHECK-NEXT: .LBB9_4:
+; CHECK-NEXT: ret
+ %cmp = icmp slt i64 %a, %b
+ %cond = select i1 %cmp, i64 %a, i64 %b
+ ret i64 %cond
+}
+
+define i32 @max_i32(i32 %a, i32 %b) nounwind {
+; CHECK-LABEL: max_i32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: max a0, a0, a1
+; CHECK-NEXT: ret
+ %cmp = icmp sgt i32 %a, %b
+ %cond = select i1 %cmp, i32 %a, i32 %b
+ ret i32 %cond
+}
+
+; As we are not matching directly i64 code patterns on RV32 some i64 patterns
+; don't have yet any matching bit manipulation instructions on RV32.
+; This test is presented here in case future expansions of the Bitmanip
+; extensions introduce instructions suitable for this pattern.
+
+define i64 @max_i64(i64 %a, i64 %b) nounwind {
+; CHECK-LABEL: max_i64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: beq a1, a3, .LBB11_2
+; CHECK-NEXT: # %bb.1:
+; CHECK-NEXT: slt a4, a3, a1
+; CHECK-NEXT: beqz a4, .LBB11_3
+; CHECK-NEXT: j .LBB11_4
+; CHECK-NEXT: .LBB11_2:
+; CHECK-NEXT: sltu a4, a2, a0
+; CHECK-NEXT: bnez a4, .LBB11_4
+; CHECK-NEXT: .LBB11_3:
+; CHECK-NEXT: mv a0, a2
+; CHECK-NEXT: mv a1, a3
+; CHECK-NEXT: .LBB11_4:
+; CHECK-NEXT: ret
+ %cmp = icmp sgt i64 %a, %b
+ %cond = select i1 %cmp, i64 %a, i64 %b
+ ret i64 %cond
+}
+
+define i32 @minu_i32(i32 %a, i32 %b) nounwind {
+; CHECK-LABEL: minu_i32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: minu a0, a0, a1
+; CHECK-NEXT: ret
+ %cmp = icmp ult i32 %a, %b
+ %cond = select i1 %cmp, i32 %a, i32 %b
+ ret i32 %cond
+}
+
+; As we are not matching directly i64 code patterns on RV32 some i64 patterns
+; don't have yet any matching bit manipulation instructions on RV32.
+; This test is presented here in case future expansions of the Bitmanip
+; extensions introduce instructions suitable for this pattern.
+
+define i64 @minu_i64(i64 %a, i64 %b) nounwind {
+; CHECK-LABEL: minu_i64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: beq a1, a3, .LBB13_2
+; CHECK-NEXT: # %bb.1:
+; CHECK-NEXT: sltu a4, a1, a3
+; CHECK-NEXT: beqz a4, .LBB13_3
+; CHECK-NEXT: j .LBB13_4
+; CHECK-NEXT: .LBB13_2:
+; CHECK-NEXT: sltu a4, a0, a2
+; CHECK-NEXT: bnez a4, .LBB13_4
+; CHECK-NEXT: .LBB13_3:
+; CHECK-NEXT: mv a0, a2
+; CHECK-NEXT: mv a1, a3
+; CHECK-NEXT: .LBB13_4:
+; CHECK-NEXT: ret
+ %cmp = icmp ult i64 %a, %b
+ %cond = select i1 %cmp, i64 %a, i64 %b
+ ret i64 %cond
+}
+
+define i32 @maxu_i32(i32 %a, i32 %b) nounwind {
+; CHECK-LABEL: maxu_i32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: maxu a0, a0, a1
+; CHECK-NEXT: ret
+ %cmp = icmp ugt i32 %a, %b
+ %cond = select i1 %cmp, i32 %a, i32 %b
+ ret i32 %cond
+}
+
+; As we are not matching directly i64 code patterns on RV32 some i64 patterns
+; don't have yet any matching bit manipulation instructions on RV32.
+; This test is presented here in case future expansions of the Bitmanip
+; extensions introduce instructions suitable for this pattern.
+
+define i64 @maxu_i64(i64 %a, i64 %b) nounwind {
+; CHECK-LABEL: maxu_i64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: beq a1, a3, .LBB15_2
+; CHECK-NEXT: # %bb.1:
+; CHECK-NEXT: sltu a4, a3, a1
+; CHECK-NEXT: beqz a4, .LBB15_3
+; CHECK-NEXT: j .LBB15_4
+; CHECK-NEXT: .LBB15_2:
+; CHECK-NEXT: sltu a4, a2, a0
+; CHECK-NEXT: bnez a4, .LBB15_4
+; CHECK-NEXT: .LBB15_3:
+; CHECK-NEXT: mv a0, a2
+; CHECK-NEXT: mv a1, a3
+; CHECK-NEXT: .LBB15_4:
+; CHECK-NEXT: ret
+ %cmp = icmp ugt i64 %a, %b
+ %cond = select i1 %cmp, i64 %a, i64 %b
+ ret i64 %cond
+}
+
+declare i32 @llvm.abs.i32(i32, i1 immarg)
+
+define i32 @abs_i32(i32 %x) {
+; CHECK-LABEL: abs_i32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: abs a0, a0
+; CHECK-NEXT: ret
+ %abs = tail call i32 @llvm.abs.i32(i32 %x, i1 true)
+ ret i32 %abs
+}
+
+declare i64 @llvm.abs.i64(i64, i1 immarg)
+
+define i64 @abs_i64(i64 %x) {
+; CHECK-LABEL: abs_i64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: bgez a1, .LBB17_2
+; CHECK-NEXT: # %bb.1:
+; CHECK-NEXT: snez a2, a0
+; CHECK-NEXT: neg a0, a0
+; CHECK-NEXT: neg a1, a1
+; CHECK-NEXT: sub a1, a1, a2
+; CHECK-NEXT: .LBB17_2:
+; CHECK-NEXT: ret
+ %abs = tail call i64 @llvm.abs.i64(i64 %x, i1 true)
+ ret i64 %abs
+}
+
+define i32 @zexth_i32(i32 %a) nounwind {
+; CHECK-LABEL: zexth_i32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: slli a0, a0, 16
+; CHECK-NEXT: srli a0, a0, 16
+; CHECK-NEXT: ret
+ %and = and i32 %a, 65535
+ ret i32 %and
+}
+
+define i64 @zexth_i64(i64 %a) nounwind {
+; CHECK-LABEL: zexth_i64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: slli a0, a0, 16
+; CHECK-NEXT: srli a0, a0, 16
+; CHECK-NEXT: li a1, 0
+; CHECK-NEXT: ret
+ %and = and i64 %a, 65535
+ ret i64 %and
+}
+
+declare i32 @llvm.bswap.i32(i32)
+
+define i32 @bswap_i32(i32 %a) nounwind {
+; CHECK-LABEL: bswap_i32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: rev8 a0, a0
+; CHECK-NEXT: ret
+ %1 = tail call i32 @llvm.bswap.i32(i32 %a)
+ ret i32 %1
+}
+
+declare i64 @llvm.bswap.i64(i64)
+
+define i64 @bswap_i64(i64 %a) {
+; CHECK-LABEL: bswap_i64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: rev8 a2, a1
+; CHECK-NEXT: rev8 a1, a0
+; CHECK-NEXT: mv a0, a2
+; CHECK-NEXT: ret
+ %1 = call i64 @llvm.bswap.i64(i64 %a)
+ ret i64 %1
+}
+
+define i32 @srai_slli(i16 signext %0) {
+; CHECK-LABEL: srai_slli:
+; CHECK: # %bb.0:
+; CHECK-NEXT: slli a0, a0, 25
+; CHECK-NEXT: srai a0, a0, 31
+; CHECK-NEXT: ret
+ %2 = shl i16 %0, 9
+ %sext = ashr i16 %2, 15
+ %3 = sext i16 %sext to i32
+ ret i32 %3
+}
+
+define i32 @srai_slli2(i16 signext %0) {
+; CHECK-LABEL: srai_slli2:
+; CHECK: # %bb.0:
+; CHECK-NEXT: slli a0, a0, 25
+; CHECK-NEXT: srai a0, a0, 30
+; CHECK-NEXT: ret
+ %2 = shl i16 %0, 9
+ %sext = ashr i16 %2, 14
+ %3 = sext i16 %sext to i32
+ ret i32 %3
+}
+define i8 @sub_if_uge_i8(i8 %x, i8 %y) {
+; CHECK-LABEL: sub_if_uge_i8:
+; CHECK: # %bb.0:
+; CHECK-NEXT: zext.b a2, a0
+; CHECK-NEXT: sub a0, a0, a1
+; CHECK-NEXT: zext.b a0, a0
+; CHECK-NEXT: minu a0, a2, a0
+; CHECK-NEXT: ret
+ %cmp = icmp ult i8 %x, %y
+ %select = select i1 %cmp, i8 0, i8 %y
+ %sub = sub nuw i8 %x, %select
+ ret i8 %sub
+}
+
+define i16 @sub_if_uge_i16(i16 %x, i16 %y) {
+; CHECK-LABEL: sub_if_uge_i16:
+; CHECK: # %bb.0:
+; CHECK-NEXT: lui a2, 16
+; CHECK-NEXT: sub a1, a0, a1
+; CHECK-NEXT: addi a2, a2, -1
+; CHECK-NEXT: and a0, a0, a2
+; CHECK-NEXT: and a1, a1, a2
+; CHECK-NEXT: minu a0, a0, a1
+; CHECK-NEXT: ret
+ %cmp = icmp ult i16 %x, %y
+ %select = select i1 %cmp, i16 0, i16 %y
+ %sub = sub nuw i16 %x, %select
+ ret i16 %sub
+}
+
+define i32 @sub_if_uge_i32(i32 %x, i32 %y) {
+; CHECK-LABEL: sub_if_uge_i32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: sub a1, a0, a1
+; CHECK-NEXT: minu a0, a0, a1
+; CHECK-NEXT: ret
+ %cmp = icmp ult i32 %x, %y
+ %select = select i1 %cmp, i32 0, i32 %y
+ %sub = sub nuw i32 %x, %select
+ ret i32 %sub
+}
+
+define i64 @sub_if_uge_i64(i64 %x, i64 %y) {
+; CHECK-LABEL: sub_if_uge_i64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: beq a1, a3, .LBB27_2
+; CHECK-NEXT: # %bb.1:
+; CHECK-NEXT: sltu a4, a1, a3
+; CHECK-NEXT: j .LBB27_3
+; CHECK-NEXT: .LBB27_2:
+; CHECK-NEXT: sltu a4, a0, a2
+; CHECK-NEXT: .LBB27_3:
+; CHECK-NEXT: addi a4, a4, -1
+; CHECK-NEXT: and a3, a4, a3
+; CHECK-NEXT: and a2, a4, a2
+; CHECK-NEXT: sltu a4, a0, a2
+; CHECK-NEXT: sub a1, a1, a3
+; CHECK-NEXT: sub a1, a1, a4
+; CHECK-NEXT: sub a0, a0, a2
+; CHECK-NEXT: ret
+ %cmp = icmp ult i64 %x, %y
+ %select = select i1 %cmp, i64 0, i64 %y
+ %sub = sub nuw i64 %x, %select
+ ret i64 %sub
+}
+
+define i128 @sub_if_uge_i128(i128 %x, i128 %y) {
+; CHECK-LABEL: sub_if_uge_i128:
+; CHECK: # %bb.0:
+; CHECK-NEXT: lw a3, 4(a1)
+; CHECK-NEXT: lw a4, 8(a1)
+; CHECK-NEXT: lw a5, 12(a1)
+; CHECK-NEXT: lw a6, 4(a2)
+; CHECK-NEXT: lw t0, 12(a2)
+; CHECK-NEXT: lw a7, 8(a2)
+; CHECK-NEXT: beq a5, t0, .LBB28_2
+; CHECK-NEXT: # %bb.1:
+; CHECK-NEXT: sltu t1, a5, t0
+; CHECK-NEXT: j .LBB28_3
+; CHECK-NEXT: .LBB28_2:
+; CHECK-NEXT: sltu t1, a4, a7
+; CHECK-NEXT: .LBB28_3:
+; CHECK-NEXT: lw a1, 0(a1)
+; CHECK-NEXT: lw a2, 0(a2)
+; CHECK-NEXT: beq a3, a6, .LBB28_5
+; CHECK-NEXT: # %bb.4:
+; CHECK-NEXT: sltu t2, a3, a6
+; CHECK-NEXT: j .LB...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/164359
More information about the llvm-commits
mailing list