From 7c0b9b379b0e36ec1ed76ab5a4386e80dafd53fc Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper@sifive.com>
Date: Mon, 18 Jul 2022 09:10:53 -0700
Subject: [PATCH] [RISCV] Add isel patterns for ineg+setge/le/uge/ule.

setge/le/uge/ule selected by themselves require an xori with 1.
If we're negating the setcc, we can fold the xori with the neg
to create an addi with -1.

This works because xori X, 1 is equivalent to 1 - X if X is either
0 or 1. So we're doing -(1 - X) which is X-1 or X+-1.

This improves the code for selecting between 0 and -1 based on a
condition for some conditions.

Reviewed By: asb

Differential Revision: https://reviews.llvm.org/D129957
---
 llvm/lib/Target/RISCV/RISCVInstrInfo.td |   7 +
 llvm/test/CodeGen/RISCV/select-const.ll | 546 ++++++++++++++++++++++++++++++++
 2 files changed, 553 insertions(+)

diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfo.td b/llvm/lib/Target/RISCV/RISCVInstrInfo.td
index 06a9043..78fd09f 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfo.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfo.td
@@ -1278,6 +1278,13 @@ def : Pat<(setgt GPR:$rs1, simm12_minus1_nonzero:$imm),
 def : Pat<(setugt GPR:$rs1, simm12_minus1_nonzero:$imm),
           (XORI (SLTIU GPR:$rs1, (ImmPlus1 simm12_minus1_nonzero:$imm)), 1)>;
 
+// If negating a pattern that requires an XORI above, we can fold the XORI with
+// the NEG. The XORI is equivalent to 1-X and negating gives X-1.
+def : Pat<(ineg (setuge GPR:$rs1, GPR:$rs2)), (ADDI (SLTU GPR:$rs1, GPR:$rs2), -1)>;
+def : Pat<(ineg (setule GPR:$rs1, GPR:$rs2)), (ADDI (SLTU GPR:$rs2, GPR:$rs1), -1)>;
+def : Pat<(ineg (setge GPR:$rs1, GPR:$rs2)), (ADDI (SLT GPR:$rs1, GPR:$rs2), -1)>;
+def : Pat<(ineg (setle GPR:$rs1, GPR:$rs2)), (ADDI (SLT GPR:$rs2, GPR:$rs1), -1)>;
+
 def IntCCtoRISCVCC : SDNodeXForm<riscv_selectcc, [{
   ISD::CondCode CC = cast<CondCodeSDNode>(N->getOperand(2))->get();
   RISCVCC::CondCode BrCC = getRISCVCCForIntCC(CC);
diff --git a/llvm/test/CodeGen/RISCV/select-const.ll b/llvm/test/CodeGen/RISCV/select-const.ll
index e3739fc..927d85d 100644
--- a/llvm/test/CodeGen/RISCV/select-const.ll
+++ b/llvm/test/CodeGen/RISCV/select-const.ll
@@ -321,3 +321,549 @@ define float @select_const_fp(i1 zeroext %a) nounwind {
   %1 = select i1 %a, float 3.0, float 4.0
   ret float %1
 }
+
+define signext i32 @select_eq_zero_negone(i32 signext %a, i32 signext %b) nounwind {
+; RV32I-LABEL: select_eq_zero_negone:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    xor a0, a0, a1
+; RV32I-NEXT:    seqz a0, a0
+; RV32I-NEXT:    neg a0, a0
+; RV32I-NEXT:    ret
+;
+; RV32IF-LABEL: select_eq_zero_negone:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    xor a0, a0, a1
+; RV32IF-NEXT:    seqz a0, a0
+; RV32IF-NEXT:    neg a0, a0
+; RV32IF-NEXT:    ret
+;
+; RV32IBT-LABEL: select_eq_zero_negone:
+; RV32IBT:       # %bb.0:
+; RV32IBT-NEXT:    xor a0, a0, a1
+; RV32IBT-NEXT:    seqz a0, a0
+; RV32IBT-NEXT:    neg a0, a0
+; RV32IBT-NEXT:    ret
+;
+; RV32IFBT-LABEL: select_eq_zero_negone:
+; RV32IFBT:       # %bb.0:
+; RV32IFBT-NEXT:    xor a0, a0, a1
+; RV32IFBT-NEXT:    seqz a0, a0
+; RV32IFBT-NEXT:    neg a0, a0
+; RV32IFBT-NEXT:    ret
+;
+; RV64I-LABEL: select_eq_zero_negone:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    xor a0, a0, a1
+; RV64I-NEXT:    seqz a0, a0
+; RV64I-NEXT:    neg a0, a0
+; RV64I-NEXT:    ret
+;
+; RV64IFD-LABEL: select_eq_zero_negone:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    xor a0, a0, a1
+; RV64IFD-NEXT:    seqz a0, a0
+; RV64IFD-NEXT:    neg a0, a0
+; RV64IFD-NEXT:    ret
+;
+; RV64IBT-LABEL: select_eq_zero_negone:
+; RV64IBT:       # %bb.0:
+; RV64IBT-NEXT:    xor a0, a0, a1
+; RV64IBT-NEXT:    seqz a0, a0
+; RV64IBT-NEXT:    neg a0, a0
+; RV64IBT-NEXT:    ret
+;
+; RV64IFDBT-LABEL: select_eq_zero_negone:
+; RV64IFDBT:       # %bb.0:
+; RV64IFDBT-NEXT:    xor a0, a0, a1
+; RV64IFDBT-NEXT:    seqz a0, a0
+; RV64IFDBT-NEXT:    neg a0, a0
+; RV64IFDBT-NEXT:    ret
+  %1 = icmp eq i32 %a, %b
+  %2 = select i1 %1, i32 -1, i32 0
+  ret i32 %2
+}
+
+define signext i32 @select_ne_zero_negone(i32 signext %a, i32 signext %b) nounwind {
+; RV32I-LABEL: select_ne_zero_negone:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    xor a0, a0, a1
+; RV32I-NEXT:    snez a0, a0
+; RV32I-NEXT:    neg a0, a0
+; RV32I-NEXT:    ret
+;
+; RV32IF-LABEL: select_ne_zero_negone:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    xor a0, a0, a1
+; RV32IF-NEXT:    snez a0, a0
+; RV32IF-NEXT:    neg a0, a0
+; RV32IF-NEXT:    ret
+;
+; RV32IBT-LABEL: select_ne_zero_negone:
+; RV32IBT:       # %bb.0:
+; RV32IBT-NEXT:    xor a0, a0, a1
+; RV32IBT-NEXT:    snez a0, a0
+; RV32IBT-NEXT:    neg a0, a0
+; RV32IBT-NEXT:    ret
+;
+; RV32IFBT-LABEL: select_ne_zero_negone:
+; RV32IFBT:       # %bb.0:
+; RV32IFBT-NEXT:    xor a0, a0, a1
+; RV32IFBT-NEXT:    snez a0, a0
+; RV32IFBT-NEXT:    neg a0, a0
+; RV32IFBT-NEXT:    ret
+;
+; RV64I-LABEL: select_ne_zero_negone:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    xor a0, a0, a1
+; RV64I-NEXT:    snez a0, a0
+; RV64I-NEXT:    neg a0, a0
+; RV64I-NEXT:    ret
+;
+; RV64IFD-LABEL: select_ne_zero_negone:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    xor a0, a0, a1
+; RV64IFD-NEXT:    snez a0, a0
+; RV64IFD-NEXT:    neg a0, a0
+; RV64IFD-NEXT:    ret
+;
+; RV64IBT-LABEL: select_ne_zero_negone:
+; RV64IBT:       # %bb.0:
+; RV64IBT-NEXT:    xor a0, a0, a1
+; RV64IBT-NEXT:    snez a0, a0
+; RV64IBT-NEXT:    neg a0, a0
+; RV64IBT-NEXT:    ret
+;
+; RV64IFDBT-LABEL: select_ne_zero_negone:
+; RV64IFDBT:       # %bb.0:
+; RV64IFDBT-NEXT:    xor a0, a0, a1
+; RV64IFDBT-NEXT:    snez a0, a0
+; RV64IFDBT-NEXT:    neg a0, a0
+; RV64IFDBT-NEXT:    ret
+  %1 = icmp ne i32 %a, %b
+  %2 = select i1 %1, i32 -1, i32 0
+  ret i32 %2
+}
+
+define signext i32 @select_sgt_zero_negone(i32 signext %a, i32 signext %b) nounwind {
+; RV32I-LABEL: select_sgt_zero_negone:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    slt a0, a1, a0
+; RV32I-NEXT:    neg a0, a0
+; RV32I-NEXT:    ret
+;
+; RV32IF-LABEL: select_sgt_zero_negone:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    slt a0, a1, a0
+; RV32IF-NEXT:    neg a0, a0
+; RV32IF-NEXT:    ret
+;
+; RV32IBT-LABEL: select_sgt_zero_negone:
+; RV32IBT:       # %bb.0:
+; RV32IBT-NEXT:    slt a0, a1, a0
+; RV32IBT-NEXT:    neg a0, a0
+; RV32IBT-NEXT:    ret
+;
+; RV32IFBT-LABEL: select_sgt_zero_negone:
+; RV32IFBT:       # %bb.0:
+; RV32IFBT-NEXT:    slt a0, a1, a0
+; RV32IFBT-NEXT:    neg a0, a0
+; RV32IFBT-NEXT:    ret
+;
+; RV64I-LABEL: select_sgt_zero_negone:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    slt a0, a1, a0
+; RV64I-NEXT:    neg a0, a0
+; RV64I-NEXT:    ret
+;
+; RV64IFD-LABEL: select_sgt_zero_negone:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    slt a0, a1, a0
+; RV64IFD-NEXT:    neg a0, a0
+; RV64IFD-NEXT:    ret
+;
+; RV64IBT-LABEL: select_sgt_zero_negone:
+; RV64IBT:       # %bb.0:
+; RV64IBT-NEXT:    slt a0, a1, a0
+; RV64IBT-NEXT:    neg a0, a0
+; RV64IBT-NEXT:    ret
+;
+; RV64IFDBT-LABEL: select_sgt_zero_negone:
+; RV64IFDBT:       # %bb.0:
+; RV64IFDBT-NEXT:    slt a0, a1, a0
+; RV64IFDBT-NEXT:    neg a0, a0
+; RV64IFDBT-NEXT:    ret
+  %1 = icmp sgt i32 %a, %b
+  %2 = select i1 %1, i32 -1, i32 0
+  ret i32 %2
+}
+
+define signext i32 @select_slt_zero_negone(i32 signext %a, i32 signext %b) nounwind {
+; RV32I-LABEL: select_slt_zero_negone:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    slt a0, a0, a1
+; RV32I-NEXT:    neg a0, a0
+; RV32I-NEXT:    ret
+;
+; RV32IF-LABEL: select_slt_zero_negone:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    slt a0, a0, a1
+; RV32IF-NEXT:    neg a0, a0
+; RV32IF-NEXT:    ret
+;
+; RV32IBT-LABEL: select_slt_zero_negone:
+; RV32IBT:       # %bb.0:
+; RV32IBT-NEXT:    slt a0, a0, a1
+; RV32IBT-NEXT:    neg a0, a0
+; RV32IBT-NEXT:    ret
+;
+; RV32IFBT-LABEL: select_slt_zero_negone:
+; RV32IFBT:       # %bb.0:
+; RV32IFBT-NEXT:    slt a0, a0, a1
+; RV32IFBT-NEXT:    neg a0, a0
+; RV32IFBT-NEXT:    ret
+;
+; RV64I-LABEL: select_slt_zero_negone:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    slt a0, a0, a1
+; RV64I-NEXT:    neg a0, a0
+; RV64I-NEXT:    ret
+;
+; RV64IFD-LABEL: select_slt_zero_negone:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    slt a0, a0, a1
+; RV64IFD-NEXT:    neg a0, a0
+; RV64IFD-NEXT:    ret
+;
+; RV64IBT-LABEL: select_slt_zero_negone:
+; RV64IBT:       # %bb.0:
+; RV64IBT-NEXT:    slt a0, a0, a1
+; RV64IBT-NEXT:    neg a0, a0
+; RV64IBT-NEXT:    ret
+;
+; RV64IFDBT-LABEL: select_slt_zero_negone:
+; RV64IFDBT:       # %bb.0:
+; RV64IFDBT-NEXT:    slt a0, a0, a1
+; RV64IFDBT-NEXT:    neg a0, a0
+; RV64IFDBT-NEXT:    ret
+  %1 = icmp slt i32 %a, %b
+  %2 = select i1 %1, i32 -1, i32 0
+  ret i32 %2
+}
+
+define signext i32 @select_sge_zero_negone(i32 signext %a, i32 signext %b) nounwind {
+; RV32I-LABEL: select_sge_zero_negone:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    slt a0, a0, a1
+; RV32I-NEXT:    addi a0, a0, -1
+; RV32I-NEXT:    ret
+;
+; RV32IF-LABEL: select_sge_zero_negone:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    slt a0, a0, a1
+; RV32IF-NEXT:    addi a0, a0, -1
+; RV32IF-NEXT:    ret
+;
+; RV32IBT-LABEL: select_sge_zero_negone:
+; RV32IBT:       # %bb.0:
+; RV32IBT-NEXT:    slt a0, a0, a1
+; RV32IBT-NEXT:    addi a0, a0, -1
+; RV32IBT-NEXT:    ret
+;
+; RV32IFBT-LABEL: select_sge_zero_negone:
+; RV32IFBT:       # %bb.0:
+; RV32IFBT-NEXT:    slt a0, a0, a1
+; RV32IFBT-NEXT:    addi a0, a0, -1
+; RV32IFBT-NEXT:    ret
+;
+; RV64I-LABEL: select_sge_zero_negone:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    slt a0, a0, a1
+; RV64I-NEXT:    addi a0, a0, -1
+; RV64I-NEXT:    ret
+;
+; RV64IFD-LABEL: select_sge_zero_negone:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    slt a0, a0, a1
+; RV64IFD-NEXT:    addi a0, a0, -1
+; RV64IFD-NEXT:    ret
+;
+; RV64IBT-LABEL: select_sge_zero_negone:
+; RV64IBT:       # %bb.0:
+; RV64IBT-NEXT:    slt a0, a0, a1
+; RV64IBT-NEXT:    addi a0, a0, -1
+; RV64IBT-NEXT:    ret
+;
+; RV64IFDBT-LABEL: select_sge_zero_negone:
+; RV64IFDBT:       # %bb.0:
+; RV64IFDBT-NEXT:    slt a0, a0, a1
+; RV64IFDBT-NEXT:    addi a0, a0, -1
+; RV64IFDBT-NEXT:    ret
+  %1 = icmp sge i32 %a, %b
+  %2 = select i1 %1, i32 -1, i32 0
+  ret i32 %2
+}
+
+define signext i32 @select_sle_zero_negone(i32 signext %a, i32 signext %b) nounwind {
+; RV32I-LABEL: select_sle_zero_negone:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    slt a0, a1, a0
+; RV32I-NEXT:    addi a0, a0, -1
+; RV32I-NEXT:    ret
+;
+; RV32IF-LABEL: select_sle_zero_negone:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    slt a0, a1, a0
+; RV32IF-NEXT:    addi a0, a0, -1
+; RV32IF-NEXT:    ret
+;
+; RV32IBT-LABEL: select_sle_zero_negone:
+; RV32IBT:       # %bb.0:
+; RV32IBT-NEXT:    slt a0, a1, a0
+; RV32IBT-NEXT:    addi a0, a0, -1
+; RV32IBT-NEXT:    ret
+;
+; RV32IFBT-LABEL: select_sle_zero_negone:
+; RV32IFBT:       # %bb.0:
+; RV32IFBT-NEXT:    slt a0, a1, a0
+; RV32IFBT-NEXT:    addi a0, a0, -1
+; RV32IFBT-NEXT:    ret
+;
+; RV64I-LABEL: select_sle_zero_negone:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    slt a0, a1, a0
+; RV64I-NEXT:    addi a0, a0, -1
+; RV64I-NEXT:    ret
+;
+; RV64IFD-LABEL: select_sle_zero_negone:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    slt a0, a1, a0
+; RV64IFD-NEXT:    addi a0, a0, -1
+; RV64IFD-NEXT:    ret
+;
+; RV64IBT-LABEL: select_sle_zero_negone:
+; RV64IBT:       # %bb.0:
+; RV64IBT-NEXT:    slt a0, a1, a0
+; RV64IBT-NEXT:    addi a0, a0, -1
+; RV64IBT-NEXT:    ret
+;
+; RV64IFDBT-LABEL: select_sle_zero_negone:
+; RV64IFDBT:       # %bb.0:
+; RV64IFDBT-NEXT:    slt a0, a1, a0
+; RV64IFDBT-NEXT:    addi a0, a0, -1
+; RV64IFDBT-NEXT:    ret
+  %1 = icmp sle i32 %a, %b
+  %2 = select i1 %1, i32 -1, i32 0
+  ret i32 %2
+}
+
+define signext i32 @select_ugt_zero_negone(i32 signext %a, i32 signext %b) nounwind {
+; RV32I-LABEL: select_ugt_zero_negone:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    sltu a0, a1, a0
+; RV32I-NEXT:    neg a0, a0
+; RV32I-NEXT:    ret
+;
+; RV32IF-LABEL: select_ugt_zero_negone:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    sltu a0, a1, a0
+; RV32IF-NEXT:    neg a0, a0
+; RV32IF-NEXT:    ret
+;
+; RV32IBT-LABEL: select_ugt_zero_negone:
+; RV32IBT:       # %bb.0:
+; RV32IBT-NEXT:    sltu a0, a1, a0
+; RV32IBT-NEXT:    neg a0, a0
+; RV32IBT-NEXT:    ret
+;
+; RV32IFBT-LABEL: select_ugt_zero_negone:
+; RV32IFBT:       # %bb.0:
+; RV32IFBT-NEXT:    sltu a0, a1, a0
+; RV32IFBT-NEXT:    neg a0, a0
+; RV32IFBT-NEXT:    ret
+;
+; RV64I-LABEL: select_ugt_zero_negone:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    sltu a0, a1, a0
+; RV64I-NEXT:    neg a0, a0
+; RV64I-NEXT:    ret
+;
+; RV64IFD-LABEL: select_ugt_zero_negone:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    sltu a0, a1, a0
+; RV64IFD-NEXT:    neg a0, a0
+; RV64IFD-NEXT:    ret
+;
+; RV64IBT-LABEL: select_ugt_zero_negone:
+; RV64IBT:       # %bb.0:
+; RV64IBT-NEXT:    sltu a0, a1, a0
+; RV64IBT-NEXT:    neg a0, a0
+; RV64IBT-NEXT:    ret
+;
+; RV64IFDBT-LABEL: select_ugt_zero_negone:
+; RV64IFDBT:       # %bb.0:
+; RV64IFDBT-NEXT:    sltu a0, a1, a0
+; RV64IFDBT-NEXT:    neg a0, a0
+; RV64IFDBT-NEXT:    ret
+  %1 = icmp ugt i32 %a, %b
+  %2 = select i1 %1, i32 -1, i32 0
+  ret i32 %2
+}
+
+define signext i32 @select_ult_zero_negone(i32 signext %a, i32 signext %b) nounwind {
+; RV32I-LABEL: select_ult_zero_negone:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    sltu a0, a0, a1
+; RV32I-NEXT:    neg a0, a0
+; RV32I-NEXT:    ret
+;
+; RV32IF-LABEL: select_ult_zero_negone:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    sltu a0, a0, a1
+; RV32IF-NEXT:    neg a0, a0
+; RV32IF-NEXT:    ret
+;
+; RV32IBT-LABEL: select_ult_zero_negone:
+; RV32IBT:       # %bb.0:
+; RV32IBT-NEXT:    sltu a0, a0, a1
+; RV32IBT-NEXT:    neg a0, a0
+; RV32IBT-NEXT:    ret
+;
+; RV32IFBT-LABEL: select_ult_zero_negone:
+; RV32IFBT:       # %bb.0:
+; RV32IFBT-NEXT:    sltu a0, a0, a1
+; RV32IFBT-NEXT:    neg a0, a0
+; RV32IFBT-NEXT:    ret
+;
+; RV64I-LABEL: select_ult_zero_negone:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    sltu a0, a0, a1
+; RV64I-NEXT:    neg a0, a0
+; RV64I-NEXT:    ret
+;
+; RV64IFD-LABEL: select_ult_zero_negone:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    sltu a0, a0, a1
+; RV64IFD-NEXT:    neg a0, a0
+; RV64IFD-NEXT:    ret
+;
+; RV64IBT-LABEL: select_ult_zero_negone:
+; RV64IBT:       # %bb.0:
+; RV64IBT-NEXT:    sltu a0, a0, a1
+; RV64IBT-NEXT:    neg a0, a0
+; RV64IBT-NEXT:    ret
+;
+; RV64IFDBT-LABEL: select_ult_zero_negone:
+; RV64IFDBT:       # %bb.0:
+; RV64IFDBT-NEXT:    sltu a0, a0, a1
+; RV64IFDBT-NEXT:    neg a0, a0
+; RV64IFDBT-NEXT:    ret
+  %1 = icmp ult i32 %a, %b
+  %2 = select i1 %1, i32 -1, i32 0
+  ret i32 %2
+}
+
+define signext i32 @select_uge_zero_negone(i32 signext %a, i32 signext %b) nounwind {
+; RV32I-LABEL: select_uge_zero_negone:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    sltu a0, a0, a1
+; RV32I-NEXT:    addi a0, a0, -1
+; RV32I-NEXT:    ret
+;
+; RV32IF-LABEL: select_uge_zero_negone:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    sltu a0, a0, a1
+; RV32IF-NEXT:    addi a0, a0, -1
+; RV32IF-NEXT:    ret
+;
+; RV32IBT-LABEL: select_uge_zero_negone:
+; RV32IBT:       # %bb.0:
+; RV32IBT-NEXT:    sltu a0, a0, a1
+; RV32IBT-NEXT:    addi a0, a0, -1
+; RV32IBT-NEXT:    ret
+;
+; RV32IFBT-LABEL: select_uge_zero_negone:
+; RV32IFBT:       # %bb.0:
+; RV32IFBT-NEXT:    sltu a0, a0, a1
+; RV32IFBT-NEXT:    addi a0, a0, -1
+; RV32IFBT-NEXT:    ret
+;
+; RV64I-LABEL: select_uge_zero_negone:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    sltu a0, a0, a1
+; RV64I-NEXT:    addi a0, a0, -1
+; RV64I-NEXT:    ret
+;
+; RV64IFD-LABEL: select_uge_zero_negone:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    sltu a0, a0, a1
+; RV64IFD-NEXT:    addi a0, a0, -1
+; RV64IFD-NEXT:    ret
+;
+; RV64IBT-LABEL: select_uge_zero_negone:
+; RV64IBT:       # %bb.0:
+; RV64IBT-NEXT:    sltu a0, a0, a1
+; RV64IBT-NEXT:    addi a0, a0, -1
+; RV64IBT-NEXT:    ret
+;
+; RV64IFDBT-LABEL: select_uge_zero_negone:
+; RV64IFDBT:       # %bb.0:
+; RV64IFDBT-NEXT:    sltu a0, a0, a1
+; RV64IFDBT-NEXT:    addi a0, a0, -1
+; RV64IFDBT-NEXT:    ret
+  %1 = icmp uge i32 %a, %b
+  %2 = select i1 %1, i32 -1, i32 0
+  ret i32 %2
+}
+
+define signext i32 @select_ule_zero_negone(i32 signext %a, i32 signext %b) nounwind {
+; RV32I-LABEL: select_ule_zero_negone:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    sltu a0, a1, a0
+; RV32I-NEXT:    addi a0, a0, -1
+; RV32I-NEXT:    ret
+;
+; RV32IF-LABEL: select_ule_zero_negone:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    sltu a0, a1, a0
+; RV32IF-NEXT:    addi a0, a0, -1
+; RV32IF-NEXT:    ret
+;
+; RV32IBT-LABEL: select_ule_zero_negone:
+; RV32IBT:       # %bb.0:
+; RV32IBT-NEXT:    sltu a0, a1, a0
+; RV32IBT-NEXT:    addi a0, a0, -1
+; RV32IBT-NEXT:    ret
+;
+; RV32IFBT-LABEL: select_ule_zero_negone:
+; RV32IFBT:       # %bb.0:
+; RV32IFBT-NEXT:    sltu a0, a1, a0
+; RV32IFBT-NEXT:    addi a0, a0, -1
+; RV32IFBT-NEXT:    ret
+;
+; RV64I-LABEL: select_ule_zero_negone:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    sltu a0, a1, a0
+; RV64I-NEXT:    addi a0, a0, -1
+; RV64I-NEXT:    ret
+;
+; RV64IFD-LABEL: select_ule_zero_negone:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    sltu a0, a1, a0
+; RV64IFD-NEXT:    addi a0, a0, -1
+; RV64IFD-NEXT:    ret
+;
+; RV64IBT-LABEL: select_ule_zero_negone:
+; RV64IBT:       # %bb.0:
+; RV64IBT-NEXT:    sltu a0, a1, a0
+; RV64IBT-NEXT:    addi a0, a0, -1
+; RV64IBT-NEXT:    ret
+;
+; RV64IFDBT-LABEL: select_ule_zero_negone:
+; RV64IFDBT:       # %bb.0:
+; RV64IFDBT-NEXT:    sltu a0, a1, a0
+; RV64IFDBT-NEXT:    addi a0, a0, -1
+; RV64IFDBT-NEXT:    ret
+  %1 = icmp ule i32 %a, %b
+  %2 = select i1 %1, i32 -1, i32 0
+  ret i32 %2
+}
-- 
2.7.4