[llvm] [RISCV]: Implemented softening of `FCANONICALIZE` (PR #169234)
Kevin Per via llvm-commits
llvm-commits at lists.llvm.org
Mon Nov 24 03:05:38 PST 2025
https://github.com/kper updated https://github.com/llvm/llvm-project/pull/169234
>From a7af5b21388d4941f9008582e081ae8b1312eb63 Mon Sep 17 00:00:00 2001
From: Kevin Per <kevin.per at protonmail.com>
Date: Sun, 23 Nov 2025 21:02:45 +0100
Subject: [PATCH 1/2] [RISCV]: Implemented softening of `FCANONICALIZE`
---
.../SelectionDAG/LegalizeFloatTypes.cpp | 8 +++
llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h | 1 +
llvm/test/CodeGen/RISCV/fcanonicalize.ll | 64 +++++++++++++++++++
3 files changed, 73 insertions(+)
create mode 100644 llvm/test/CodeGen/RISCV/fcanonicalize.ll
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
index 383a025a4d916..d405295229203 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
@@ -70,6 +70,8 @@ void DAGTypeLegalizer::SoftenFloatResult(SDNode *N, unsigned ResNo) {
case ISD::EXTRACT_VECTOR_ELT:
R = SoftenFloatRes_EXTRACT_VECTOR_ELT(N, ResNo); break;
case ISD::FABS: R = SoftenFloatRes_FABS(N); break;
+ case ISD::FCANONICALIZE:
+ R = SoftenFloatRes_FCANONICALIZE(N); break;
case ISD::STRICT_FMINNUM:
case ISD::FMINNUM: R = SoftenFloatRes_FMINNUM(N); break;
case ISD::STRICT_FMAXNUM:
@@ -311,6 +313,12 @@ SDValue DAGTypeLegalizer::SoftenFloatRes_FABS(SDNode *N) {
return DAG.getNode(ISD::AND, SDLoc(N), NVT, Op, Mask);
}
+SDValue DAGTypeLegalizer::SoftenFloatRes_FCANONICALIZE(SDNode *N) {
+ return SoftenFloatRes_Unary(
+ N, GetFPLibCall(N->getValueType(0), RTLIB::FMIN_F32, RTLIB::FMIN_F64,
+ RTLIB::FMIN_F80, RTLIB::FMIN_F128, RTLIB::FMIN_PPCF128));
+}
+
SDValue DAGTypeLegalizer::SoftenFloatRes_FMINNUM(SDNode *N) {
if (SDValue SelCC = TLI.createSelectForFMINNUM_FMAXNUM(N, DAG))
return SoftenFloatRes_SELECT_CC(SelCC.getNode());
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
index ede522eff6df3..c90cb7bc88f57 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
@@ -585,6 +585,7 @@ class LLVM_LIBRARY_VISIBILITY DAGTypeLegalizer {
SDValue SoftenFloatRes_FASIN(SDNode *N);
SDValue SoftenFloatRes_FATAN(SDNode *N);
SDValue SoftenFloatRes_FATAN2(SDNode *N);
+ SDValue SoftenFloatRes_FCANONICALIZE(SDNode *N);
SDValue SoftenFloatRes_FMINNUM(SDNode *N);
SDValue SoftenFloatRes_FMAXNUM(SDNode *N);
SDValue SoftenFloatRes_FMINIMUMNUM(SDNode *N);
diff --git a/llvm/test/CodeGen/RISCV/fcanonicalize.ll b/llvm/test/CodeGen/RISCV/fcanonicalize.ll
new file mode 100644
index 0000000000000..334657bdf92de
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/fcanonicalize.ll
@@ -0,0 +1,64 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=riscv64 < %s | FileCheck -check-prefix=RV64I %s
+; RUN: llc -mtriple=riscv64 -mattr=+d < %s | FileCheck -check-prefix=RV64D %s
+
+define double @max(double, double) unnamed_addr #0 {
+; RV64I-LABEL: max:
+; RV64I: # %bb.0: # %start
+; RV64I-NEXT: addi sp, sp, -32
+; RV64I-NEXT: .cfi_def_cfa_offset 32
+; RV64I-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
+; RV64I-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
+; RV64I-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT: sd s2, 0(sp) # 8-byte Folded Spill
+; RV64I-NEXT: .cfi_offset ra, -8
+; RV64I-NEXT: .cfi_offset s0, -16
+; RV64I-NEXT: .cfi_offset s1, -24
+; RV64I-NEXT: .cfi_offset s2, -32
+; RV64I-NEXT: mv s0, a1
+; RV64I-NEXT: mv s1, a0
+; RV64I-NEXT: call __ltdf2
+; RV64I-NEXT: srli s2, a0, 63
+; RV64I-NEXT: mv a0, s1
+; RV64I-NEXT: mv a1, s1
+; RV64I-NEXT: call __unorddf2
+; RV64I-NEXT: snez a0, a0
+; RV64I-NEXT: or a0, a0, s2
+; RV64I-NEXT: bnez a0, .LBB0_2
+; RV64I-NEXT: # %bb.1: # %start
+; RV64I-NEXT: mv s0, s1
+; RV64I-NEXT: .LBB0_2: # %start
+; RV64I-NEXT: mv a0, s0
+; RV64I-NEXT: call fmin
+; RV64I-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
+; RV64I-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
+; RV64I-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT: ld s2, 0(sp) # 8-byte Folded Reload
+; RV64I-NEXT: .cfi_restore ra
+; RV64I-NEXT: .cfi_restore s0
+; RV64I-NEXT: .cfi_restore s1
+; RV64I-NEXT: .cfi_restore s2
+; RV64I-NEXT: addi sp, sp, 32
+; RV64I-NEXT: .cfi_def_cfa_offset 0
+; RV64I-NEXT: ret
+;
+; RV64D-LABEL: max:
+; RV64D: # %bb.0: # %start
+; RV64D-NEXT: flt.d a0, fa0, fa1
+; RV64D-NEXT: feq.d a1, fa0, fa0
+; RV64D-NEXT: xori a1, a1, 1
+; RV64D-NEXT: or a0, a1, a0
+; RV64D-NEXT: bnez a0, .LBB0_2
+; RV64D-NEXT: # %bb.1: # %start
+; RV64D-NEXT: fmv.d fa1, fa0
+; RV64D-NEXT: .LBB0_2: # %start
+; RV64D-NEXT: fmin.d fa0, fa1, fa1
+; RV64D-NEXT: ret
+start:
+ %2 = fcmp olt double %0, %1
+ %3 = fcmp uno double %0, 0.000000e+00
+ %or.cond.i.i = or i1 %3, %2
+ %4 = select i1 %or.cond.i.i, double %1, double %0
+ %5 = tail call double @llvm.canonicalize.f64(double %4) #2
+ ret double %5
+}
>From 8aa740662963a377bf0c83845d6c6523832b44a7 Mon Sep 17 00:00:00 2001
From: Kevin Per <kevin.per at protonmail.com>
Date: Mon, 24 Nov 2025 11:53:17 +0100
Subject: [PATCH 2/2] [RISCV]: Map to FMINIMUMNUM
---
.../SelectionDAG/LegalizeFloatTypes.cpp | 11 +-
llvm/test/CodeGen/RISCV/fcanonicalize.ll | 64 -
llvm/test/CodeGen/RISCV/fp-fcanonicalize.ll | 3657 ++++++++++++++++-
3 files changed, 3606 insertions(+), 126 deletions(-)
delete mode 100644 llvm/test/CodeGen/RISCV/fcanonicalize.ll
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
index d405295229203..a162bfdc52189 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
@@ -314,9 +314,14 @@ SDValue DAGTypeLegalizer::SoftenFloatRes_FABS(SDNode *N) {
}
SDValue DAGTypeLegalizer::SoftenFloatRes_FCANONICALIZE(SDNode *N) {
- return SoftenFloatRes_Unary(
- N, GetFPLibCall(N->getValueType(0), RTLIB::FMIN_F32, RTLIB::FMIN_F64,
- RTLIB::FMIN_F80, RTLIB::FMIN_F128, RTLIB::FMIN_PPCF128));
+ SDLoc dl(N);
+ auto Node = DAG.getNode(ISD::FMINIMUMNUM, dl, N->getValueType(0),
+ N->getOperand(0), N->getOperand(0));
+ return SoftenFloatRes_Binary(
+ Node.getNode(),
+ GetFPLibCall(N->getValueType(0), RTLIB::FMINIMUM_NUM_F32,
+ RTLIB::FMINIMUM_NUM_F64, RTLIB::FMINIMUM_NUM_F80,
+ RTLIB::FMINIMUM_NUM_F128, RTLIB::FMINIMUM_NUM_PPCF128));
}
SDValue DAGTypeLegalizer::SoftenFloatRes_FMINNUM(SDNode *N) {
diff --git a/llvm/test/CodeGen/RISCV/fcanonicalize.ll b/llvm/test/CodeGen/RISCV/fcanonicalize.ll
deleted file mode 100644
index 334657bdf92de..0000000000000
--- a/llvm/test/CodeGen/RISCV/fcanonicalize.ll
+++ /dev/null
@@ -1,64 +0,0 @@
-; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
-; RUN: llc -mtriple=riscv64 < %s | FileCheck -check-prefix=RV64I %s
-; RUN: llc -mtriple=riscv64 -mattr=+d < %s | FileCheck -check-prefix=RV64D %s
-
-define double @max(double, double) unnamed_addr #0 {
-; RV64I-LABEL: max:
-; RV64I: # %bb.0: # %start
-; RV64I-NEXT: addi sp, sp, -32
-; RV64I-NEXT: .cfi_def_cfa_offset 32
-; RV64I-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
-; RV64I-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
-; RV64I-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
-; RV64I-NEXT: sd s2, 0(sp) # 8-byte Folded Spill
-; RV64I-NEXT: .cfi_offset ra, -8
-; RV64I-NEXT: .cfi_offset s0, -16
-; RV64I-NEXT: .cfi_offset s1, -24
-; RV64I-NEXT: .cfi_offset s2, -32
-; RV64I-NEXT: mv s0, a1
-; RV64I-NEXT: mv s1, a0
-; RV64I-NEXT: call __ltdf2
-; RV64I-NEXT: srli s2, a0, 63
-; RV64I-NEXT: mv a0, s1
-; RV64I-NEXT: mv a1, s1
-; RV64I-NEXT: call __unorddf2
-; RV64I-NEXT: snez a0, a0
-; RV64I-NEXT: or a0, a0, s2
-; RV64I-NEXT: bnez a0, .LBB0_2
-; RV64I-NEXT: # %bb.1: # %start
-; RV64I-NEXT: mv s0, s1
-; RV64I-NEXT: .LBB0_2: # %start
-; RV64I-NEXT: mv a0, s0
-; RV64I-NEXT: call fmin
-; RV64I-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
-; RV64I-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
-; RV64I-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
-; RV64I-NEXT: ld s2, 0(sp) # 8-byte Folded Reload
-; RV64I-NEXT: .cfi_restore ra
-; RV64I-NEXT: .cfi_restore s0
-; RV64I-NEXT: .cfi_restore s1
-; RV64I-NEXT: .cfi_restore s2
-; RV64I-NEXT: addi sp, sp, 32
-; RV64I-NEXT: .cfi_def_cfa_offset 0
-; RV64I-NEXT: ret
-;
-; RV64D-LABEL: max:
-; RV64D: # %bb.0: # %start
-; RV64D-NEXT: flt.d a0, fa0, fa1
-; RV64D-NEXT: feq.d a1, fa0, fa0
-; RV64D-NEXT: xori a1, a1, 1
-; RV64D-NEXT: or a0, a1, a0
-; RV64D-NEXT: bnez a0, .LBB0_2
-; RV64D-NEXT: # %bb.1: # %start
-; RV64D-NEXT: fmv.d fa1, fa0
-; RV64D-NEXT: .LBB0_2: # %start
-; RV64D-NEXT: fmin.d fa0, fa1, fa1
-; RV64D-NEXT: ret
-start:
- %2 = fcmp olt double %0, %1
- %3 = fcmp uno double %0, 0.000000e+00
- %or.cond.i.i = or i1 %3, %2
- %4 = select i1 %or.cond.i.i, double %1, double %0
- %5 = tail call double @llvm.canonicalize.f64(double %4) #2
- ret double %5
-}
diff --git a/llvm/test/CodeGen/RISCV/fp-fcanonicalize.ll b/llvm/test/CodeGen/RISCV/fp-fcanonicalize.ll
index e9b771a0698de..5b0b77156ebb0 100644
--- a/llvm/test/CodeGen/RISCV/fp-fcanonicalize.ll
+++ b/llvm/test/CodeGen/RISCV/fp-fcanonicalize.ll
@@ -1,7 +1,8 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; FIXME: @llvm.canonicalize doesn't support soft-float abi yet.
+; RUN: llc --mtriple=riscv64 < %s | FileCheck %s --check-prefixes=CHECK,CHECK-SOFT-RV64
; RUN: llc --mtriple=riscv64 --mattr=+d,+zfh < %s | FileCheck %s --check-prefixes=CHECK,CHECK-FP16-RV64
; RUN: llc --mtriple=riscv64 --mattr=+d,-zfh < %s | FileCheck %s --check-prefixes=CHECK,CHECK-NOFP16-RV64
+; RUN: llc --mtriple=riscv32 < %s | FileCheck %s --check-prefixes=CHECK,CHECK-SOFT-RV32
; RUN: llc --mtriple=riscv32 --mattr=+d,+zfh < %s | FileCheck %s --check-prefixes=CHECK,CHECK-FP16-RV32
; RUN: llc --mtriple=riscv32 --mattr=+d,-zfh < %s | FileCheck %s --check-prefixes=CHECK,CHECK-NOFP16-RV32
@@ -10,6 +11,42 @@ declare float @llvm.fcanonicalize.f32(float)
declare double @llvm.fcanonicalize.f64(double)
define half @fcanonicalize_f16(half %x) {
+; RV64-SOFT-LABEL: fcanonicalize_f16:
+; RV64-SOFT: # %bb.0:
+; RV64-SOFT-NEXT: addi sp, sp, -16
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 16
+; RV64-SOFT-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: .cfi_offset ra, -8
+; RV64-SOFT-NEXT: slli a0, a0, 48
+; RV64-SOFT-NEXT: srli a0, a0, 48
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: .cfi_restore ra
+; RV64-SOFT-NEXT: addi sp, sp, 16
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV64-SOFT-NEXT: ret
+;
+; CHECK-SOFT-RV64-LABEL: fcanonicalize_f16:
+; CHECK-SOFT-RV64: # %bb.0:
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, -16
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 16
+; CHECK-SOFT-RV64-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: .cfi_offset ra, -8
+; CHECK-SOFT-RV64-NEXT: slli a0, a0, 48
+; CHECK-SOFT-RV64-NEXT: srli a0, a0, 48
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, 16
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV64-NEXT: ret
+;
; CHECK-FP16-RV64-LABEL: fcanonicalize_f16:
; CHECK-FP16-RV64: # %bb.0:
; CHECK-FP16-RV64-NEXT: fmin.h fa0, fa0, fa0
@@ -34,6 +71,24 @@ define half @fcanonicalize_f16(half %x) {
; CHECK-NOFP16-RV64-NEXT: .cfi_def_cfa_offset 0
; CHECK-NOFP16-RV64-NEXT: ret
;
+; CHECK-SOFT-RV32-LABEL: fcanonicalize_f16:
+; CHECK-SOFT-RV32: # %bb.0:
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, -16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 16
+; CHECK-SOFT-RV32-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: .cfi_offset ra, -4
+; CHECK-SOFT-RV32-NEXT: slli a0, a0, 16
+; CHECK-SOFT-RV32-NEXT: srli a0, a0, 16
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, 16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV32-NEXT: ret
+;
; CHECK-FP16-RV32-LABEL: fcanonicalize_f16:
; CHECK-FP16-RV32: # %bb.0:
; CHECK-FP16-RV32-NEXT: fmin.h fa0, fa0, fa0
@@ -57,11 +112,64 @@ define half @fcanonicalize_f16(half %x) {
; CHECK-NOFP16-RV32-NEXT: addi sp, sp, 16
; CHECK-NOFP16-RV32-NEXT: .cfi_def_cfa_offset 0
; CHECK-NOFP16-RV32-NEXT: ret
+; RV32-SOFT-LABEL: fcanonicalize_f16:
+; RV32-SOFT: # %bb.0:
+; RV32-SOFT-NEXT: addi sp, sp, -16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 16
+; RV32-SOFT-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: .cfi_offset ra, -4
+; RV32-SOFT-NEXT: slli a0, a0, 16
+; RV32-SOFT-NEXT: srli a0, a0, 16
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: .cfi_restore ra
+; RV32-SOFT-NEXT: addi sp, sp, 16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV32-SOFT-NEXT: ret
%z = call half @llvm.canonicalize.f16(half %x)
ret half %z
}
define half @fcanonicalize_f16_nnan(half %x) {
+; RV64-SOFT-LABEL: fcanonicalize_f16_nnan:
+; RV64-SOFT: # %bb.0:
+; RV64-SOFT-NEXT: addi sp, sp, -16
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 16
+; RV64-SOFT-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: .cfi_offset ra, -8
+; RV64-SOFT-NEXT: slli a0, a0, 48
+; RV64-SOFT-NEXT: srli a0, a0, 48
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: .cfi_restore ra
+; RV64-SOFT-NEXT: addi sp, sp, 16
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV64-SOFT-NEXT: ret
+;
+; CHECK-SOFT-RV64-LABEL: fcanonicalize_f16_nnan:
+; CHECK-SOFT-RV64: # %bb.0:
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, -16
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 16
+; CHECK-SOFT-RV64-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: .cfi_offset ra, -8
+; CHECK-SOFT-RV64-NEXT: slli a0, a0, 48
+; CHECK-SOFT-RV64-NEXT: srli a0, a0, 48
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, 16
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV64-NEXT: ret
+;
; CHECK-FP16-RV64-LABEL: fcanonicalize_f16_nnan:
; CHECK-FP16-RV64: # %bb.0:
; CHECK-FP16-RV64-NEXT: fmin.h fa0, fa0, fa0
@@ -86,6 +194,24 @@ define half @fcanonicalize_f16_nnan(half %x) {
; CHECK-NOFP16-RV64-NEXT: .cfi_def_cfa_offset 0
; CHECK-NOFP16-RV64-NEXT: ret
;
+; CHECK-SOFT-RV32-LABEL: fcanonicalize_f16_nnan:
+; CHECK-SOFT-RV32: # %bb.0:
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, -16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 16
+; CHECK-SOFT-RV32-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: .cfi_offset ra, -4
+; CHECK-SOFT-RV32-NEXT: slli a0, a0, 16
+; CHECK-SOFT-RV32-NEXT: srli a0, a0, 16
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, 16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV32-NEXT: ret
+;
; CHECK-FP16-RV32-LABEL: fcanonicalize_f16_nnan:
; CHECK-FP16-RV32: # %bb.0:
; CHECK-FP16-RV32-NEXT: fmin.h fa0, fa0, fa0
@@ -109,11 +235,108 @@ define half @fcanonicalize_f16_nnan(half %x) {
; CHECK-NOFP16-RV32-NEXT: addi sp, sp, 16
; CHECK-NOFP16-RV32-NEXT: .cfi_def_cfa_offset 0
; CHECK-NOFP16-RV32-NEXT: ret
+; RV32-SOFT-LABEL: fcanonicalize_f16_nnan:
+; RV32-SOFT: # %bb.0:
+; RV32-SOFT-NEXT: addi sp, sp, -16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 16
+; RV32-SOFT-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: .cfi_offset ra, -4
+; RV32-SOFT-NEXT: slli a0, a0, 16
+; RV32-SOFT-NEXT: srli a0, a0, 16
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: .cfi_restore ra
+; RV32-SOFT-NEXT: addi sp, sp, 16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV32-SOFT-NEXT: ret
%z = call nnan half @llvm.canonicalize.f16(half %x)
ret half %z
}
define <2 x half> @fcanonicalize_v2f16(<2 x half> %x) {
+; RV64-SOFT-LABEL: fcanonicalize_v2f16:
+; RV64-SOFT: # %bb.0:
+; RV64-SOFT-NEXT: addi sp, sp, -32
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 32
+; RV64-SOFT-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s2, 0(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: .cfi_offset ra, -8
+; RV64-SOFT-NEXT: .cfi_offset s0, -16
+; RV64-SOFT-NEXT: .cfi_offset s1, -24
+; RV64-SOFT-NEXT: .cfi_offset s2, -32
+; RV64-SOFT-NEXT: mv s0, a1
+; RV64-SOFT-NEXT: lui a1, 16
+; RV64-SOFT-NEXT: addi s2, a1, -1
+; RV64-SOFT-NEXT: and a0, a0, s2
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s1, a0
+; RV64-SOFT-NEXT: and a0, s0, s2
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: mv a0, s1
+; RV64-SOFT-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s2, 0(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: .cfi_restore ra
+; RV64-SOFT-NEXT: .cfi_restore s0
+; RV64-SOFT-NEXT: .cfi_restore s1
+; RV64-SOFT-NEXT: .cfi_restore s2
+; RV64-SOFT-NEXT: addi sp, sp, 32
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV64-SOFT-NEXT: ret
+;
+; CHECK-SOFT-RV64-LABEL: fcanonicalize_v2f16:
+; CHECK-SOFT-RV64: # %bb.0:
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, -32
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 32
+; CHECK-SOFT-RV64-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s2, 0(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: .cfi_offset ra, -8
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s0, -16
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s1, -24
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s2, -32
+; CHECK-SOFT-RV64-NEXT: mv s0, a1
+; CHECK-SOFT-RV64-NEXT: lui a1, 16
+; CHECK-SOFT-RV64-NEXT: addi s2, a1, -1
+; CHECK-SOFT-RV64-NEXT: and a0, a0, s2
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s1, a0
+; CHECK-SOFT-RV64-NEXT: and a0, s0, s2
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s1
+; CHECK-SOFT-RV64-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s2, 0(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, 32
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV64-NEXT: ret
+;
; CHECK-FP16-RV64-LABEL: fcanonicalize_v2f16:
; CHECK-FP16-RV64: # %bb.0:
; CHECK-FP16-RV64-NEXT: fmin.h fa0, fa0, fa0
@@ -152,6 +375,46 @@ define <2 x half> @fcanonicalize_v2f16(<2 x half> %x) {
; CHECK-NOFP16-RV64-NEXT: .cfi_def_cfa_offset 0
; CHECK-NOFP16-RV64-NEXT: ret
;
+; CHECK-SOFT-RV32-LABEL: fcanonicalize_v2f16:
+; CHECK-SOFT-RV32: # %bb.0:
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, -16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 16
+; CHECK-SOFT-RV32-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s1, 4(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s2, 0(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: .cfi_offset ra, -4
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s0, -8
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s1, -12
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s2, -16
+; CHECK-SOFT-RV32-NEXT: mv s0, a1
+; CHECK-SOFT-RV32-NEXT: lui a1, 16
+; CHECK-SOFT-RV32-NEXT: addi s2, a1, -1
+; CHECK-SOFT-RV32-NEXT: and a0, a0, s2
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s1, a0
+; CHECK-SOFT-RV32-NEXT: and a0, s0, s2
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s1
+; CHECK-SOFT-RV32-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s1, 4(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s2, 0(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, 16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV32-NEXT: ret
+;
; CHECK-FP16-RV32-LABEL: fcanonicalize_v2f16:
; CHECK-FP16-RV32: # %bb.0:
; CHECK-FP16-RV32-NEXT: fmin.h fa0, fa0, fa0
@@ -189,11 +452,130 @@ define <2 x half> @fcanonicalize_v2f16(<2 x half> %x) {
; CHECK-NOFP16-RV32-NEXT: addi sp, sp, 16
; CHECK-NOFP16-RV32-NEXT: .cfi_def_cfa_offset 0
; CHECK-NOFP16-RV32-NEXT: ret
+; RV32-SOFT-LABEL: fcanonicalize_v2f16:
+; RV32-SOFT: # %bb.0:
+; RV32-SOFT-NEXT: addi sp, sp, -16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 16
+; RV32-SOFT-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s1, 4(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s2, 0(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: .cfi_offset ra, -4
+; RV32-SOFT-NEXT: .cfi_offset s0, -8
+; RV32-SOFT-NEXT: .cfi_offset s1, -12
+; RV32-SOFT-NEXT: .cfi_offset s2, -16
+; RV32-SOFT-NEXT: mv s0, a1
+; RV32-SOFT-NEXT: lui a1, 16
+; RV32-SOFT-NEXT: addi s2, a1, -1
+; RV32-SOFT-NEXT: and a0, a0, s2
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s1, a0
+; RV32-SOFT-NEXT: and a0, s0, s2
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: mv a0, s1
+; RV32-SOFT-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s1, 4(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s2, 0(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: .cfi_restore ra
+; RV32-SOFT-NEXT: .cfi_restore s0
+; RV32-SOFT-NEXT: .cfi_restore s1
+; RV32-SOFT-NEXT: .cfi_restore s2
+; RV32-SOFT-NEXT: addi sp, sp, 16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV32-SOFT-NEXT: ret
%z = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %x)
ret <2 x half> %z
}
define <2 x half> @fcanonicalize_v2f16_nnan(<2 x half> %x) {
+; RV64-SOFT-LABEL: fcanonicalize_v2f16_nnan:
+; RV64-SOFT: # %bb.0:
+; RV64-SOFT-NEXT: addi sp, sp, -32
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 32
+; RV64-SOFT-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s2, 0(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: .cfi_offset ra, -8
+; RV64-SOFT-NEXT: .cfi_offset s0, -16
+; RV64-SOFT-NEXT: .cfi_offset s1, -24
+; RV64-SOFT-NEXT: .cfi_offset s2, -32
+; RV64-SOFT-NEXT: mv s0, a1
+; RV64-SOFT-NEXT: lui a1, 16
+; RV64-SOFT-NEXT: addi s2, a1, -1
+; RV64-SOFT-NEXT: and a0, a0, s2
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s1, a0
+; RV64-SOFT-NEXT: and a0, s0, s2
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: mv a0, s1
+; RV64-SOFT-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s2, 0(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: .cfi_restore ra
+; RV64-SOFT-NEXT: .cfi_restore s0
+; RV64-SOFT-NEXT: .cfi_restore s1
+; RV64-SOFT-NEXT: .cfi_restore s2
+; RV64-SOFT-NEXT: addi sp, sp, 32
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV64-SOFT-NEXT: ret
+;
+; CHECK-SOFT-RV64-LABEL: fcanonicalize_v2f16_nnan:
+; CHECK-SOFT-RV64: # %bb.0:
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, -32
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 32
+; CHECK-SOFT-RV64-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s2, 0(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: .cfi_offset ra, -8
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s0, -16
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s1, -24
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s2, -32
+; CHECK-SOFT-RV64-NEXT: mv s0, a1
+; CHECK-SOFT-RV64-NEXT: lui a1, 16
+; CHECK-SOFT-RV64-NEXT: addi s2, a1, -1
+; CHECK-SOFT-RV64-NEXT: and a0, a0, s2
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s1, a0
+; CHECK-SOFT-RV64-NEXT: and a0, s0, s2
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s1
+; CHECK-SOFT-RV64-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s2, 0(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, 32
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV64-NEXT: ret
+;
; CHECK-FP16-RV64-LABEL: fcanonicalize_v2f16_nnan:
; CHECK-FP16-RV64: # %bb.0:
; CHECK-FP16-RV64-NEXT: fmin.h fa0, fa0, fa0
@@ -232,6 +614,46 @@ define <2 x half> @fcanonicalize_v2f16_nnan(<2 x half> %x) {
; CHECK-NOFP16-RV64-NEXT: .cfi_def_cfa_offset 0
; CHECK-NOFP16-RV64-NEXT: ret
;
+; CHECK-SOFT-RV32-LABEL: fcanonicalize_v2f16_nnan:
+; CHECK-SOFT-RV32: # %bb.0:
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, -16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 16
+; CHECK-SOFT-RV32-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s1, 4(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s2, 0(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: .cfi_offset ra, -4
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s0, -8
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s1, -12
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s2, -16
+; CHECK-SOFT-RV32-NEXT: mv s0, a1
+; CHECK-SOFT-RV32-NEXT: lui a1, 16
+; CHECK-SOFT-RV32-NEXT: addi s2, a1, -1
+; CHECK-SOFT-RV32-NEXT: and a0, a0, s2
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s1, a0
+; CHECK-SOFT-RV32-NEXT: and a0, s0, s2
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s1
+; CHECK-SOFT-RV32-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s1, 4(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s2, 0(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, 16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV32-NEXT: ret
+;
; CHECK-FP16-RV32-LABEL: fcanonicalize_v2f16_nnan:
; CHECK-FP16-RV32: # %bb.0:
; CHECK-FP16-RV32-NEXT: fmin.h fa0, fa0, fa0
@@ -269,11 +691,178 @@ define <2 x half> @fcanonicalize_v2f16_nnan(<2 x half> %x) {
; CHECK-NOFP16-RV32-NEXT: addi sp, sp, 16
; CHECK-NOFP16-RV32-NEXT: .cfi_def_cfa_offset 0
; CHECK-NOFP16-RV32-NEXT: ret
+; RV32-SOFT-LABEL: fcanonicalize_v2f16_nnan:
+; RV32-SOFT: # %bb.0:
+; RV32-SOFT-NEXT: addi sp, sp, -16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 16
+; RV32-SOFT-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s1, 4(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s2, 0(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: .cfi_offset ra, -4
+; RV32-SOFT-NEXT: .cfi_offset s0, -8
+; RV32-SOFT-NEXT: .cfi_offset s1, -12
+; RV32-SOFT-NEXT: .cfi_offset s2, -16
+; RV32-SOFT-NEXT: mv s0, a1
+; RV32-SOFT-NEXT: lui a1, 16
+; RV32-SOFT-NEXT: addi s2, a1, -1
+; RV32-SOFT-NEXT: and a0, a0, s2
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s1, a0
+; RV32-SOFT-NEXT: and a0, s0, s2
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: mv a0, s1
+; RV32-SOFT-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s1, 4(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s2, 0(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: .cfi_restore ra
+; RV32-SOFT-NEXT: .cfi_restore s0
+; RV32-SOFT-NEXT: .cfi_restore s1
+; RV32-SOFT-NEXT: .cfi_restore s2
+; RV32-SOFT-NEXT: addi sp, sp, 16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV32-SOFT-NEXT: ret
%z = call nnan <2 x half> @llvm.canonicalize.v2f16(<2 x half> %x)
ret <2 x half> %z
}
define <4 x half> @fcanonicalize_v4f16(<4 x half> %x) {
+; RV64-SOFT-LABEL: fcanonicalize_v4f16:
+; RV64-SOFT: # %bb.0:
+; RV64-SOFT-NEXT: addi sp, sp, -48
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 48
+; RV64-SOFT-NEXT: sd ra, 40(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s0, 32(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s1, 24(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s2, 16(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s3, 8(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s4, 0(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: .cfi_offset ra, -8
+; RV64-SOFT-NEXT: .cfi_offset s0, -16
+; RV64-SOFT-NEXT: .cfi_offset s1, -24
+; RV64-SOFT-NEXT: .cfi_offset s2, -32
+; RV64-SOFT-NEXT: .cfi_offset s3, -40
+; RV64-SOFT-NEXT: .cfi_offset s4, -48
+; RV64-SOFT-NEXT: lhu a2, 0(a1)
+; RV64-SOFT-NEXT: lhu s1, 8(a1)
+; RV64-SOFT-NEXT: lhu s2, 16(a1)
+; RV64-SOFT-NEXT: lhu s3, 24(a1)
+; RV64-SOFT-NEXT: mv s0, a0
+; RV64-SOFT-NEXT: mv a0, a2
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s4, a0
+; RV64-SOFT-NEXT: mv a0, s1
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s1, a0
+; RV64-SOFT-NEXT: mv a0, s2
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s2, a0
+; RV64-SOFT-NEXT: mv a0, s3
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: sh s4, 0(s0)
+; RV64-SOFT-NEXT: sh s1, 2(s0)
+; RV64-SOFT-NEXT: sh s2, 4(s0)
+; RV64-SOFT-NEXT: sh a0, 6(s0)
+; RV64-SOFT-NEXT: ld ra, 40(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s0, 32(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s1, 24(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s2, 16(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s3, 8(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s4, 0(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: .cfi_restore ra
+; RV64-SOFT-NEXT: .cfi_restore s0
+; RV64-SOFT-NEXT: .cfi_restore s1
+; RV64-SOFT-NEXT: .cfi_restore s2
+; RV64-SOFT-NEXT: .cfi_restore s3
+; RV64-SOFT-NEXT: .cfi_restore s4
+; RV64-SOFT-NEXT: addi sp, sp, 48
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV64-SOFT-NEXT: ret
+;
+; CHECK-SOFT-RV64-LABEL: fcanonicalize_v4f16:
+; CHECK-SOFT-RV64: # %bb.0:
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, -48
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 48
+; CHECK-SOFT-RV64-NEXT: sd ra, 40(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s0, 32(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s1, 24(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s2, 16(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s3, 8(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s4, 0(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: .cfi_offset ra, -8
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s0, -16
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s1, -24
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s2, -32
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s3, -40
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s4, -48
+; CHECK-SOFT-RV64-NEXT: lhu a2, 0(a1)
+; CHECK-SOFT-RV64-NEXT: lhu s1, 8(a1)
+; CHECK-SOFT-RV64-NEXT: lhu s2, 16(a1)
+; CHECK-SOFT-RV64-NEXT: lhu s3, 24(a1)
+; CHECK-SOFT-RV64-NEXT: mv s0, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, a2
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s4, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s1
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s1, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s2
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s2, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s3
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: sh s4, 0(s0)
+; CHECK-SOFT-RV64-NEXT: sh s1, 2(s0)
+; CHECK-SOFT-RV64-NEXT: sh s2, 4(s0)
+; CHECK-SOFT-RV64-NEXT: sh a0, 6(s0)
+; CHECK-SOFT-RV64-NEXT: ld ra, 40(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s0, 32(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s1, 24(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s2, 16(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s3, 8(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s4, 0(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s3
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s4
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, 48
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV64-NEXT: ret
+;
; CHECK-FP16-RV64-LABEL: fcanonicalize_v4f16:
; CHECK-FP16-RV64: # %bb.0:
; CHECK-FP16-RV64-NEXT: fmin.h fa5, fa0, fa0
@@ -358,6 +947,70 @@ define <4 x half> @fcanonicalize_v4f16(<4 x half> %x) {
; CHECK-NOFP16-RV64-NEXT: .cfi_def_cfa_offset 0
; CHECK-NOFP16-RV64-NEXT: ret
;
+; CHECK-SOFT-RV32-LABEL: fcanonicalize_v4f16:
+; CHECK-SOFT-RV32: # %bb.0:
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, -32
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 32
+; CHECK-SOFT-RV32-NEXT: sw ra, 28(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s0, 24(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s1, 20(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s2, 16(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s3, 12(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s4, 8(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: .cfi_offset ra, -4
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s0, -8
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s1, -12
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s2, -16
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s3, -20
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s4, -24
+; CHECK-SOFT-RV32-NEXT: lhu a2, 0(a1)
+; CHECK-SOFT-RV32-NEXT: lhu s1, 4(a1)
+; CHECK-SOFT-RV32-NEXT: lhu s2, 8(a1)
+; CHECK-SOFT-RV32-NEXT: lhu s3, 12(a1)
+; CHECK-SOFT-RV32-NEXT: mv s0, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, a2
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s4, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s1
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s1, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s2
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s2, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s3
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: sh s4, 0(s0)
+; CHECK-SOFT-RV32-NEXT: sh s1, 2(s0)
+; CHECK-SOFT-RV32-NEXT: sh s2, 4(s0)
+; CHECK-SOFT-RV32-NEXT: sh a0, 6(s0)
+; CHECK-SOFT-RV32-NEXT: lw ra, 28(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s0, 24(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s1, 20(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s2, 16(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s3, 12(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s4, 8(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s3
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s4
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, 32
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV32-NEXT: ret
+;
; CHECK-FP16-RV32-LABEL: fcanonicalize_v4f16:
; CHECK-FP16-RV32: # %bb.0:
; CHECK-FP16-RV32-NEXT: fmin.h fa5, fa0, fa0
@@ -448,11 +1101,202 @@ define <4 x half> @fcanonicalize_v4f16(<4 x half> %x) {
; CHECK-NOFP16-RV32-NEXT: addi sp, sp, 64
; CHECK-NOFP16-RV32-NEXT: .cfi_def_cfa_offset 0
; CHECK-NOFP16-RV32-NEXT: ret
+; RV32-SOFT-LABEL: fcanonicalize_v4f16:
+; RV32-SOFT: # %bb.0:
+; RV32-SOFT-NEXT: addi sp, sp, -32
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 32
+; RV32-SOFT-NEXT: sw ra, 28(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s0, 24(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s1, 20(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s2, 16(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s3, 12(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s4, 8(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: .cfi_offset ra, -4
+; RV32-SOFT-NEXT: .cfi_offset s0, -8
+; RV32-SOFT-NEXT: .cfi_offset s1, -12
+; RV32-SOFT-NEXT: .cfi_offset s2, -16
+; RV32-SOFT-NEXT: .cfi_offset s3, -20
+; RV32-SOFT-NEXT: .cfi_offset s4, -24
+; RV32-SOFT-NEXT: lhu a2, 0(a1)
+; RV32-SOFT-NEXT: lhu s1, 4(a1)
+; RV32-SOFT-NEXT: lhu s2, 8(a1)
+; RV32-SOFT-NEXT: lhu s3, 12(a1)
+; RV32-SOFT-NEXT: mv s0, a0
+; RV32-SOFT-NEXT: mv a0, a2
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s4, a0
+; RV32-SOFT-NEXT: mv a0, s1
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s1, a0
+; RV32-SOFT-NEXT: mv a0, s2
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s2, a0
+; RV32-SOFT-NEXT: mv a0, s3
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: sh s4, 0(s0)
+; RV32-SOFT-NEXT: sh s1, 2(s0)
+; RV32-SOFT-NEXT: sh s2, 4(s0)
+; RV32-SOFT-NEXT: sh a0, 6(s0)
+; RV32-SOFT-NEXT: lw ra, 28(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s0, 24(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s1, 20(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s2, 16(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s3, 12(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s4, 8(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: .cfi_restore ra
+; RV32-SOFT-NEXT: .cfi_restore s0
+; RV32-SOFT-NEXT: .cfi_restore s1
+; RV32-SOFT-NEXT: .cfi_restore s2
+; RV32-SOFT-NEXT: .cfi_restore s3
+; RV32-SOFT-NEXT: .cfi_restore s4
+; RV32-SOFT-NEXT: addi sp, sp, 32
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV32-SOFT-NEXT: ret
%z = call <4 x half> @llvm.canonicalize.v4f16(<4 x half> %x)
ret <4 x half> %z
}
define <4 x half> @fcanonicalize_v4f16_nnan(<4 x half> %x) {
+; RV64-SOFT-LABEL: fcanonicalize_v4f16_nnan:
+; RV64-SOFT: # %bb.0:
+; RV64-SOFT-NEXT: addi sp, sp, -48
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 48
+; RV64-SOFT-NEXT: sd ra, 40(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s0, 32(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s1, 24(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s2, 16(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s3, 8(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s4, 0(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: .cfi_offset ra, -8
+; RV64-SOFT-NEXT: .cfi_offset s0, -16
+; RV64-SOFT-NEXT: .cfi_offset s1, -24
+; RV64-SOFT-NEXT: .cfi_offset s2, -32
+; RV64-SOFT-NEXT: .cfi_offset s3, -40
+; RV64-SOFT-NEXT: .cfi_offset s4, -48
+; RV64-SOFT-NEXT: lhu a2, 0(a1)
+; RV64-SOFT-NEXT: lhu s1, 8(a1)
+; RV64-SOFT-NEXT: lhu s2, 16(a1)
+; RV64-SOFT-NEXT: lhu s3, 24(a1)
+; RV64-SOFT-NEXT: mv s0, a0
+; RV64-SOFT-NEXT: mv a0, a2
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s4, a0
+; RV64-SOFT-NEXT: mv a0, s1
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s1, a0
+; RV64-SOFT-NEXT: mv a0, s2
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s2, a0
+; RV64-SOFT-NEXT: mv a0, s3
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: sh s4, 0(s0)
+; RV64-SOFT-NEXT: sh s1, 2(s0)
+; RV64-SOFT-NEXT: sh s2, 4(s0)
+; RV64-SOFT-NEXT: sh a0, 6(s0)
+; RV64-SOFT-NEXT: ld ra, 40(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s0, 32(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s1, 24(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s2, 16(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s3, 8(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s4, 0(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: .cfi_restore ra
+; RV64-SOFT-NEXT: .cfi_restore s0
+; RV64-SOFT-NEXT: .cfi_restore s1
+; RV64-SOFT-NEXT: .cfi_restore s2
+; RV64-SOFT-NEXT: .cfi_restore s3
+; RV64-SOFT-NEXT: .cfi_restore s4
+; RV64-SOFT-NEXT: addi sp, sp, 48
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV64-SOFT-NEXT: ret
+;
+; CHECK-SOFT-RV64-LABEL: fcanonicalize_v4f16_nnan:
+; CHECK-SOFT-RV64: # %bb.0:
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, -48
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 48
+; CHECK-SOFT-RV64-NEXT: sd ra, 40(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s0, 32(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s1, 24(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s2, 16(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s3, 8(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s4, 0(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: .cfi_offset ra, -8
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s0, -16
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s1, -24
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s2, -32
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s3, -40
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s4, -48
+; CHECK-SOFT-RV64-NEXT: lhu a2, 0(a1)
+; CHECK-SOFT-RV64-NEXT: lhu s1, 8(a1)
+; CHECK-SOFT-RV64-NEXT: lhu s2, 16(a1)
+; CHECK-SOFT-RV64-NEXT: lhu s3, 24(a1)
+; CHECK-SOFT-RV64-NEXT: mv s0, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, a2
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s4, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s1
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s1, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s2
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s2, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s3
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: sh s4, 0(s0)
+; CHECK-SOFT-RV64-NEXT: sh s1, 2(s0)
+; CHECK-SOFT-RV64-NEXT: sh s2, 4(s0)
+; CHECK-SOFT-RV64-NEXT: sh a0, 6(s0)
+; CHECK-SOFT-RV64-NEXT: ld ra, 40(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s0, 32(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s1, 24(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s2, 16(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s3, 8(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s4, 0(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s3
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s4
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, 48
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV64-NEXT: ret
+;
; CHECK-FP16-RV64-LABEL: fcanonicalize_v4f16_nnan:
; CHECK-FP16-RV64: # %bb.0:
; CHECK-FP16-RV64-NEXT: fmin.h fa5, fa0, fa0
@@ -537,6 +1381,70 @@ define <4 x half> @fcanonicalize_v4f16_nnan(<4 x half> %x) {
; CHECK-NOFP16-RV64-NEXT: .cfi_def_cfa_offset 0
; CHECK-NOFP16-RV64-NEXT: ret
;
+; CHECK-SOFT-RV32-LABEL: fcanonicalize_v4f16_nnan:
+; CHECK-SOFT-RV32: # %bb.0:
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, -32
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 32
+; CHECK-SOFT-RV32-NEXT: sw ra, 28(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s0, 24(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s1, 20(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s2, 16(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s3, 12(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s4, 8(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: .cfi_offset ra, -4
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s0, -8
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s1, -12
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s2, -16
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s3, -20
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s4, -24
+; CHECK-SOFT-RV32-NEXT: lhu a2, 0(a1)
+; CHECK-SOFT-RV32-NEXT: lhu s1, 4(a1)
+; CHECK-SOFT-RV32-NEXT: lhu s2, 8(a1)
+; CHECK-SOFT-RV32-NEXT: lhu s3, 12(a1)
+; CHECK-SOFT-RV32-NEXT: mv s0, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, a2
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s4, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s1
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s1, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s2
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s2, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s3
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: sh s4, 0(s0)
+; CHECK-SOFT-RV32-NEXT: sh s1, 2(s0)
+; CHECK-SOFT-RV32-NEXT: sh s2, 4(s0)
+; CHECK-SOFT-RV32-NEXT: sh a0, 6(s0)
+; CHECK-SOFT-RV32-NEXT: lw ra, 28(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s0, 24(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s1, 20(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s2, 16(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s3, 12(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s4, 8(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s3
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s4
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, 32
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV32-NEXT: ret
+;
; CHECK-FP16-RV32-LABEL: fcanonicalize_v4f16_nnan:
; CHECK-FP16-RV32: # %bb.0:
; CHECK-FP16-RV32-NEXT: fmin.h fa5, fa0, fa0
@@ -627,11 +1535,298 @@ define <4 x half> @fcanonicalize_v4f16_nnan(<4 x half> %x) {
; CHECK-NOFP16-RV32-NEXT: addi sp, sp, 64
; CHECK-NOFP16-RV32-NEXT: .cfi_def_cfa_offset 0
; CHECK-NOFP16-RV32-NEXT: ret
+; RV32-SOFT-LABEL: fcanonicalize_v4f16_nnan:
+; RV32-SOFT: # %bb.0:
+; RV32-SOFT-NEXT: addi sp, sp, -32
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 32
+; RV32-SOFT-NEXT: sw ra, 28(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s0, 24(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s1, 20(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s2, 16(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s3, 12(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s4, 8(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: .cfi_offset ra, -4
+; RV32-SOFT-NEXT: .cfi_offset s0, -8
+; RV32-SOFT-NEXT: .cfi_offset s1, -12
+; RV32-SOFT-NEXT: .cfi_offset s2, -16
+; RV32-SOFT-NEXT: .cfi_offset s3, -20
+; RV32-SOFT-NEXT: .cfi_offset s4, -24
+; RV32-SOFT-NEXT: lhu a2, 0(a1)
+; RV32-SOFT-NEXT: lhu s1, 4(a1)
+; RV32-SOFT-NEXT: lhu s2, 8(a1)
+; RV32-SOFT-NEXT: lhu s3, 12(a1)
+; RV32-SOFT-NEXT: mv s0, a0
+; RV32-SOFT-NEXT: mv a0, a2
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s4, a0
+; RV32-SOFT-NEXT: mv a0, s1
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s1, a0
+; RV32-SOFT-NEXT: mv a0, s2
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s2, a0
+; RV32-SOFT-NEXT: mv a0, s3
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: sh s4, 0(s0)
+; RV32-SOFT-NEXT: sh s1, 2(s0)
+; RV32-SOFT-NEXT: sh s2, 4(s0)
+; RV32-SOFT-NEXT: sh a0, 6(s0)
+; RV32-SOFT-NEXT: lw ra, 28(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s0, 24(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s1, 20(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s2, 16(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s3, 12(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s4, 8(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: .cfi_restore ra
+; RV32-SOFT-NEXT: .cfi_restore s0
+; RV32-SOFT-NEXT: .cfi_restore s1
+; RV32-SOFT-NEXT: .cfi_restore s2
+; RV32-SOFT-NEXT: .cfi_restore s3
+; RV32-SOFT-NEXT: .cfi_restore s4
+; RV32-SOFT-NEXT: addi sp, sp, 32
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV32-SOFT-NEXT: ret
%z = call nnan <4 x half> @llvm.canonicalize.v4f16(<4 x half> %x)
ret <4 x half> %z
}
define <8 x half> @fcanonicalize_v8f16(<8 x half> %x) {
+; RV64-SOFT-LABEL: fcanonicalize_v8f16:
+; RV64-SOFT: # %bb.0:
+; RV64-SOFT-NEXT: addi sp, sp, -80
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 80
+; RV64-SOFT-NEXT: sd ra, 72(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s0, 64(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s1, 56(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s2, 48(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s3, 40(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s4, 32(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s5, 24(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s6, 16(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s7, 8(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s8, 0(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: .cfi_offset ra, -8
+; RV64-SOFT-NEXT: .cfi_offset s0, -16
+; RV64-SOFT-NEXT: .cfi_offset s1, -24
+; RV64-SOFT-NEXT: .cfi_offset s2, -32
+; RV64-SOFT-NEXT: .cfi_offset s3, -40
+; RV64-SOFT-NEXT: .cfi_offset s4, -48
+; RV64-SOFT-NEXT: .cfi_offset s5, -56
+; RV64-SOFT-NEXT: .cfi_offset s6, -64
+; RV64-SOFT-NEXT: .cfi_offset s7, -72
+; RV64-SOFT-NEXT: .cfi_offset s8, -80
+; RV64-SOFT-NEXT: lhu s7, 32(a1)
+; RV64-SOFT-NEXT: lhu s5, 40(a1)
+; RV64-SOFT-NEXT: lhu s3, 48(a1)
+; RV64-SOFT-NEXT: lhu s1, 56(a1)
+; RV64-SOFT-NEXT: lhu a2, 0(a1)
+; RV64-SOFT-NEXT: lhu s4, 8(a1)
+; RV64-SOFT-NEXT: lhu s6, 16(a1)
+; RV64-SOFT-NEXT: lhu s8, 24(a1)
+; RV64-SOFT-NEXT: mv s0, a0
+; RV64-SOFT-NEXT: mv a0, a2
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s2, a0
+; RV64-SOFT-NEXT: mv a0, s4
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s4, a0
+; RV64-SOFT-NEXT: mv a0, s6
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s6, a0
+; RV64-SOFT-NEXT: mv a0, s8
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s8, a0
+; RV64-SOFT-NEXT: mv a0, s7
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s7, a0
+; RV64-SOFT-NEXT: mv a0, s5
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s5, a0
+; RV64-SOFT-NEXT: mv a0, s3
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s3, a0
+; RV64-SOFT-NEXT: mv a0, s1
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: sh s7, 8(s0)
+; RV64-SOFT-NEXT: sh s5, 10(s0)
+; RV64-SOFT-NEXT: sh s3, 12(s0)
+; RV64-SOFT-NEXT: sh a0, 14(s0)
+; RV64-SOFT-NEXT: sh s2, 0(s0)
+; RV64-SOFT-NEXT: sh s4, 2(s0)
+; RV64-SOFT-NEXT: sh s6, 4(s0)
+; RV64-SOFT-NEXT: sh s8, 6(s0)
+; RV64-SOFT-NEXT: ld ra, 72(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s0, 64(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s1, 56(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s2, 48(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s3, 40(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s4, 32(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s5, 24(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s6, 16(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s7, 8(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s8, 0(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: .cfi_restore ra
+; RV64-SOFT-NEXT: .cfi_restore s0
+; RV64-SOFT-NEXT: .cfi_restore s1
+; RV64-SOFT-NEXT: .cfi_restore s2
+; RV64-SOFT-NEXT: .cfi_restore s3
+; RV64-SOFT-NEXT: .cfi_restore s4
+; RV64-SOFT-NEXT: .cfi_restore s5
+; RV64-SOFT-NEXT: .cfi_restore s6
+; RV64-SOFT-NEXT: .cfi_restore s7
+; RV64-SOFT-NEXT: .cfi_restore s8
+; RV64-SOFT-NEXT: addi sp, sp, 80
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV64-SOFT-NEXT: ret
+;
+; CHECK-SOFT-RV64-LABEL: fcanonicalize_v8f16:
+; CHECK-SOFT-RV64: # %bb.0:
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, -80
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 80
+; CHECK-SOFT-RV64-NEXT: sd ra, 72(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s0, 64(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s1, 56(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s2, 48(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s3, 40(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s4, 32(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s5, 24(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s6, 16(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s7, 8(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s8, 0(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: .cfi_offset ra, -8
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s0, -16
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s1, -24
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s2, -32
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s3, -40
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s4, -48
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s5, -56
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s6, -64
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s7, -72
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s8, -80
+; CHECK-SOFT-RV64-NEXT: lhu s7, 32(a1)
+; CHECK-SOFT-RV64-NEXT: lhu s5, 40(a1)
+; CHECK-SOFT-RV64-NEXT: lhu s3, 48(a1)
+; CHECK-SOFT-RV64-NEXT: lhu s1, 56(a1)
+; CHECK-SOFT-RV64-NEXT: lhu a2, 0(a1)
+; CHECK-SOFT-RV64-NEXT: lhu s4, 8(a1)
+; CHECK-SOFT-RV64-NEXT: lhu s6, 16(a1)
+; CHECK-SOFT-RV64-NEXT: lhu s8, 24(a1)
+; CHECK-SOFT-RV64-NEXT: mv s0, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, a2
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s2, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s4
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s4, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s6
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s6, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s8
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s8, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s7
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s7, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s5
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s5, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s3
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s3, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s1
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: sh s7, 8(s0)
+; CHECK-SOFT-RV64-NEXT: sh s5, 10(s0)
+; CHECK-SOFT-RV64-NEXT: sh s3, 12(s0)
+; CHECK-SOFT-RV64-NEXT: sh a0, 14(s0)
+; CHECK-SOFT-RV64-NEXT: sh s2, 0(s0)
+; CHECK-SOFT-RV64-NEXT: sh s4, 2(s0)
+; CHECK-SOFT-RV64-NEXT: sh s6, 4(s0)
+; CHECK-SOFT-RV64-NEXT: sh s8, 6(s0)
+; CHECK-SOFT-RV64-NEXT: ld ra, 72(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s0, 64(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s1, 56(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s2, 48(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s3, 40(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s4, 32(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s5, 24(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s6, 16(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s7, 8(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s8, 0(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s3
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s4
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s5
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s6
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s7
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s8
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, 80
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV64-NEXT: ret
+;
; CHECK-FP16-RV64-LABEL: fcanonicalize_v8f16:
; CHECK-FP16-RV64: # %bb.0:
; CHECK-FP16-RV64-NEXT: fmin.h fa0, fa0, fa0
@@ -788,6 +1983,118 @@ define <8 x half> @fcanonicalize_v8f16(<8 x half> %x) {
; CHECK-NOFP16-RV64-NEXT: .cfi_def_cfa_offset 0
; CHECK-NOFP16-RV64-NEXT: ret
;
+; CHECK-SOFT-RV32-LABEL: fcanonicalize_v8f16:
+; CHECK-SOFT-RV32: # %bb.0:
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, -48
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 48
+; CHECK-SOFT-RV32-NEXT: sw ra, 44(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s0, 40(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s1, 36(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s2, 32(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s3, 28(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s4, 24(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s5, 20(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s6, 16(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s7, 12(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s8, 8(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: .cfi_offset ra, -4
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s0, -8
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s1, -12
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s2, -16
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s3, -20
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s4, -24
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s5, -28
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s6, -32
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s7, -36
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s8, -40
+; CHECK-SOFT-RV32-NEXT: lhu s7, 16(a1)
+; CHECK-SOFT-RV32-NEXT: lhu s5, 20(a1)
+; CHECK-SOFT-RV32-NEXT: lhu s3, 24(a1)
+; CHECK-SOFT-RV32-NEXT: lhu s1, 28(a1)
+; CHECK-SOFT-RV32-NEXT: lhu a2, 0(a1)
+; CHECK-SOFT-RV32-NEXT: lhu s4, 4(a1)
+; CHECK-SOFT-RV32-NEXT: lhu s6, 8(a1)
+; CHECK-SOFT-RV32-NEXT: lhu s8, 12(a1)
+; CHECK-SOFT-RV32-NEXT: mv s0, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, a2
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s2, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s4
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s4, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s6
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s6, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s8
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s8, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s7
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s7, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s5
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s5, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s3
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s3, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s1
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: sh s7, 8(s0)
+; CHECK-SOFT-RV32-NEXT: sh s5, 10(s0)
+; CHECK-SOFT-RV32-NEXT: sh s3, 12(s0)
+; CHECK-SOFT-RV32-NEXT: sh a0, 14(s0)
+; CHECK-SOFT-RV32-NEXT: sh s2, 0(s0)
+; CHECK-SOFT-RV32-NEXT: sh s4, 2(s0)
+; CHECK-SOFT-RV32-NEXT: sh s6, 4(s0)
+; CHECK-SOFT-RV32-NEXT: sh s8, 6(s0)
+; CHECK-SOFT-RV32-NEXT: lw ra, 44(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s0, 40(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s1, 36(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s2, 32(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s3, 28(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s4, 24(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s5, 20(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s6, 16(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s7, 12(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s8, 8(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s3
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s4
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s5
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s6
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s7
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s8
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, 48
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV32-NEXT: ret
+;
; CHECK-FP16-RV32-LABEL: fcanonicalize_v8f16:
; CHECK-FP16-RV32: # %bb.0:
; CHECK-FP16-RV32-NEXT: fmin.h fa0, fa0, fa0
@@ -954,11 +2261,346 @@ define <8 x half> @fcanonicalize_v8f16(<8 x half> %x) {
; CHECK-NOFP16-RV32-NEXT: addi sp, sp, 112
; CHECK-NOFP16-RV32-NEXT: .cfi_def_cfa_offset 0
; CHECK-NOFP16-RV32-NEXT: ret
+; RV32-SOFT-LABEL: fcanonicalize_v8f16:
+; RV32-SOFT: # %bb.0:
+; RV32-SOFT-NEXT: addi sp, sp, -48
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 48
+; RV32-SOFT-NEXT: sw ra, 44(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s0, 40(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s1, 36(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s2, 32(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s3, 28(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s4, 24(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s5, 20(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s6, 16(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s7, 12(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s8, 8(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: .cfi_offset ra, -4
+; RV32-SOFT-NEXT: .cfi_offset s0, -8
+; RV32-SOFT-NEXT: .cfi_offset s1, -12
+; RV32-SOFT-NEXT: .cfi_offset s2, -16
+; RV32-SOFT-NEXT: .cfi_offset s3, -20
+; RV32-SOFT-NEXT: .cfi_offset s4, -24
+; RV32-SOFT-NEXT: .cfi_offset s5, -28
+; RV32-SOFT-NEXT: .cfi_offset s6, -32
+; RV32-SOFT-NEXT: .cfi_offset s7, -36
+; RV32-SOFT-NEXT: .cfi_offset s8, -40
+; RV32-SOFT-NEXT: lhu s7, 16(a1)
+; RV32-SOFT-NEXT: lhu s5, 20(a1)
+; RV32-SOFT-NEXT: lhu s3, 24(a1)
+; RV32-SOFT-NEXT: lhu s1, 28(a1)
+; RV32-SOFT-NEXT: lhu a2, 0(a1)
+; RV32-SOFT-NEXT: lhu s4, 4(a1)
+; RV32-SOFT-NEXT: lhu s6, 8(a1)
+; RV32-SOFT-NEXT: lhu s8, 12(a1)
+; RV32-SOFT-NEXT: mv s0, a0
+; RV32-SOFT-NEXT: mv a0, a2
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s2, a0
+; RV32-SOFT-NEXT: mv a0, s4
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s4, a0
+; RV32-SOFT-NEXT: mv a0, s6
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s6, a0
+; RV32-SOFT-NEXT: mv a0, s8
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s8, a0
+; RV32-SOFT-NEXT: mv a0, s7
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s7, a0
+; RV32-SOFT-NEXT: mv a0, s5
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s5, a0
+; RV32-SOFT-NEXT: mv a0, s3
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s3, a0
+; RV32-SOFT-NEXT: mv a0, s1
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: sh s7, 8(s0)
+; RV32-SOFT-NEXT: sh s5, 10(s0)
+; RV32-SOFT-NEXT: sh s3, 12(s0)
+; RV32-SOFT-NEXT: sh a0, 14(s0)
+; RV32-SOFT-NEXT: sh s2, 0(s0)
+; RV32-SOFT-NEXT: sh s4, 2(s0)
+; RV32-SOFT-NEXT: sh s6, 4(s0)
+; RV32-SOFT-NEXT: sh s8, 6(s0)
+; RV32-SOFT-NEXT: lw ra, 44(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s0, 40(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s1, 36(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s2, 32(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s3, 28(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s4, 24(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s5, 20(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s6, 16(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s7, 12(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s8, 8(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: .cfi_restore ra
+; RV32-SOFT-NEXT: .cfi_restore s0
+; RV32-SOFT-NEXT: .cfi_restore s1
+; RV32-SOFT-NEXT: .cfi_restore s2
+; RV32-SOFT-NEXT: .cfi_restore s3
+; RV32-SOFT-NEXT: .cfi_restore s4
+; RV32-SOFT-NEXT: .cfi_restore s5
+; RV32-SOFT-NEXT: .cfi_restore s6
+; RV32-SOFT-NEXT: .cfi_restore s7
+; RV32-SOFT-NEXT: .cfi_restore s8
+; RV32-SOFT-NEXT: addi sp, sp, 48
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV32-SOFT-NEXT: ret
%z = call <8 x half> @llvm.canonicalize.v8f16(<8 x half> %x)
ret <8 x half> %z
}
define <8 x half> @fcanonicalize_v8f16_nnan(<8 x half> %x) {
+; RV64-SOFT-LABEL: fcanonicalize_v8f16_nnan:
+; RV64-SOFT: # %bb.0:
+; RV64-SOFT-NEXT: addi sp, sp, -80
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 80
+; RV64-SOFT-NEXT: sd ra, 72(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s0, 64(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s1, 56(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s2, 48(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s3, 40(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s4, 32(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s5, 24(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s6, 16(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s7, 8(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s8, 0(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: .cfi_offset ra, -8
+; RV64-SOFT-NEXT: .cfi_offset s0, -16
+; RV64-SOFT-NEXT: .cfi_offset s1, -24
+; RV64-SOFT-NEXT: .cfi_offset s2, -32
+; RV64-SOFT-NEXT: .cfi_offset s3, -40
+; RV64-SOFT-NEXT: .cfi_offset s4, -48
+; RV64-SOFT-NEXT: .cfi_offset s5, -56
+; RV64-SOFT-NEXT: .cfi_offset s6, -64
+; RV64-SOFT-NEXT: .cfi_offset s7, -72
+; RV64-SOFT-NEXT: .cfi_offset s8, -80
+; RV64-SOFT-NEXT: lhu s7, 32(a1)
+; RV64-SOFT-NEXT: lhu s5, 40(a1)
+; RV64-SOFT-NEXT: lhu s3, 48(a1)
+; RV64-SOFT-NEXT: lhu s1, 56(a1)
+; RV64-SOFT-NEXT: lhu a2, 0(a1)
+; RV64-SOFT-NEXT: lhu s4, 8(a1)
+; RV64-SOFT-NEXT: lhu s6, 16(a1)
+; RV64-SOFT-NEXT: lhu s8, 24(a1)
+; RV64-SOFT-NEXT: mv s0, a0
+; RV64-SOFT-NEXT: mv a0, a2
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s2, a0
+; RV64-SOFT-NEXT: mv a0, s4
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s4, a0
+; RV64-SOFT-NEXT: mv a0, s6
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s6, a0
+; RV64-SOFT-NEXT: mv a0, s8
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s8, a0
+; RV64-SOFT-NEXT: mv a0, s7
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s7, a0
+; RV64-SOFT-NEXT: mv a0, s5
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s5, a0
+; RV64-SOFT-NEXT: mv a0, s3
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: mv s3, a0
+; RV64-SOFT-NEXT: mv a0, s1
+; RV64-SOFT-NEXT: call __extendhfsf2
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: call __truncsfhf2
+; RV64-SOFT-NEXT: sh s7, 8(s0)
+; RV64-SOFT-NEXT: sh s5, 10(s0)
+; RV64-SOFT-NEXT: sh s3, 12(s0)
+; RV64-SOFT-NEXT: sh a0, 14(s0)
+; RV64-SOFT-NEXT: sh s2, 0(s0)
+; RV64-SOFT-NEXT: sh s4, 2(s0)
+; RV64-SOFT-NEXT: sh s6, 4(s0)
+; RV64-SOFT-NEXT: sh s8, 6(s0)
+; RV64-SOFT-NEXT: ld ra, 72(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s0, 64(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s1, 56(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s2, 48(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s3, 40(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s4, 32(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s5, 24(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s6, 16(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s7, 8(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s8, 0(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: .cfi_restore ra
+; RV64-SOFT-NEXT: .cfi_restore s0
+; RV64-SOFT-NEXT: .cfi_restore s1
+; RV64-SOFT-NEXT: .cfi_restore s2
+; RV64-SOFT-NEXT: .cfi_restore s3
+; RV64-SOFT-NEXT: .cfi_restore s4
+; RV64-SOFT-NEXT: .cfi_restore s5
+; RV64-SOFT-NEXT: .cfi_restore s6
+; RV64-SOFT-NEXT: .cfi_restore s7
+; RV64-SOFT-NEXT: .cfi_restore s8
+; RV64-SOFT-NEXT: addi sp, sp, 80
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV64-SOFT-NEXT: ret
+;
+; CHECK-SOFT-RV64-LABEL: fcanonicalize_v8f16_nnan:
+; CHECK-SOFT-RV64: # %bb.0:
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, -80
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 80
+; CHECK-SOFT-RV64-NEXT: sd ra, 72(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s0, 64(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s1, 56(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s2, 48(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s3, 40(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s4, 32(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s5, 24(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s6, 16(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s7, 8(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s8, 0(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: .cfi_offset ra, -8
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s0, -16
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s1, -24
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s2, -32
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s3, -40
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s4, -48
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s5, -56
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s6, -64
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s7, -72
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s8, -80
+; CHECK-SOFT-RV64-NEXT: lhu s7, 32(a1)
+; CHECK-SOFT-RV64-NEXT: lhu s5, 40(a1)
+; CHECK-SOFT-RV64-NEXT: lhu s3, 48(a1)
+; CHECK-SOFT-RV64-NEXT: lhu s1, 56(a1)
+; CHECK-SOFT-RV64-NEXT: lhu a2, 0(a1)
+; CHECK-SOFT-RV64-NEXT: lhu s4, 8(a1)
+; CHECK-SOFT-RV64-NEXT: lhu s6, 16(a1)
+; CHECK-SOFT-RV64-NEXT: lhu s8, 24(a1)
+; CHECK-SOFT-RV64-NEXT: mv s0, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, a2
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s2, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s4
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s4, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s6
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s6, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s8
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s8, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s7
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s7, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s5
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s5, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s3
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: mv s3, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s1
+; CHECK-SOFT-RV64-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV64-NEXT: sh s7, 8(s0)
+; CHECK-SOFT-RV64-NEXT: sh s5, 10(s0)
+; CHECK-SOFT-RV64-NEXT: sh s3, 12(s0)
+; CHECK-SOFT-RV64-NEXT: sh a0, 14(s0)
+; CHECK-SOFT-RV64-NEXT: sh s2, 0(s0)
+; CHECK-SOFT-RV64-NEXT: sh s4, 2(s0)
+; CHECK-SOFT-RV64-NEXT: sh s6, 4(s0)
+; CHECK-SOFT-RV64-NEXT: sh s8, 6(s0)
+; CHECK-SOFT-RV64-NEXT: ld ra, 72(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s0, 64(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s1, 56(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s2, 48(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s3, 40(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s4, 32(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s5, 24(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s6, 16(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s7, 8(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s8, 0(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s3
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s4
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s5
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s6
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s7
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s8
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, 80
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV64-NEXT: ret
+;
; CHECK-FP16-RV64-LABEL: fcanonicalize_v8f16_nnan:
; CHECK-FP16-RV64: # %bb.0:
; CHECK-FP16-RV64-NEXT: fmin.h fa0, fa0, fa0
@@ -1115,6 +2757,118 @@ define <8 x half> @fcanonicalize_v8f16_nnan(<8 x half> %x) {
; CHECK-NOFP16-RV64-NEXT: .cfi_def_cfa_offset 0
; CHECK-NOFP16-RV64-NEXT: ret
;
+; CHECK-SOFT-RV32-LABEL: fcanonicalize_v8f16_nnan:
+; CHECK-SOFT-RV32: # %bb.0:
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, -48
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 48
+; CHECK-SOFT-RV32-NEXT: sw ra, 44(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s0, 40(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s1, 36(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s2, 32(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s3, 28(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s4, 24(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s5, 20(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s6, 16(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s7, 12(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s8, 8(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: .cfi_offset ra, -4
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s0, -8
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s1, -12
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s2, -16
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s3, -20
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s4, -24
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s5, -28
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s6, -32
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s7, -36
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s8, -40
+; CHECK-SOFT-RV32-NEXT: lhu s7, 16(a1)
+; CHECK-SOFT-RV32-NEXT: lhu s5, 20(a1)
+; CHECK-SOFT-RV32-NEXT: lhu s3, 24(a1)
+; CHECK-SOFT-RV32-NEXT: lhu s1, 28(a1)
+; CHECK-SOFT-RV32-NEXT: lhu a2, 0(a1)
+; CHECK-SOFT-RV32-NEXT: lhu s4, 4(a1)
+; CHECK-SOFT-RV32-NEXT: lhu s6, 8(a1)
+; CHECK-SOFT-RV32-NEXT: lhu s8, 12(a1)
+; CHECK-SOFT-RV32-NEXT: mv s0, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, a2
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s2, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s4
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s4, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s6
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s6, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s8
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s8, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s7
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s7, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s5
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s5, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s3
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: mv s3, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s1
+; CHECK-SOFT-RV32-NEXT: call __extendhfsf2
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: call __truncsfhf2
+; CHECK-SOFT-RV32-NEXT: sh s7, 8(s0)
+; CHECK-SOFT-RV32-NEXT: sh s5, 10(s0)
+; CHECK-SOFT-RV32-NEXT: sh s3, 12(s0)
+; CHECK-SOFT-RV32-NEXT: sh a0, 14(s0)
+; CHECK-SOFT-RV32-NEXT: sh s2, 0(s0)
+; CHECK-SOFT-RV32-NEXT: sh s4, 2(s0)
+; CHECK-SOFT-RV32-NEXT: sh s6, 4(s0)
+; CHECK-SOFT-RV32-NEXT: sh s8, 6(s0)
+; CHECK-SOFT-RV32-NEXT: lw ra, 44(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s0, 40(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s1, 36(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s2, 32(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s3, 28(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s4, 24(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s5, 20(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s6, 16(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s7, 12(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s8, 8(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s3
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s4
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s5
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s6
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s7
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s8
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, 48
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV32-NEXT: ret
+;
; CHECK-FP16-RV32-LABEL: fcanonicalize_v8f16_nnan:
; CHECK-FP16-RV32: # %bb.0:
; CHECK-FP16-RV32-NEXT: fmin.h fa0, fa0, fa0
@@ -1281,114 +3035,1899 @@ define <8 x half> @fcanonicalize_v8f16_nnan(<8 x half> %x) {
; CHECK-NOFP16-RV32-NEXT: addi sp, sp, 112
; CHECK-NOFP16-RV32-NEXT: .cfi_def_cfa_offset 0
; CHECK-NOFP16-RV32-NEXT: ret
+; RV32-SOFT-LABEL: fcanonicalize_v8f16_nnan:
+; RV32-SOFT: # %bb.0:
+; RV32-SOFT-NEXT: addi sp, sp, -48
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 48
+; RV32-SOFT-NEXT: sw ra, 44(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s0, 40(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s1, 36(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s2, 32(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s3, 28(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s4, 24(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s5, 20(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s6, 16(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s7, 12(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s8, 8(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: .cfi_offset ra, -4
+; RV32-SOFT-NEXT: .cfi_offset s0, -8
+; RV32-SOFT-NEXT: .cfi_offset s1, -12
+; RV32-SOFT-NEXT: .cfi_offset s2, -16
+; RV32-SOFT-NEXT: .cfi_offset s3, -20
+; RV32-SOFT-NEXT: .cfi_offset s4, -24
+; RV32-SOFT-NEXT: .cfi_offset s5, -28
+; RV32-SOFT-NEXT: .cfi_offset s6, -32
+; RV32-SOFT-NEXT: .cfi_offset s7, -36
+; RV32-SOFT-NEXT: .cfi_offset s8, -40
+; RV32-SOFT-NEXT: lhu s7, 16(a1)
+; RV32-SOFT-NEXT: lhu s5, 20(a1)
+; RV32-SOFT-NEXT: lhu s3, 24(a1)
+; RV32-SOFT-NEXT: lhu s1, 28(a1)
+; RV32-SOFT-NEXT: lhu a2, 0(a1)
+; RV32-SOFT-NEXT: lhu s4, 4(a1)
+; RV32-SOFT-NEXT: lhu s6, 8(a1)
+; RV32-SOFT-NEXT: lhu s8, 12(a1)
+; RV32-SOFT-NEXT: mv s0, a0
+; RV32-SOFT-NEXT: mv a0, a2
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s2, a0
+; RV32-SOFT-NEXT: mv a0, s4
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s4, a0
+; RV32-SOFT-NEXT: mv a0, s6
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s6, a0
+; RV32-SOFT-NEXT: mv a0, s8
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s8, a0
+; RV32-SOFT-NEXT: mv a0, s7
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s7, a0
+; RV32-SOFT-NEXT: mv a0, s5
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s5, a0
+; RV32-SOFT-NEXT: mv a0, s3
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: mv s3, a0
+; RV32-SOFT-NEXT: mv a0, s1
+; RV32-SOFT-NEXT: call __extendhfsf2
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: call __truncsfhf2
+; RV32-SOFT-NEXT: sh s7, 8(s0)
+; RV32-SOFT-NEXT: sh s5, 10(s0)
+; RV32-SOFT-NEXT: sh s3, 12(s0)
+; RV32-SOFT-NEXT: sh a0, 14(s0)
+; RV32-SOFT-NEXT: sh s2, 0(s0)
+; RV32-SOFT-NEXT: sh s4, 2(s0)
+; RV32-SOFT-NEXT: sh s6, 4(s0)
+; RV32-SOFT-NEXT: sh s8, 6(s0)
+; RV32-SOFT-NEXT: lw ra, 44(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s0, 40(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s1, 36(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s2, 32(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s3, 28(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s4, 24(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s5, 20(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s6, 16(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s7, 12(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s8, 8(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: .cfi_restore ra
+; RV32-SOFT-NEXT: .cfi_restore s0
+; RV32-SOFT-NEXT: .cfi_restore s1
+; RV32-SOFT-NEXT: .cfi_restore s2
+; RV32-SOFT-NEXT: .cfi_restore s3
+; RV32-SOFT-NEXT: .cfi_restore s4
+; RV32-SOFT-NEXT: .cfi_restore s5
+; RV32-SOFT-NEXT: .cfi_restore s6
+; RV32-SOFT-NEXT: .cfi_restore s7
+; RV32-SOFT-NEXT: .cfi_restore s8
+; RV32-SOFT-NEXT: addi sp, sp, 48
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV32-SOFT-NEXT: ret
%z = call nnan <8 x half> @llvm.canonicalize.v8f16(<8 x half> %x)
ret <8 x half> %z
}
define float @fcanonicalize_f32(float %x) {
-; CHECK-LABEL: fcanonicalize_f32:
-; CHECK: # %bb.0:
-; CHECK-NEXT: fmin.s fa0, fa0, fa0
-; CHECK-NEXT: ret
+; RV64-SOFT-LABEL: fcanonicalize_f32:
+; RV64-SOFT: # %bb.0:
+; RV64-SOFT-NEXT: addi sp, sp, -16
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 16
+; RV64-SOFT-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: .cfi_offset ra, -8
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: .cfi_restore ra
+; RV64-SOFT-NEXT: addi sp, sp, 16
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV64-SOFT-NEXT: ret
+;
+; CHECK-SOFT-RV64-LABEL: fcanonicalize_f32:
+; CHECK-SOFT-RV64: # %bb.0:
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, -16
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 16
+; CHECK-SOFT-RV64-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: .cfi_offset ra, -8
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, 16
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV64-NEXT: ret
+;
+; CHECK-FP16-RV64-LABEL: fcanonicalize_f32:
+; CHECK-FP16-RV64: # %bb.0:
+; CHECK-FP16-RV64-NEXT: fmin.s fa0, fa0, fa0
+; CHECK-FP16-RV64-NEXT: ret
+;
+; CHECK-NOFP16-RV64-LABEL: fcanonicalize_f32:
+; CHECK-NOFP16-RV64: # %bb.0:
+; CHECK-NOFP16-RV64-NEXT: fmin.s fa0, fa0, fa0
+; CHECK-NOFP16-RV64-NEXT: ret
+;
+; CHECK-SOFT-RV32-LABEL: fcanonicalize_f32:
+; CHECK-SOFT-RV32: # %bb.0:
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, -16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 16
+; CHECK-SOFT-RV32-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: .cfi_offset ra, -4
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, 16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV32-NEXT: ret
+;
+; CHECK-FP16-RV32-LABEL: fcanonicalize_f32:
+; CHECK-FP16-RV32: # %bb.0:
+; CHECK-FP16-RV32-NEXT: fmin.s fa0, fa0, fa0
+; CHECK-FP16-RV32-NEXT: ret
+;
+; CHECK-NOFP16-RV32-LABEL: fcanonicalize_f32:
+; CHECK-NOFP16-RV32: # %bb.0:
+; CHECK-NOFP16-RV32-NEXT: fmin.s fa0, fa0, fa0
+; CHECK-NOFP16-RV32-NEXT: ret
+; RV32-SOFT-LABEL: fcanonicalize_f32:
+; RV32-SOFT: # %bb.0:
+; RV32-SOFT-NEXT: addi sp, sp, -16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 16
+; RV32-SOFT-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: .cfi_offset ra, -4
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: .cfi_restore ra
+; RV32-SOFT-NEXT: addi sp, sp, 16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV32-SOFT-NEXT: ret
%z = call float @llvm.canonicalize.f32(float %x)
ret float %z
}
define float @fcanonicalize_f32_nnan(float %x) {
-; CHECK-LABEL: fcanonicalize_f32_nnan:
-; CHECK: # %bb.0:
-; CHECK-NEXT: fmin.s fa0, fa0, fa0
-; CHECK-NEXT: ret
+; RV64-SOFT-LABEL: fcanonicalize_f32_nnan:
+; RV64-SOFT: # %bb.0:
+; RV64-SOFT-NEXT: addi sp, sp, -16
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 16
+; RV64-SOFT-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: .cfi_offset ra, -8
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: .cfi_restore ra
+; RV64-SOFT-NEXT: addi sp, sp, 16
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV64-SOFT-NEXT: ret
+;
+; CHECK-SOFT-RV64-LABEL: fcanonicalize_f32_nnan:
+; CHECK-SOFT-RV64: # %bb.0:
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, -16
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 16
+; CHECK-SOFT-RV64-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: .cfi_offset ra, -8
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, 16
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV64-NEXT: ret
+;
+; CHECK-FP16-RV64-LABEL: fcanonicalize_f32_nnan:
+; CHECK-FP16-RV64: # %bb.0:
+; CHECK-FP16-RV64-NEXT: fmin.s fa0, fa0, fa0
+; CHECK-FP16-RV64-NEXT: ret
+;
+; CHECK-NOFP16-RV64-LABEL: fcanonicalize_f32_nnan:
+; CHECK-NOFP16-RV64: # %bb.0:
+; CHECK-NOFP16-RV64-NEXT: fmin.s fa0, fa0, fa0
+; CHECK-NOFP16-RV64-NEXT: ret
+;
+; CHECK-SOFT-RV32-LABEL: fcanonicalize_f32_nnan:
+; CHECK-SOFT-RV32: # %bb.0:
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, -16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 16
+; CHECK-SOFT-RV32-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: .cfi_offset ra, -4
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, 16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV32-NEXT: ret
+;
+; CHECK-FP16-RV32-LABEL: fcanonicalize_f32_nnan:
+; CHECK-FP16-RV32: # %bb.0:
+; CHECK-FP16-RV32-NEXT: fmin.s fa0, fa0, fa0
+; CHECK-FP16-RV32-NEXT: ret
+;
+; CHECK-NOFP16-RV32-LABEL: fcanonicalize_f32_nnan:
+; CHECK-NOFP16-RV32: # %bb.0:
+; CHECK-NOFP16-RV32-NEXT: fmin.s fa0, fa0, fa0
+; CHECK-NOFP16-RV32-NEXT: ret
+; RV32-SOFT-LABEL: fcanonicalize_f32_nnan:
+; RV32-SOFT: # %bb.0:
+; RV32-SOFT-NEXT: addi sp, sp, -16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 16
+; RV32-SOFT-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: .cfi_offset ra, -4
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: .cfi_restore ra
+; RV32-SOFT-NEXT: addi sp, sp, 16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV32-SOFT-NEXT: ret
%z = call nnan float @llvm.canonicalize.f32(float %x)
ret float %z
}
define <2 x float> @fcanonicalize_v2f32(<2 x float> %x) {
-; CHECK-LABEL: fcanonicalize_v2f32:
-; CHECK: # %bb.0:
-; CHECK-NEXT: fmin.s fa0, fa0, fa0
-; CHECK-NEXT: fmin.s fa1, fa1, fa1
-; CHECK-NEXT: ret
+; RV64-SOFT-LABEL: fcanonicalize_v2f32:
+; RV64-SOFT: # %bb.0:
+; RV64-SOFT-NEXT: addi sp, sp, -32
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 32
+; RV64-SOFT-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: .cfi_offset ra, -8
+; RV64-SOFT-NEXT: .cfi_offset s0, -16
+; RV64-SOFT-NEXT: .cfi_offset s1, -24
+; RV64-SOFT-NEXT: mv s0, a1
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: mv s1, a0
+; RV64-SOFT-NEXT: mv a0, s0
+; RV64-SOFT-NEXT: mv a1, s0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: mv a0, s1
+; RV64-SOFT-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: .cfi_restore ra
+; RV64-SOFT-NEXT: .cfi_restore s0
+; RV64-SOFT-NEXT: .cfi_restore s1
+; RV64-SOFT-NEXT: addi sp, sp, 32
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV64-SOFT-NEXT: ret
+;
+; CHECK-SOFT-RV64-LABEL: fcanonicalize_v2f32:
+; CHECK-SOFT-RV64: # %bb.0:
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, -32
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 32
+; CHECK-SOFT-RV64-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: .cfi_offset ra, -8
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s0, -16
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s1, -24
+; CHECK-SOFT-RV64-NEXT: mv s0, a1
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: mv s1, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s0
+; CHECK-SOFT-RV64-NEXT: mv a1, s0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s1
+; CHECK-SOFT-RV64-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, 32
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV64-NEXT: ret
+;
+; CHECK-FP16-RV64-LABEL: fcanonicalize_v2f32:
+; CHECK-FP16-RV64: # %bb.0:
+; CHECK-FP16-RV64-NEXT: fmin.s fa0, fa0, fa0
+; CHECK-FP16-RV64-NEXT: fmin.s fa1, fa1, fa1
+; CHECK-FP16-RV64-NEXT: ret
+;
+; CHECK-NOFP16-RV64-LABEL: fcanonicalize_v2f32:
+; CHECK-NOFP16-RV64: # %bb.0:
+; CHECK-NOFP16-RV64-NEXT: fmin.s fa0, fa0, fa0
+; CHECK-NOFP16-RV64-NEXT: fmin.s fa1, fa1, fa1
+; CHECK-NOFP16-RV64-NEXT: ret
+;
+; CHECK-SOFT-RV32-LABEL: fcanonicalize_v2f32:
+; CHECK-SOFT-RV32: # %bb.0:
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, -16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 16
+; CHECK-SOFT-RV32-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s1, 4(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: .cfi_offset ra, -4
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s0, -8
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s1, -12
+; CHECK-SOFT-RV32-NEXT: mv s0, a1
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: mv s1, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s0
+; CHECK-SOFT-RV32-NEXT: mv a1, s0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s1
+; CHECK-SOFT-RV32-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s1, 4(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, 16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV32-NEXT: ret
+;
+; CHECK-FP16-RV32-LABEL: fcanonicalize_v2f32:
+; CHECK-FP16-RV32: # %bb.0:
+; CHECK-FP16-RV32-NEXT: fmin.s fa0, fa0, fa0
+; CHECK-FP16-RV32-NEXT: fmin.s fa1, fa1, fa1
+; CHECK-FP16-RV32-NEXT: ret
+;
+; CHECK-NOFP16-RV32-LABEL: fcanonicalize_v2f32:
+; CHECK-NOFP16-RV32: # %bb.0:
+; CHECK-NOFP16-RV32-NEXT: fmin.s fa0, fa0, fa0
+; CHECK-NOFP16-RV32-NEXT: fmin.s fa1, fa1, fa1
+; CHECK-NOFP16-RV32-NEXT: ret
+; RV32-SOFT-LABEL: fcanonicalize_v2f32:
+; RV32-SOFT: # %bb.0:
+; RV32-SOFT-NEXT: addi sp, sp, -16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 16
+; RV32-SOFT-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s1, 4(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: .cfi_offset ra, -4
+; RV32-SOFT-NEXT: .cfi_offset s0, -8
+; RV32-SOFT-NEXT: .cfi_offset s1, -12
+; RV32-SOFT-NEXT: mv s0, a1
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: mv s1, a0
+; RV32-SOFT-NEXT: mv a0, s0
+; RV32-SOFT-NEXT: mv a1, s0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: mv a0, s1
+; RV32-SOFT-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s1, 4(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: .cfi_restore ra
+; RV32-SOFT-NEXT: .cfi_restore s0
+; RV32-SOFT-NEXT: .cfi_restore s1
+; RV32-SOFT-NEXT: addi sp, sp, 16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV32-SOFT-NEXT: ret
%z = call <2 x float> @llvm.canonicalize.v2f32(<2 x float> %x)
ret <2 x float> %z
}
define <2 x float> @fcanonicalize_v2f32_nnan(<2 x float> %x) {
-; CHECK-LABEL: fcanonicalize_v2f32_nnan:
-; CHECK: # %bb.0:
-; CHECK-NEXT: fmin.s fa0, fa0, fa0
-; CHECK-NEXT: fmin.s fa1, fa1, fa1
-; CHECK-NEXT: ret
+; RV64-SOFT-LABEL: fcanonicalize_v2f32_nnan:
+; RV64-SOFT: # %bb.0:
+; RV64-SOFT-NEXT: addi sp, sp, -32
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 32
+; RV64-SOFT-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: .cfi_offset ra, -8
+; RV64-SOFT-NEXT: .cfi_offset s0, -16
+; RV64-SOFT-NEXT: .cfi_offset s1, -24
+; RV64-SOFT-NEXT: mv s0, a1
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: mv s1, a0
+; RV64-SOFT-NEXT: mv a0, s0
+; RV64-SOFT-NEXT: mv a1, s0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: mv a0, s1
+; RV64-SOFT-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: .cfi_restore ra
+; RV64-SOFT-NEXT: .cfi_restore s0
+; RV64-SOFT-NEXT: .cfi_restore s1
+; RV64-SOFT-NEXT: addi sp, sp, 32
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV64-SOFT-NEXT: ret
+;
+; CHECK-SOFT-RV64-LABEL: fcanonicalize_v2f32_nnan:
+; CHECK-SOFT-RV64: # %bb.0:
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, -32
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 32
+; CHECK-SOFT-RV64-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: .cfi_offset ra, -8
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s0, -16
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s1, -24
+; CHECK-SOFT-RV64-NEXT: mv s0, a1
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: mv s1, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s0
+; CHECK-SOFT-RV64-NEXT: mv a1, s0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s1
+; CHECK-SOFT-RV64-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, 32
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV64-NEXT: ret
+;
+; CHECK-FP16-RV64-LABEL: fcanonicalize_v2f32_nnan:
+; CHECK-FP16-RV64: # %bb.0:
+; CHECK-FP16-RV64-NEXT: fmin.s fa0, fa0, fa0
+; CHECK-FP16-RV64-NEXT: fmin.s fa1, fa1, fa1
+; CHECK-FP16-RV64-NEXT: ret
+;
+; CHECK-NOFP16-RV64-LABEL: fcanonicalize_v2f32_nnan:
+; CHECK-NOFP16-RV64: # %bb.0:
+; CHECK-NOFP16-RV64-NEXT: fmin.s fa0, fa0, fa0
+; CHECK-NOFP16-RV64-NEXT: fmin.s fa1, fa1, fa1
+; CHECK-NOFP16-RV64-NEXT: ret
+;
+; CHECK-SOFT-RV32-LABEL: fcanonicalize_v2f32_nnan:
+; CHECK-SOFT-RV32: # %bb.0:
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, -16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 16
+; CHECK-SOFT-RV32-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s1, 4(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: .cfi_offset ra, -4
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s0, -8
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s1, -12
+; CHECK-SOFT-RV32-NEXT: mv s0, a1
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: mv s1, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s0
+; CHECK-SOFT-RV32-NEXT: mv a1, s0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: mv a1, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s1
+; CHECK-SOFT-RV32-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s1, 4(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, 16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV32-NEXT: ret
+;
+; CHECK-FP16-RV32-LABEL: fcanonicalize_v2f32_nnan:
+; CHECK-FP16-RV32: # %bb.0:
+; CHECK-FP16-RV32-NEXT: fmin.s fa0, fa0, fa0
+; CHECK-FP16-RV32-NEXT: fmin.s fa1, fa1, fa1
+; CHECK-FP16-RV32-NEXT: ret
+;
+; CHECK-NOFP16-RV32-LABEL: fcanonicalize_v2f32_nnan:
+; CHECK-NOFP16-RV32: # %bb.0:
+; CHECK-NOFP16-RV32-NEXT: fmin.s fa0, fa0, fa0
+; CHECK-NOFP16-RV32-NEXT: fmin.s fa1, fa1, fa1
+; CHECK-NOFP16-RV32-NEXT: ret
+; RV32-SOFT-LABEL: fcanonicalize_v2f32_nnan:
+; RV32-SOFT: # %bb.0:
+; RV32-SOFT-NEXT: addi sp, sp, -16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 16
+; RV32-SOFT-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s1, 4(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: .cfi_offset ra, -4
+; RV32-SOFT-NEXT: .cfi_offset s0, -8
+; RV32-SOFT-NEXT: .cfi_offset s1, -12
+; RV32-SOFT-NEXT: mv s0, a1
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: mv s1, a0
+; RV32-SOFT-NEXT: mv a0, s0
+; RV32-SOFT-NEXT: mv a1, s0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: mv a1, a0
+; RV32-SOFT-NEXT: mv a0, s1
+; RV32-SOFT-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s1, 4(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: .cfi_restore ra
+; RV32-SOFT-NEXT: .cfi_restore s0
+; RV32-SOFT-NEXT: .cfi_restore s1
+; RV32-SOFT-NEXT: addi sp, sp, 16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV32-SOFT-NEXT: ret
%z = call nnan <2 x float> @llvm.canonicalize.v2f32(<2 x float> %x)
ret <2 x float> %z
}
define <4 x float> @fcanonicalize_v4f32(<4 x float> %x) {
-; CHECK-LABEL: fcanonicalize_v4f32:
-; CHECK: # %bb.0:
-; CHECK-NEXT: fmin.s fa5, fa0, fa0
-; CHECK-NEXT: fmin.s fa4, fa1, fa1
-; CHECK-NEXT: fmin.s fa2, fa2, fa2
-; CHECK-NEXT: fmin.s fa3, fa3, fa3
-; CHECK-NEXT: fsw fa5, 0(a0)
-; CHECK-NEXT: fsw fa4, 4(a0)
-; CHECK-NEXT: fsw fa2, 8(a0)
-; CHECK-NEXT: fsw fa3, 12(a0)
-; CHECK-NEXT: ret
+; RV64-SOFT-LABEL: fcanonicalize_v4f32:
+; RV64-SOFT: # %bb.0:
+; RV64-SOFT-NEXT: addi sp, sp, -48
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 48
+; RV64-SOFT-NEXT: sd ra, 40(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s0, 32(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s1, 24(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s2, 16(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s3, 8(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s4, 0(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: .cfi_offset ra, -8
+; RV64-SOFT-NEXT: .cfi_offset s0, -16
+; RV64-SOFT-NEXT: .cfi_offset s1, -24
+; RV64-SOFT-NEXT: .cfi_offset s2, -32
+; RV64-SOFT-NEXT: .cfi_offset s3, -40
+; RV64-SOFT-NEXT: .cfi_offset s4, -48
+; RV64-SOFT-NEXT: lw a2, 0(a1)
+; RV64-SOFT-NEXT: lw s0, 8(a1)
+; RV64-SOFT-NEXT: lw s1, 16(a1)
+; RV64-SOFT-NEXT: lw s2, 24(a1)
+; RV64-SOFT-NEXT: mv s3, a0
+; RV64-SOFT-NEXT: mv a0, a2
+; RV64-SOFT-NEXT: mv a1, a2
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: mv s4, a0
+; RV64-SOFT-NEXT: mv a0, s0
+; RV64-SOFT-NEXT: mv a1, s0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: mv s0, a0
+; RV64-SOFT-NEXT: mv a0, s1
+; RV64-SOFT-NEXT: mv a1, s1
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: mv s1, a0
+; RV64-SOFT-NEXT: mv a0, s2
+; RV64-SOFT-NEXT: mv a1, s2
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: sw s4, 0(s3)
+; RV64-SOFT-NEXT: sw s0, 4(s3)
+; RV64-SOFT-NEXT: sw s1, 8(s3)
+; RV64-SOFT-NEXT: sw a0, 12(s3)
+; RV64-SOFT-NEXT: ld ra, 40(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s0, 32(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s1, 24(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s2, 16(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s3, 8(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s4, 0(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: .cfi_restore ra
+; RV64-SOFT-NEXT: .cfi_restore s0
+; RV64-SOFT-NEXT: .cfi_restore s1
+; RV64-SOFT-NEXT: .cfi_restore s2
+; RV64-SOFT-NEXT: .cfi_restore s3
+; RV64-SOFT-NEXT: .cfi_restore s4
+; RV64-SOFT-NEXT: addi sp, sp, 48
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV64-SOFT-NEXT: ret
+;
+; CHECK-SOFT-RV64-LABEL: fcanonicalize_v4f32:
+; CHECK-SOFT-RV64: # %bb.0:
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, -48
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 48
+; CHECK-SOFT-RV64-NEXT: sd ra, 40(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s0, 32(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s1, 24(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s2, 16(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s3, 8(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s4, 0(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: .cfi_offset ra, -8
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s0, -16
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s1, -24
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s2, -32
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s3, -40
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s4, -48
+; CHECK-SOFT-RV64-NEXT: lw a2, 0(a1)
+; CHECK-SOFT-RV64-NEXT: lw s0, 8(a1)
+; CHECK-SOFT-RV64-NEXT: lw s1, 16(a1)
+; CHECK-SOFT-RV64-NEXT: lw s2, 24(a1)
+; CHECK-SOFT-RV64-NEXT: mv s3, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, a2
+; CHECK-SOFT-RV64-NEXT: mv a1, a2
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: mv s4, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s0
+; CHECK-SOFT-RV64-NEXT: mv a1, s0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: mv s0, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s1
+; CHECK-SOFT-RV64-NEXT: mv a1, s1
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: mv s1, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s2
+; CHECK-SOFT-RV64-NEXT: mv a1, s2
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: sw s4, 0(s3)
+; CHECK-SOFT-RV64-NEXT: sw s0, 4(s3)
+; CHECK-SOFT-RV64-NEXT: sw s1, 8(s3)
+; CHECK-SOFT-RV64-NEXT: sw a0, 12(s3)
+; CHECK-SOFT-RV64-NEXT: ld ra, 40(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s0, 32(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s1, 24(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s2, 16(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s3, 8(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s4, 0(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s3
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s4
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, 48
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV64-NEXT: ret
+;
+; CHECK-FP16-RV64-LABEL: fcanonicalize_v4f32:
+; CHECK-FP16-RV64: # %bb.0:
+; CHECK-FP16-RV64-NEXT: fmin.s fa5, fa0, fa0
+; CHECK-FP16-RV64-NEXT: fmin.s fa4, fa1, fa1
+; CHECK-FP16-RV64-NEXT: fmin.s fa2, fa2, fa2
+; CHECK-FP16-RV64-NEXT: fmin.s fa3, fa3, fa3
+; CHECK-FP16-RV64-NEXT: fsw fa5, 0(a0)
+; CHECK-FP16-RV64-NEXT: fsw fa4, 4(a0)
+; CHECK-FP16-RV64-NEXT: fsw fa2, 8(a0)
+; CHECK-FP16-RV64-NEXT: fsw fa3, 12(a0)
+; CHECK-FP16-RV64-NEXT: ret
+;
+; CHECK-NOFP16-RV64-LABEL: fcanonicalize_v4f32:
+; CHECK-NOFP16-RV64: # %bb.0:
+; CHECK-NOFP16-RV64-NEXT: fmin.s fa5, fa0, fa0
+; CHECK-NOFP16-RV64-NEXT: fmin.s fa4, fa1, fa1
+; CHECK-NOFP16-RV64-NEXT: fmin.s fa2, fa2, fa2
+; CHECK-NOFP16-RV64-NEXT: fmin.s fa3, fa3, fa3
+; CHECK-NOFP16-RV64-NEXT: fsw fa5, 0(a0)
+; CHECK-NOFP16-RV64-NEXT: fsw fa4, 4(a0)
+; CHECK-NOFP16-RV64-NEXT: fsw fa2, 8(a0)
+; CHECK-NOFP16-RV64-NEXT: fsw fa3, 12(a0)
+; CHECK-NOFP16-RV64-NEXT: ret
+;
+; CHECK-SOFT-RV32-LABEL: fcanonicalize_v4f32:
+; CHECK-SOFT-RV32: # %bb.0:
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, -32
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 32
+; CHECK-SOFT-RV32-NEXT: sw ra, 28(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s0, 24(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s1, 20(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s2, 16(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s3, 12(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s4, 8(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: .cfi_offset ra, -4
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s0, -8
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s1, -12
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s2, -16
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s3, -20
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s4, -24
+; CHECK-SOFT-RV32-NEXT: lw a2, 0(a1)
+; CHECK-SOFT-RV32-NEXT: lw s0, 4(a1)
+; CHECK-SOFT-RV32-NEXT: lw s1, 8(a1)
+; CHECK-SOFT-RV32-NEXT: lw s2, 12(a1)
+; CHECK-SOFT-RV32-NEXT: mv s3, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, a2
+; CHECK-SOFT-RV32-NEXT: mv a1, a2
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: mv s4, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s0
+; CHECK-SOFT-RV32-NEXT: mv a1, s0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: mv s0, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s1
+; CHECK-SOFT-RV32-NEXT: mv a1, s1
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: mv s1, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s2
+; CHECK-SOFT-RV32-NEXT: mv a1, s2
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: sw s4, 0(s3)
+; CHECK-SOFT-RV32-NEXT: sw s0, 4(s3)
+; CHECK-SOFT-RV32-NEXT: sw s1, 8(s3)
+; CHECK-SOFT-RV32-NEXT: sw a0, 12(s3)
+; CHECK-SOFT-RV32-NEXT: lw ra, 28(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s0, 24(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s1, 20(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s2, 16(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s3, 12(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s4, 8(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s3
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s4
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, 32
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV32-NEXT: ret
+;
+; CHECK-FP16-RV32-LABEL: fcanonicalize_v4f32:
+; CHECK-FP16-RV32: # %bb.0:
+; CHECK-FP16-RV32-NEXT: fmin.s fa5, fa0, fa0
+; CHECK-FP16-RV32-NEXT: fmin.s fa4, fa1, fa1
+; CHECK-FP16-RV32-NEXT: fmin.s fa2, fa2, fa2
+; CHECK-FP16-RV32-NEXT: fmin.s fa3, fa3, fa3
+; CHECK-FP16-RV32-NEXT: fsw fa5, 0(a0)
+; CHECK-FP16-RV32-NEXT: fsw fa4, 4(a0)
+; CHECK-FP16-RV32-NEXT: fsw fa2, 8(a0)
+; CHECK-FP16-RV32-NEXT: fsw fa3, 12(a0)
+; CHECK-FP16-RV32-NEXT: ret
+;
+; CHECK-NOFP16-RV32-LABEL: fcanonicalize_v4f32:
+; CHECK-NOFP16-RV32: # %bb.0:
+; CHECK-NOFP16-RV32-NEXT: fmin.s fa5, fa0, fa0
+; CHECK-NOFP16-RV32-NEXT: fmin.s fa4, fa1, fa1
+; CHECK-NOFP16-RV32-NEXT: fmin.s fa2, fa2, fa2
+; CHECK-NOFP16-RV32-NEXT: fmin.s fa3, fa3, fa3
+; CHECK-NOFP16-RV32-NEXT: fsw fa5, 0(a0)
+; CHECK-NOFP16-RV32-NEXT: fsw fa4, 4(a0)
+; CHECK-NOFP16-RV32-NEXT: fsw fa2, 8(a0)
+; CHECK-NOFP16-RV32-NEXT: fsw fa3, 12(a0)
+; CHECK-NOFP16-RV32-NEXT: ret
+; RV32-SOFT-LABEL: fcanonicalize_v4f32:
+; RV32-SOFT: # %bb.0:
+; RV32-SOFT-NEXT: addi sp, sp, -32
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 32
+; RV32-SOFT-NEXT: sw ra, 28(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s0, 24(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s1, 20(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s2, 16(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s3, 12(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s4, 8(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: .cfi_offset ra, -4
+; RV32-SOFT-NEXT: .cfi_offset s0, -8
+; RV32-SOFT-NEXT: .cfi_offset s1, -12
+; RV32-SOFT-NEXT: .cfi_offset s2, -16
+; RV32-SOFT-NEXT: .cfi_offset s3, -20
+; RV32-SOFT-NEXT: .cfi_offset s4, -24
+; RV32-SOFT-NEXT: lw a2, 0(a1)
+; RV32-SOFT-NEXT: lw s0, 4(a1)
+; RV32-SOFT-NEXT: lw s1, 8(a1)
+; RV32-SOFT-NEXT: lw s2, 12(a1)
+; RV32-SOFT-NEXT: mv s3, a0
+; RV32-SOFT-NEXT: mv a0, a2
+; RV32-SOFT-NEXT: mv a1, a2
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: mv s4, a0
+; RV32-SOFT-NEXT: mv a0, s0
+; RV32-SOFT-NEXT: mv a1, s0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: mv s0, a0
+; RV32-SOFT-NEXT: mv a0, s1
+; RV32-SOFT-NEXT: mv a1, s1
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: mv s1, a0
+; RV32-SOFT-NEXT: mv a0, s2
+; RV32-SOFT-NEXT: mv a1, s2
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: sw s4, 0(s3)
+; RV32-SOFT-NEXT: sw s0, 4(s3)
+; RV32-SOFT-NEXT: sw s1, 8(s3)
+; RV32-SOFT-NEXT: sw a0, 12(s3)
+; RV32-SOFT-NEXT: lw ra, 28(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s0, 24(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s1, 20(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s2, 16(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s3, 12(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s4, 8(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: .cfi_restore ra
+; RV32-SOFT-NEXT: .cfi_restore s0
+; RV32-SOFT-NEXT: .cfi_restore s1
+; RV32-SOFT-NEXT: .cfi_restore s2
+; RV32-SOFT-NEXT: .cfi_restore s3
+; RV32-SOFT-NEXT: .cfi_restore s4
+; RV32-SOFT-NEXT: addi sp, sp, 32
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV32-SOFT-NEXT: ret
%z = call <4 x float> @llvm.canonicalize.v4f32(<4 x float> %x)
ret <4 x float> %z
}
define <4 x float> @fcanonicalize_v4f32_nnan(<4 x float> %x) {
-; CHECK-LABEL: fcanonicalize_v4f32_nnan:
-; CHECK: # %bb.0:
-; CHECK-NEXT: fmin.s fa5, fa0, fa0
-; CHECK-NEXT: fmin.s fa4, fa1, fa1
-; CHECK-NEXT: fmin.s fa2, fa2, fa2
-; CHECK-NEXT: fmin.s fa3, fa3, fa3
-; CHECK-NEXT: fsw fa5, 0(a0)
-; CHECK-NEXT: fsw fa4, 4(a0)
-; CHECK-NEXT: fsw fa2, 8(a0)
-; CHECK-NEXT: fsw fa3, 12(a0)
-; CHECK-NEXT: ret
+; RV64-SOFT-LABEL: fcanonicalize_v4f32_nnan:
+; RV64-SOFT: # %bb.0:
+; RV64-SOFT-NEXT: addi sp, sp, -48
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 48
+; RV64-SOFT-NEXT: sd ra, 40(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s0, 32(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s1, 24(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s2, 16(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s3, 8(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s4, 0(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: .cfi_offset ra, -8
+; RV64-SOFT-NEXT: .cfi_offset s0, -16
+; RV64-SOFT-NEXT: .cfi_offset s1, -24
+; RV64-SOFT-NEXT: .cfi_offset s2, -32
+; RV64-SOFT-NEXT: .cfi_offset s3, -40
+; RV64-SOFT-NEXT: .cfi_offset s4, -48
+; RV64-SOFT-NEXT: lw a2, 0(a1)
+; RV64-SOFT-NEXT: lw s0, 8(a1)
+; RV64-SOFT-NEXT: lw s1, 16(a1)
+; RV64-SOFT-NEXT: lw s2, 24(a1)
+; RV64-SOFT-NEXT: mv s3, a0
+; RV64-SOFT-NEXT: mv a0, a2
+; RV64-SOFT-NEXT: mv a1, a2
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: mv s4, a0
+; RV64-SOFT-NEXT: mv a0, s0
+; RV64-SOFT-NEXT: mv a1, s0
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: mv s0, a0
+; RV64-SOFT-NEXT: mv a0, s1
+; RV64-SOFT-NEXT: mv a1, s1
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: mv s1, a0
+; RV64-SOFT-NEXT: mv a0, s2
+; RV64-SOFT-NEXT: mv a1, s2
+; RV64-SOFT-NEXT: call fminimum_numf
+; RV64-SOFT-NEXT: sw s4, 0(s3)
+; RV64-SOFT-NEXT: sw s0, 4(s3)
+; RV64-SOFT-NEXT: sw s1, 8(s3)
+; RV64-SOFT-NEXT: sw a0, 12(s3)
+; RV64-SOFT-NEXT: ld ra, 40(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s0, 32(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s1, 24(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s2, 16(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s3, 8(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s4, 0(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: .cfi_restore ra
+; RV64-SOFT-NEXT: .cfi_restore s0
+; RV64-SOFT-NEXT: .cfi_restore s1
+; RV64-SOFT-NEXT: .cfi_restore s2
+; RV64-SOFT-NEXT: .cfi_restore s3
+; RV64-SOFT-NEXT: .cfi_restore s4
+; RV64-SOFT-NEXT: addi sp, sp, 48
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV64-SOFT-NEXT: ret
+;
+; CHECK-SOFT-RV64-LABEL: fcanonicalize_v4f32_nnan:
+; CHECK-SOFT-RV64: # %bb.0:
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, -48
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 48
+; CHECK-SOFT-RV64-NEXT: sd ra, 40(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s0, 32(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s1, 24(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s2, 16(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s3, 8(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s4, 0(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: .cfi_offset ra, -8
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s0, -16
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s1, -24
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s2, -32
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s3, -40
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s4, -48
+; CHECK-SOFT-RV64-NEXT: lw a2, 0(a1)
+; CHECK-SOFT-RV64-NEXT: lw s0, 8(a1)
+; CHECK-SOFT-RV64-NEXT: lw s1, 16(a1)
+; CHECK-SOFT-RV64-NEXT: lw s2, 24(a1)
+; CHECK-SOFT-RV64-NEXT: mv s3, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, a2
+; CHECK-SOFT-RV64-NEXT: mv a1, a2
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: mv s4, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s0
+; CHECK-SOFT-RV64-NEXT: mv a1, s0
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: mv s0, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s1
+; CHECK-SOFT-RV64-NEXT: mv a1, s1
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: mv s1, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s2
+; CHECK-SOFT-RV64-NEXT: mv a1, s2
+; CHECK-SOFT-RV64-NEXT: call fminimum_numf
+; CHECK-SOFT-RV64-NEXT: sw s4, 0(s3)
+; CHECK-SOFT-RV64-NEXT: sw s0, 4(s3)
+; CHECK-SOFT-RV64-NEXT: sw s1, 8(s3)
+; CHECK-SOFT-RV64-NEXT: sw a0, 12(s3)
+; CHECK-SOFT-RV64-NEXT: ld ra, 40(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s0, 32(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s1, 24(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s2, 16(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s3, 8(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s4, 0(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s3
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s4
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, 48
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV64-NEXT: ret
+;
+; CHECK-FP16-RV64-LABEL: fcanonicalize_v4f32_nnan:
+; CHECK-FP16-RV64: # %bb.0:
+; CHECK-FP16-RV64-NEXT: fmin.s fa5, fa0, fa0
+; CHECK-FP16-RV64-NEXT: fmin.s fa4, fa1, fa1
+; CHECK-FP16-RV64-NEXT: fmin.s fa2, fa2, fa2
+; CHECK-FP16-RV64-NEXT: fmin.s fa3, fa3, fa3
+; CHECK-FP16-RV64-NEXT: fsw fa5, 0(a0)
+; CHECK-FP16-RV64-NEXT: fsw fa4, 4(a0)
+; CHECK-FP16-RV64-NEXT: fsw fa2, 8(a0)
+; CHECK-FP16-RV64-NEXT: fsw fa3, 12(a0)
+; CHECK-FP16-RV64-NEXT: ret
+;
+; CHECK-NOFP16-RV64-LABEL: fcanonicalize_v4f32_nnan:
+; CHECK-NOFP16-RV64: # %bb.0:
+; CHECK-NOFP16-RV64-NEXT: fmin.s fa5, fa0, fa0
+; CHECK-NOFP16-RV64-NEXT: fmin.s fa4, fa1, fa1
+; CHECK-NOFP16-RV64-NEXT: fmin.s fa2, fa2, fa2
+; CHECK-NOFP16-RV64-NEXT: fmin.s fa3, fa3, fa3
+; CHECK-NOFP16-RV64-NEXT: fsw fa5, 0(a0)
+; CHECK-NOFP16-RV64-NEXT: fsw fa4, 4(a0)
+; CHECK-NOFP16-RV64-NEXT: fsw fa2, 8(a0)
+; CHECK-NOFP16-RV64-NEXT: fsw fa3, 12(a0)
+; CHECK-NOFP16-RV64-NEXT: ret
+;
+; CHECK-SOFT-RV32-LABEL: fcanonicalize_v4f32_nnan:
+; CHECK-SOFT-RV32: # %bb.0:
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, -32
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 32
+; CHECK-SOFT-RV32-NEXT: sw ra, 28(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s0, 24(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s1, 20(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s2, 16(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s3, 12(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s4, 8(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: .cfi_offset ra, -4
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s0, -8
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s1, -12
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s2, -16
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s3, -20
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s4, -24
+; CHECK-SOFT-RV32-NEXT: lw a2, 0(a1)
+; CHECK-SOFT-RV32-NEXT: lw s0, 4(a1)
+; CHECK-SOFT-RV32-NEXT: lw s1, 8(a1)
+; CHECK-SOFT-RV32-NEXT: lw s2, 12(a1)
+; CHECK-SOFT-RV32-NEXT: mv s3, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, a2
+; CHECK-SOFT-RV32-NEXT: mv a1, a2
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: mv s4, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s0
+; CHECK-SOFT-RV32-NEXT: mv a1, s0
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: mv s0, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s1
+; CHECK-SOFT-RV32-NEXT: mv a1, s1
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: mv s1, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, s2
+; CHECK-SOFT-RV32-NEXT: mv a1, s2
+; CHECK-SOFT-RV32-NEXT: call fminimum_numf
+; CHECK-SOFT-RV32-NEXT: sw s4, 0(s3)
+; CHECK-SOFT-RV32-NEXT: sw s0, 4(s3)
+; CHECK-SOFT-RV32-NEXT: sw s1, 8(s3)
+; CHECK-SOFT-RV32-NEXT: sw a0, 12(s3)
+; CHECK-SOFT-RV32-NEXT: lw ra, 28(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s0, 24(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s1, 20(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s2, 16(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s3, 12(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s4, 8(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s3
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s4
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, 32
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV32-NEXT: ret
+;
+; CHECK-FP16-RV32-LABEL: fcanonicalize_v4f32_nnan:
+; CHECK-FP16-RV32: # %bb.0:
+; CHECK-FP16-RV32-NEXT: fmin.s fa5, fa0, fa0
+; CHECK-FP16-RV32-NEXT: fmin.s fa4, fa1, fa1
+; CHECK-FP16-RV32-NEXT: fmin.s fa2, fa2, fa2
+; CHECK-FP16-RV32-NEXT: fmin.s fa3, fa3, fa3
+; CHECK-FP16-RV32-NEXT: fsw fa5, 0(a0)
+; CHECK-FP16-RV32-NEXT: fsw fa4, 4(a0)
+; CHECK-FP16-RV32-NEXT: fsw fa2, 8(a0)
+; CHECK-FP16-RV32-NEXT: fsw fa3, 12(a0)
+; CHECK-FP16-RV32-NEXT: ret
+;
+; CHECK-NOFP16-RV32-LABEL: fcanonicalize_v4f32_nnan:
+; CHECK-NOFP16-RV32: # %bb.0:
+; CHECK-NOFP16-RV32-NEXT: fmin.s fa5, fa0, fa0
+; CHECK-NOFP16-RV32-NEXT: fmin.s fa4, fa1, fa1
+; CHECK-NOFP16-RV32-NEXT: fmin.s fa2, fa2, fa2
+; CHECK-NOFP16-RV32-NEXT: fmin.s fa3, fa3, fa3
+; CHECK-NOFP16-RV32-NEXT: fsw fa5, 0(a0)
+; CHECK-NOFP16-RV32-NEXT: fsw fa4, 4(a0)
+; CHECK-NOFP16-RV32-NEXT: fsw fa2, 8(a0)
+; CHECK-NOFP16-RV32-NEXT: fsw fa3, 12(a0)
+; CHECK-NOFP16-RV32-NEXT: ret
+; RV32-SOFT-LABEL: fcanonicalize_v4f32_nnan:
+; RV32-SOFT: # %bb.0:
+; RV32-SOFT-NEXT: addi sp, sp, -32
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 32
+; RV32-SOFT-NEXT: sw ra, 28(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s0, 24(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s1, 20(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s2, 16(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s3, 12(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s4, 8(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: .cfi_offset ra, -4
+; RV32-SOFT-NEXT: .cfi_offset s0, -8
+; RV32-SOFT-NEXT: .cfi_offset s1, -12
+; RV32-SOFT-NEXT: .cfi_offset s2, -16
+; RV32-SOFT-NEXT: .cfi_offset s3, -20
+; RV32-SOFT-NEXT: .cfi_offset s4, -24
+; RV32-SOFT-NEXT: lw a2, 0(a1)
+; RV32-SOFT-NEXT: lw s0, 4(a1)
+; RV32-SOFT-NEXT: lw s1, 8(a1)
+; RV32-SOFT-NEXT: lw s2, 12(a1)
+; RV32-SOFT-NEXT: mv s3, a0
+; RV32-SOFT-NEXT: mv a0, a2
+; RV32-SOFT-NEXT: mv a1, a2
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: mv s4, a0
+; RV32-SOFT-NEXT: mv a0, s0
+; RV32-SOFT-NEXT: mv a1, s0
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: mv s0, a0
+; RV32-SOFT-NEXT: mv a0, s1
+; RV32-SOFT-NEXT: mv a1, s1
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: mv s1, a0
+; RV32-SOFT-NEXT: mv a0, s2
+; RV32-SOFT-NEXT: mv a1, s2
+; RV32-SOFT-NEXT: call fminimum_numf
+; RV32-SOFT-NEXT: sw s4, 0(s3)
+; RV32-SOFT-NEXT: sw s0, 4(s3)
+; RV32-SOFT-NEXT: sw s1, 8(s3)
+; RV32-SOFT-NEXT: sw a0, 12(s3)
+; RV32-SOFT-NEXT: lw ra, 28(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s0, 24(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s1, 20(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s2, 16(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s3, 12(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s4, 8(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: .cfi_restore ra
+; RV32-SOFT-NEXT: .cfi_restore s0
+; RV32-SOFT-NEXT: .cfi_restore s1
+; RV32-SOFT-NEXT: .cfi_restore s2
+; RV32-SOFT-NEXT: .cfi_restore s3
+; RV32-SOFT-NEXT: .cfi_restore s4
+; RV32-SOFT-NEXT: addi sp, sp, 32
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV32-SOFT-NEXT: ret
%z = call nnan <4 x float> @llvm.canonicalize.v4f32(<4 x float> %x)
ret <4 x float> %z
}
define double @fcanonicalize_f64(double %x) {
-; CHECK-LABEL: fcanonicalize_f64:
-; CHECK: # %bb.0:
-; CHECK-NEXT: fmin.d fa0, fa0, fa0
-; CHECK-NEXT: ret
+; RV64-SOFT-LABEL: fcanonicalize_f64:
+; RV64-SOFT: # %bb.0:
+; RV64-SOFT-NEXT: addi sp, sp, -16
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 16
+; RV64-SOFT-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: .cfi_offset ra, -8
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_num
+; RV64-SOFT-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: .cfi_restore ra
+; RV64-SOFT-NEXT: addi sp, sp, 16
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV64-SOFT-NEXT: ret
+;
+; CHECK-SOFT-RV64-LABEL: fcanonicalize_f64:
+; CHECK-SOFT-RV64: # %bb.0:
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, -16
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 16
+; CHECK-SOFT-RV64-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: .cfi_offset ra, -8
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_num
+; CHECK-SOFT-RV64-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, 16
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV64-NEXT: ret
+;
+; CHECK-FP16-RV64-LABEL: fcanonicalize_f64:
+; CHECK-FP16-RV64: # %bb.0:
+; CHECK-FP16-RV64-NEXT: fmin.d fa0, fa0, fa0
+; CHECK-FP16-RV64-NEXT: ret
+;
+; CHECK-NOFP16-RV64-LABEL: fcanonicalize_f64:
+; CHECK-NOFP16-RV64: # %bb.0:
+; CHECK-NOFP16-RV64-NEXT: fmin.d fa0, fa0, fa0
+; CHECK-NOFP16-RV64-NEXT: ret
+;
+; CHECK-SOFT-RV32-LABEL: fcanonicalize_f64:
+; CHECK-SOFT-RV32: # %bb.0:
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, -16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 16
+; CHECK-SOFT-RV32-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: .cfi_offset ra, -4
+; CHECK-SOFT-RV32-NEXT: mv a2, a0
+; CHECK-SOFT-RV32-NEXT: mv a3, a1
+; CHECK-SOFT-RV32-NEXT: call fminimum_num
+; CHECK-SOFT-RV32-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, 16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV32-NEXT: ret
+;
+; CHECK-FP16-RV32-LABEL: fcanonicalize_f64:
+; CHECK-FP16-RV32: # %bb.0:
+; CHECK-FP16-RV32-NEXT: fmin.d fa0, fa0, fa0
+; CHECK-FP16-RV32-NEXT: ret
+;
+; CHECK-NOFP16-RV32-LABEL: fcanonicalize_f64:
+; CHECK-NOFP16-RV32: # %bb.0:
+; CHECK-NOFP16-RV32-NEXT: fmin.d fa0, fa0, fa0
+; CHECK-NOFP16-RV32-NEXT: ret
+; RV32-SOFT-LABEL: fcanonicalize_f64:
+; RV32-SOFT: # %bb.0:
+; RV32-SOFT-NEXT: addi sp, sp, -16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 16
+; RV32-SOFT-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: .cfi_offset ra, -4
+; RV32-SOFT-NEXT: mv a2, a0
+; RV32-SOFT-NEXT: mv a3, a1
+; RV32-SOFT-NEXT: call fminimum_num
+; RV32-SOFT-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: .cfi_restore ra
+; RV32-SOFT-NEXT: addi sp, sp, 16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV32-SOFT-NEXT: ret
%z = call double @llvm.canonicalize.f64(double %x)
ret double %z
}
define double @fcanonicalize_f64_nnan(double %x) {
-; CHECK-LABEL: fcanonicalize_f64_nnan:
-; CHECK: # %bb.0:
-; CHECK-NEXT: fmin.d fa0, fa0, fa0
-; CHECK-NEXT: ret
+; RV64-SOFT-LABEL: fcanonicalize_f64_nnan:
+; RV64-SOFT: # %bb.0:
+; RV64-SOFT-NEXT: addi sp, sp, -16
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 16
+; RV64-SOFT-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: .cfi_offset ra, -8
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_num
+; RV64-SOFT-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: .cfi_restore ra
+; RV64-SOFT-NEXT: addi sp, sp, 16
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV64-SOFT-NEXT: ret
+;
+; CHECK-SOFT-RV64-LABEL: fcanonicalize_f64_nnan:
+; CHECK-SOFT-RV64: # %bb.0:
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, -16
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 16
+; CHECK-SOFT-RV64-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: .cfi_offset ra, -8
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_num
+; CHECK-SOFT-RV64-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, 16
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV64-NEXT: ret
+;
+; CHECK-FP16-RV64-LABEL: fcanonicalize_f64_nnan:
+; CHECK-FP16-RV64: # %bb.0:
+; CHECK-FP16-RV64-NEXT: fmin.d fa0, fa0, fa0
+; CHECK-FP16-RV64-NEXT: ret
+;
+; CHECK-NOFP16-RV64-LABEL: fcanonicalize_f64_nnan:
+; CHECK-NOFP16-RV64: # %bb.0:
+; CHECK-NOFP16-RV64-NEXT: fmin.d fa0, fa0, fa0
+; CHECK-NOFP16-RV64-NEXT: ret
+;
+; CHECK-SOFT-RV32-LABEL: fcanonicalize_f64_nnan:
+; CHECK-SOFT-RV32: # %bb.0:
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, -16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 16
+; CHECK-SOFT-RV32-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: .cfi_offset ra, -4
+; CHECK-SOFT-RV32-NEXT: mv a2, a0
+; CHECK-SOFT-RV32-NEXT: mv a3, a1
+; CHECK-SOFT-RV32-NEXT: call fminimum_num
+; CHECK-SOFT-RV32-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, 16
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV32-NEXT: ret
+;
+; CHECK-FP16-RV32-LABEL: fcanonicalize_f64_nnan:
+; CHECK-FP16-RV32: # %bb.0:
+; CHECK-FP16-RV32-NEXT: fmin.d fa0, fa0, fa0
+; CHECK-FP16-RV32-NEXT: ret
+;
+; CHECK-NOFP16-RV32-LABEL: fcanonicalize_f64_nnan:
+; CHECK-NOFP16-RV32: # %bb.0:
+; CHECK-NOFP16-RV32-NEXT: fmin.d fa0, fa0, fa0
+; CHECK-NOFP16-RV32-NEXT: ret
+; RV32-SOFT-LABEL: fcanonicalize_f64_nnan:
+; RV32-SOFT: # %bb.0:
+; RV32-SOFT-NEXT: addi sp, sp, -16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 16
+; RV32-SOFT-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: .cfi_offset ra, -4
+; RV32-SOFT-NEXT: mv a2, a0
+; RV32-SOFT-NEXT: mv a3, a1
+; RV32-SOFT-NEXT: call fminimum_num
+; RV32-SOFT-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: .cfi_restore ra
+; RV32-SOFT-NEXT: addi sp, sp, 16
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV32-SOFT-NEXT: ret
%z = call nnan double @llvm.canonicalize.f64(double %x)
ret double %z
}
define <2 x double> @fcanonicalize_v2f64(<2 x double> %x) {
-; CHECK-LABEL: fcanonicalize_v2f64:
-; CHECK: # %bb.0:
-; CHECK-NEXT: fmin.d fa0, fa0, fa0
-; CHECK-NEXT: fmin.d fa1, fa1, fa1
-; CHECK-NEXT: ret
+; RV64-SOFT-LABEL: fcanonicalize_v2f64:
+; RV64-SOFT: # %bb.0:
+; RV64-SOFT-NEXT: addi sp, sp, -32
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 32
+; RV64-SOFT-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: .cfi_offset ra, -8
+; RV64-SOFT-NEXT: .cfi_offset s0, -16
+; RV64-SOFT-NEXT: .cfi_offset s1, -24
+; RV64-SOFT-NEXT: mv s0, a1
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_num
+; RV64-SOFT-NEXT: mv s1, a0
+; RV64-SOFT-NEXT: mv a0, s0
+; RV64-SOFT-NEXT: mv a1, s0
+; RV64-SOFT-NEXT: call fminimum_num
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: mv a0, s1
+; RV64-SOFT-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: .cfi_restore ra
+; RV64-SOFT-NEXT: .cfi_restore s0
+; RV64-SOFT-NEXT: .cfi_restore s1
+; RV64-SOFT-NEXT: addi sp, sp, 32
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV64-SOFT-NEXT: ret
+;
+; CHECK-SOFT-RV64-LABEL: fcanonicalize_v2f64:
+; CHECK-SOFT-RV64: # %bb.0:
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, -32
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 32
+; CHECK-SOFT-RV64-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: .cfi_offset ra, -8
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s0, -16
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s1, -24
+; CHECK-SOFT-RV64-NEXT: mv s0, a1
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_num
+; CHECK-SOFT-RV64-NEXT: mv s1, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s0
+; CHECK-SOFT-RV64-NEXT: mv a1, s0
+; CHECK-SOFT-RV64-NEXT: call fminimum_num
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s1
+; CHECK-SOFT-RV64-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, 32
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV64-NEXT: ret
+;
+; CHECK-FP16-RV64-LABEL: fcanonicalize_v2f64:
+; CHECK-FP16-RV64: # %bb.0:
+; CHECK-FP16-RV64-NEXT: fmin.d fa0, fa0, fa0
+; CHECK-FP16-RV64-NEXT: fmin.d fa1, fa1, fa1
+; CHECK-FP16-RV64-NEXT: ret
+;
+; CHECK-NOFP16-RV64-LABEL: fcanonicalize_v2f64:
+; CHECK-NOFP16-RV64: # %bb.0:
+; CHECK-NOFP16-RV64-NEXT: fmin.d fa0, fa0, fa0
+; CHECK-NOFP16-RV64-NEXT: fmin.d fa1, fa1, fa1
+; CHECK-NOFP16-RV64-NEXT: ret
+;
+; CHECK-SOFT-RV32-LABEL: fcanonicalize_v2f64:
+; CHECK-SOFT-RV32: # %bb.0:
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, -32
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 32
+; CHECK-SOFT-RV32-NEXT: sw ra, 28(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s0, 24(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s1, 20(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s2, 16(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s3, 12(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s4, 8(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: .cfi_offset ra, -4
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s0, -8
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s1, -12
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s2, -16
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s3, -20
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s4, -24
+; CHECK-SOFT-RV32-NEXT: lw a2, 0(a1)
+; CHECK-SOFT-RV32-NEXT: lw a3, 4(a1)
+; CHECK-SOFT-RV32-NEXT: lw s0, 8(a1)
+; CHECK-SOFT-RV32-NEXT: lw s1, 12(a1)
+; CHECK-SOFT-RV32-NEXT: mv s2, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, a2
+; CHECK-SOFT-RV32-NEXT: mv a1, a3
+; CHECK-SOFT-RV32-NEXT: call fminimum_num
+; CHECK-SOFT-RV32-NEXT: mv s3, a0
+; CHECK-SOFT-RV32-NEXT: mv s4, a1
+; CHECK-SOFT-RV32-NEXT: mv a0, s0
+; CHECK-SOFT-RV32-NEXT: mv a1, s1
+; CHECK-SOFT-RV32-NEXT: mv a2, s0
+; CHECK-SOFT-RV32-NEXT: mv a3, s1
+; CHECK-SOFT-RV32-NEXT: call fminimum_num
+; CHECK-SOFT-RV32-NEXT: sw s3, 0(s2)
+; CHECK-SOFT-RV32-NEXT: sw s4, 4(s2)
+; CHECK-SOFT-RV32-NEXT: sw a0, 8(s2)
+; CHECK-SOFT-RV32-NEXT: sw a1, 12(s2)
+; CHECK-SOFT-RV32-NEXT: lw ra, 28(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s0, 24(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s1, 20(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s2, 16(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s3, 12(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s4, 8(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s3
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s4
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, 32
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV32-NEXT: ret
+;
+; CHECK-FP16-RV32-LABEL: fcanonicalize_v2f64:
+; CHECK-FP16-RV32: # %bb.0:
+; CHECK-FP16-RV32-NEXT: fmin.d fa0, fa0, fa0
+; CHECK-FP16-RV32-NEXT: fmin.d fa1, fa1, fa1
+; CHECK-FP16-RV32-NEXT: ret
+;
+; CHECK-NOFP16-RV32-LABEL: fcanonicalize_v2f64:
+; CHECK-NOFP16-RV32: # %bb.0:
+; CHECK-NOFP16-RV32-NEXT: fmin.d fa0, fa0, fa0
+; CHECK-NOFP16-RV32-NEXT: fmin.d fa1, fa1, fa1
+; CHECK-NOFP16-RV32-NEXT: ret
+; RV32-SOFT-LABEL: fcanonicalize_v2f64:
+; RV32-SOFT: # %bb.0:
+; RV32-SOFT-NEXT: addi sp, sp, -32
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 32
+; RV32-SOFT-NEXT: sw ra, 28(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s0, 24(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s1, 20(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s2, 16(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s3, 12(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s4, 8(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: .cfi_offset ra, -4
+; RV32-SOFT-NEXT: .cfi_offset s0, -8
+; RV32-SOFT-NEXT: .cfi_offset s1, -12
+; RV32-SOFT-NEXT: .cfi_offset s2, -16
+; RV32-SOFT-NEXT: .cfi_offset s3, -20
+; RV32-SOFT-NEXT: .cfi_offset s4, -24
+; RV32-SOFT-NEXT: lw a2, 0(a1)
+; RV32-SOFT-NEXT: lw a3, 4(a1)
+; RV32-SOFT-NEXT: lw s0, 8(a1)
+; RV32-SOFT-NEXT: lw s1, 12(a1)
+; RV32-SOFT-NEXT: mv s2, a0
+; RV32-SOFT-NEXT: mv a0, a2
+; RV32-SOFT-NEXT: mv a1, a3
+; RV32-SOFT-NEXT: call fminimum_num
+; RV32-SOFT-NEXT: mv s3, a0
+; RV32-SOFT-NEXT: mv s4, a1
+; RV32-SOFT-NEXT: mv a0, s0
+; RV32-SOFT-NEXT: mv a1, s1
+; RV32-SOFT-NEXT: mv a2, s0
+; RV32-SOFT-NEXT: mv a3, s1
+; RV32-SOFT-NEXT: call fminimum_num
+; RV32-SOFT-NEXT: sw s3, 0(s2)
+; RV32-SOFT-NEXT: sw s4, 4(s2)
+; RV32-SOFT-NEXT: sw a0, 8(s2)
+; RV32-SOFT-NEXT: sw a1, 12(s2)
+; RV32-SOFT-NEXT: lw ra, 28(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s0, 24(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s1, 20(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s2, 16(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s3, 12(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s4, 8(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: .cfi_restore ra
+; RV32-SOFT-NEXT: .cfi_restore s0
+; RV32-SOFT-NEXT: .cfi_restore s1
+; RV32-SOFT-NEXT: .cfi_restore s2
+; RV32-SOFT-NEXT: .cfi_restore s3
+; RV32-SOFT-NEXT: .cfi_restore s4
+; RV32-SOFT-NEXT: addi sp, sp, 32
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV32-SOFT-NEXT: ret
%z = call <2 x double> @llvm.canonicalize.v2f64(<2 x double> %x)
ret <2 x double> %z
}
define <2 x double> @fcanonicalize_v2f64_nnan(<2 x double> %x) {
-; CHECK-LABEL: fcanonicalize_v2f64_nnan:
-; CHECK: # %bb.0:
-; CHECK-NEXT: fmin.d fa0, fa0, fa0
-; CHECK-NEXT: fmin.d fa1, fa1, fa1
-; CHECK-NEXT: ret
+; RV64-SOFT-LABEL: fcanonicalize_v2f64_nnan:
+; RV64-SOFT: # %bb.0:
+; RV64-SOFT-NEXT: addi sp, sp, -32
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 32
+; RV64-SOFT-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: .cfi_offset ra, -8
+; RV64-SOFT-NEXT: .cfi_offset s0, -16
+; RV64-SOFT-NEXT: .cfi_offset s1, -24
+; RV64-SOFT-NEXT: mv s0, a1
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: call fminimum_num
+; RV64-SOFT-NEXT: mv s1, a0
+; RV64-SOFT-NEXT: mv a0, s0
+; RV64-SOFT-NEXT: mv a1, s0
+; RV64-SOFT-NEXT: call fminimum_num
+; RV64-SOFT-NEXT: mv a1, a0
+; RV64-SOFT-NEXT: mv a0, s1
+; RV64-SOFT-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: .cfi_restore ra
+; RV64-SOFT-NEXT: .cfi_restore s0
+; RV64-SOFT-NEXT: .cfi_restore s1
+; RV64-SOFT-NEXT: addi sp, sp, 32
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV64-SOFT-NEXT: ret
+;
+; CHECK-SOFT-RV64-LABEL: fcanonicalize_v2f64_nnan:
+; CHECK-SOFT-RV64: # %bb.0:
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, -32
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 32
+; CHECK-SOFT-RV64-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: .cfi_offset ra, -8
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s0, -16
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s1, -24
+; CHECK-SOFT-RV64-NEXT: mv s0, a1
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: call fminimum_num
+; CHECK-SOFT-RV64-NEXT: mv s1, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s0
+; CHECK-SOFT-RV64-NEXT: mv a1, s0
+; CHECK-SOFT-RV64-NEXT: call fminimum_num
+; CHECK-SOFT-RV64-NEXT: mv a1, a0
+; CHECK-SOFT-RV64-NEXT: mv a0, s1
+; CHECK-SOFT-RV64-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, 32
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV64-NEXT: ret
+;
+; CHECK-FP16-RV64-LABEL: fcanonicalize_v2f64_nnan:
+; CHECK-FP16-RV64: # %bb.0:
+; CHECK-FP16-RV64-NEXT: fmin.d fa0, fa0, fa0
+; CHECK-FP16-RV64-NEXT: fmin.d fa1, fa1, fa1
+; CHECK-FP16-RV64-NEXT: ret
+;
+; CHECK-NOFP16-RV64-LABEL: fcanonicalize_v2f64_nnan:
+; CHECK-NOFP16-RV64: # %bb.0:
+; CHECK-NOFP16-RV64-NEXT: fmin.d fa0, fa0, fa0
+; CHECK-NOFP16-RV64-NEXT: fmin.d fa1, fa1, fa1
+; CHECK-NOFP16-RV64-NEXT: ret
+;
+; CHECK-SOFT-RV32-LABEL: fcanonicalize_v2f64_nnan:
+; CHECK-SOFT-RV32: # %bb.0:
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, -32
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 32
+; CHECK-SOFT-RV32-NEXT: sw ra, 28(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s0, 24(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s1, 20(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s2, 16(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s3, 12(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s4, 8(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: .cfi_offset ra, -4
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s0, -8
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s1, -12
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s2, -16
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s3, -20
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s4, -24
+; CHECK-SOFT-RV32-NEXT: lw a2, 0(a1)
+; CHECK-SOFT-RV32-NEXT: lw a3, 4(a1)
+; CHECK-SOFT-RV32-NEXT: lw s0, 8(a1)
+; CHECK-SOFT-RV32-NEXT: lw s1, 12(a1)
+; CHECK-SOFT-RV32-NEXT: mv s2, a0
+; CHECK-SOFT-RV32-NEXT: mv a0, a2
+; CHECK-SOFT-RV32-NEXT: mv a1, a3
+; CHECK-SOFT-RV32-NEXT: call fminimum_num
+; CHECK-SOFT-RV32-NEXT: mv s3, a0
+; CHECK-SOFT-RV32-NEXT: mv s4, a1
+; CHECK-SOFT-RV32-NEXT: mv a0, s0
+; CHECK-SOFT-RV32-NEXT: mv a1, s1
+; CHECK-SOFT-RV32-NEXT: mv a2, s0
+; CHECK-SOFT-RV32-NEXT: mv a3, s1
+; CHECK-SOFT-RV32-NEXT: call fminimum_num
+; CHECK-SOFT-RV32-NEXT: sw s3, 0(s2)
+; CHECK-SOFT-RV32-NEXT: sw s4, 4(s2)
+; CHECK-SOFT-RV32-NEXT: sw a0, 8(s2)
+; CHECK-SOFT-RV32-NEXT: sw a1, 12(s2)
+; CHECK-SOFT-RV32-NEXT: lw ra, 28(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s0, 24(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s1, 20(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s2, 16(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s3, 12(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s4, 8(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s3
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s4
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, 32
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV32-NEXT: ret
+;
+; CHECK-FP16-RV32-LABEL: fcanonicalize_v2f64_nnan:
+; CHECK-FP16-RV32: # %bb.0:
+; CHECK-FP16-RV32-NEXT: fmin.d fa0, fa0, fa0
+; CHECK-FP16-RV32-NEXT: fmin.d fa1, fa1, fa1
+; CHECK-FP16-RV32-NEXT: ret
+;
+; CHECK-NOFP16-RV32-LABEL: fcanonicalize_v2f64_nnan:
+; CHECK-NOFP16-RV32: # %bb.0:
+; CHECK-NOFP16-RV32-NEXT: fmin.d fa0, fa0, fa0
+; CHECK-NOFP16-RV32-NEXT: fmin.d fa1, fa1, fa1
+; CHECK-NOFP16-RV32-NEXT: ret
+; RV32-SOFT-LABEL: fcanonicalize_v2f64_nnan:
+; RV32-SOFT: # %bb.0:
+; RV32-SOFT-NEXT: addi sp, sp, -32
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 32
+; RV32-SOFT-NEXT: sw ra, 28(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s0, 24(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s1, 20(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s2, 16(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s3, 12(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s4, 8(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: .cfi_offset ra, -4
+; RV32-SOFT-NEXT: .cfi_offset s0, -8
+; RV32-SOFT-NEXT: .cfi_offset s1, -12
+; RV32-SOFT-NEXT: .cfi_offset s2, -16
+; RV32-SOFT-NEXT: .cfi_offset s3, -20
+; RV32-SOFT-NEXT: .cfi_offset s4, -24
+; RV32-SOFT-NEXT: lw a2, 0(a1)
+; RV32-SOFT-NEXT: lw a3, 4(a1)
+; RV32-SOFT-NEXT: lw s0, 8(a1)
+; RV32-SOFT-NEXT: lw s1, 12(a1)
+; RV32-SOFT-NEXT: mv s2, a0
+; RV32-SOFT-NEXT: mv a0, a2
+; RV32-SOFT-NEXT: mv a1, a3
+; RV32-SOFT-NEXT: call fminimum_num
+; RV32-SOFT-NEXT: mv s3, a0
+; RV32-SOFT-NEXT: mv s4, a1
+; RV32-SOFT-NEXT: mv a0, s0
+; RV32-SOFT-NEXT: mv a1, s1
+; RV32-SOFT-NEXT: mv a2, s0
+; RV32-SOFT-NEXT: mv a3, s1
+; RV32-SOFT-NEXT: call fminimum_num
+; RV32-SOFT-NEXT: sw s3, 0(s2)
+; RV32-SOFT-NEXT: sw s4, 4(s2)
+; RV32-SOFT-NEXT: sw a0, 8(s2)
+; RV32-SOFT-NEXT: sw a1, 12(s2)
+; RV32-SOFT-NEXT: lw ra, 28(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s0, 24(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s1, 20(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s2, 16(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s3, 12(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s4, 8(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: .cfi_restore ra
+; RV32-SOFT-NEXT: .cfi_restore s0
+; RV32-SOFT-NEXT: .cfi_restore s1
+; RV32-SOFT-NEXT: .cfi_restore s2
+; RV32-SOFT-NEXT: .cfi_restore s3
+; RV32-SOFT-NEXT: .cfi_restore s4
+; RV32-SOFT-NEXT: addi sp, sp, 32
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV32-SOFT-NEXT: ret
%z = call nnan <2 x double> @llvm.canonicalize.v2f64(<2 x double> %x)
ret <2 x double> %z
}
+
+define double @fcanonicalize_softfloat(double, double) unnamed_addr #0 {
+; RV64-SOFT-LABEL: fcanonicalize_softfloat:
+; RV64-SOFT: # %bb.0: # %start
+; RV64-SOFT-NEXT: addi sp, sp, -32
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 32
+; RV64-SOFT-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: sd s2, 0(sp) # 8-byte Folded Spill
+; RV64-SOFT-NEXT: .cfi_offset ra, -8
+; RV64-SOFT-NEXT: .cfi_offset s0, -16
+; RV64-SOFT-NEXT: .cfi_offset s1, -24
+; RV64-SOFT-NEXT: .cfi_offset s2, -32
+; RV64-SOFT-NEXT: mv s0, a1
+; RV64-SOFT-NEXT: mv s1, a0
+; RV64-SOFT-NEXT: call __ltdf2
+; RV64-SOFT-NEXT: srli s2, a0, 63
+; RV64-SOFT-NEXT: mv a0, s1
+; RV64-SOFT-NEXT: mv a1, s1
+; RV64-SOFT-NEXT: call __unorddf2
+; RV64-SOFT-NEXT: snez a0, a0
+; RV64-SOFT-NEXT: or a0, a0, s2
+; RV64-SOFT-NEXT: bnez a0, .LBB18_2
+; RV64-SOFT-NEXT: # %bb.1: # %start
+; RV64-SOFT-NEXT: mv s0, s1
+; RV64-SOFT-NEXT: .LBB18_2: # %start
+; RV64-SOFT-NEXT: mv a0, s0
+; RV64-SOFT-NEXT: mv a1, s0
+; RV64-SOFT-NEXT: call fminimum_num
+; RV64-SOFT-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: ld s2, 0(sp) # 8-byte Folded Reload
+; RV64-SOFT-NEXT: .cfi_restore ra
+; RV64-SOFT-NEXT: .cfi_restore s0
+; RV64-SOFT-NEXT: .cfi_restore s1
+; RV64-SOFT-NEXT: .cfi_restore s2
+; RV64-SOFT-NEXT: addi sp, sp, 32
+; RV64-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV64-SOFT-NEXT: ret
+;
+; CHECK-SOFT-RV64-LABEL: fcanonicalize_softfloat:
+; CHECK-SOFT-RV64: # %bb.0: # %start
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, -32
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 32
+; CHECK-SOFT-RV64-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: sd s2, 0(sp) # 8-byte Folded Spill
+; CHECK-SOFT-RV64-NEXT: .cfi_offset ra, -8
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s0, -16
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s1, -24
+; CHECK-SOFT-RV64-NEXT: .cfi_offset s2, -32
+; CHECK-SOFT-RV64-NEXT: mv s0, a1
+; CHECK-SOFT-RV64-NEXT: mv s1, a0
+; CHECK-SOFT-RV64-NEXT: call __ltdf2
+; CHECK-SOFT-RV64-NEXT: srli s2, a0, 63
+; CHECK-SOFT-RV64-NEXT: mv a0, s1
+; CHECK-SOFT-RV64-NEXT: mv a1, s1
+; CHECK-SOFT-RV64-NEXT: call __unorddf2
+; CHECK-SOFT-RV64-NEXT: snez a0, a0
+; CHECK-SOFT-RV64-NEXT: or a0, a0, s2
+; CHECK-SOFT-RV64-NEXT: bnez a0, .LBB18_2
+; CHECK-SOFT-RV64-NEXT: # %bb.1: # %start
+; CHECK-SOFT-RV64-NEXT: mv s0, s1
+; CHECK-SOFT-RV64-NEXT: .LBB18_2: # %start
+; CHECK-SOFT-RV64-NEXT: mv a0, s0
+; CHECK-SOFT-RV64-NEXT: mv a1, s0
+; CHECK-SOFT-RV64-NEXT: call fminimum_num
+; CHECK-SOFT-RV64-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: ld s2, 0(sp) # 8-byte Folded Reload
+; CHECK-SOFT-RV64-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV64-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV64-NEXT: addi sp, sp, 32
+; CHECK-SOFT-RV64-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV64-NEXT: ret
+;
+; CHECK-FP16-RV64-LABEL: fcanonicalize_softfloat:
+; CHECK-FP16-RV64: # %bb.0: # %start
+; CHECK-FP16-RV64-NEXT: flt.d a0, fa0, fa1
+; CHECK-FP16-RV64-NEXT: feq.d a1, fa0, fa0
+; CHECK-FP16-RV64-NEXT: xori a1, a1, 1
+; CHECK-FP16-RV64-NEXT: or a0, a1, a0
+; CHECK-FP16-RV64-NEXT: bnez a0, .LBB18_2
+; CHECK-FP16-RV64-NEXT: # %bb.1: # %start
+; CHECK-FP16-RV64-NEXT: fmv.d fa1, fa0
+; CHECK-FP16-RV64-NEXT: .LBB18_2: # %start
+; CHECK-FP16-RV64-NEXT: fmin.d fa0, fa1, fa1
+; CHECK-FP16-RV64-NEXT: ret
+;
+; CHECK-NOFP16-RV64-LABEL: fcanonicalize_softfloat:
+; CHECK-NOFP16-RV64: # %bb.0: # %start
+; CHECK-NOFP16-RV64-NEXT: flt.d a0, fa0, fa1
+; CHECK-NOFP16-RV64-NEXT: feq.d a1, fa0, fa0
+; CHECK-NOFP16-RV64-NEXT: xori a1, a1, 1
+; CHECK-NOFP16-RV64-NEXT: or a0, a1, a0
+; CHECK-NOFP16-RV64-NEXT: bnez a0, .LBB18_2
+; CHECK-NOFP16-RV64-NEXT: # %bb.1: # %start
+; CHECK-NOFP16-RV64-NEXT: fmv.d fa1, fa0
+; CHECK-NOFP16-RV64-NEXT: .LBB18_2: # %start
+; CHECK-NOFP16-RV64-NEXT: fmin.d fa0, fa1, fa1
+; CHECK-NOFP16-RV64-NEXT: ret
+;
+; CHECK-SOFT-RV32-LABEL: fcanonicalize_softfloat:
+; CHECK-SOFT-RV32: # %bb.0: # %start
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, -32
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 32
+; CHECK-SOFT-RV32-NEXT: sw ra, 28(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s0, 24(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s1, 20(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s2, 16(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s3, 12(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: sw s4, 8(sp) # 4-byte Folded Spill
+; CHECK-SOFT-RV32-NEXT: .cfi_offset ra, -4
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s0, -8
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s1, -12
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s2, -16
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s3, -20
+; CHECK-SOFT-RV32-NEXT: .cfi_offset s4, -24
+; CHECK-SOFT-RV32-NEXT: mv s0, a3
+; CHECK-SOFT-RV32-NEXT: mv s1, a2
+; CHECK-SOFT-RV32-NEXT: mv s2, a1
+; CHECK-SOFT-RV32-NEXT: mv s3, a0
+; CHECK-SOFT-RV32-NEXT: call __ltdf2
+; CHECK-SOFT-RV32-NEXT: srli s4, a0, 31
+; CHECK-SOFT-RV32-NEXT: mv a0, s3
+; CHECK-SOFT-RV32-NEXT: mv a1, s2
+; CHECK-SOFT-RV32-NEXT: mv a2, s3
+; CHECK-SOFT-RV32-NEXT: mv a3, s2
+; CHECK-SOFT-RV32-NEXT: call __unorddf2
+; CHECK-SOFT-RV32-NEXT: snez a0, a0
+; CHECK-SOFT-RV32-NEXT: or a0, a0, s4
+; CHECK-SOFT-RV32-NEXT: bnez a0, .LBB18_2
+; CHECK-SOFT-RV32-NEXT: # %bb.1: # %start
+; CHECK-SOFT-RV32-NEXT: mv s1, s3
+; CHECK-SOFT-RV32-NEXT: mv s0, s2
+; CHECK-SOFT-RV32-NEXT: .LBB18_2: # %start
+; CHECK-SOFT-RV32-NEXT: mv a0, s1
+; CHECK-SOFT-RV32-NEXT: mv a1, s0
+; CHECK-SOFT-RV32-NEXT: mv a2, s1
+; CHECK-SOFT-RV32-NEXT: mv a3, s0
+; CHECK-SOFT-RV32-NEXT: call fminimum_num
+; CHECK-SOFT-RV32-NEXT: lw ra, 28(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s0, 24(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s1, 20(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s2, 16(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s3, 12(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: lw s4, 8(sp) # 4-byte Folded Reload
+; CHECK-SOFT-RV32-NEXT: .cfi_restore ra
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s0
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s1
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s2
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s3
+; CHECK-SOFT-RV32-NEXT: .cfi_restore s4
+; CHECK-SOFT-RV32-NEXT: addi sp, sp, 32
+; CHECK-SOFT-RV32-NEXT: .cfi_def_cfa_offset 0
+; CHECK-SOFT-RV32-NEXT: ret
+;
+; CHECK-FP16-RV32-LABEL: fcanonicalize_softfloat:
+; CHECK-FP16-RV32: # %bb.0: # %start
+; CHECK-FP16-RV32-NEXT: flt.d a0, fa0, fa1
+; CHECK-FP16-RV32-NEXT: feq.d a1, fa0, fa0
+; CHECK-FP16-RV32-NEXT: xori a1, a1, 1
+; CHECK-FP16-RV32-NEXT: or a0, a1, a0
+; CHECK-FP16-RV32-NEXT: bnez a0, .LBB18_2
+; CHECK-FP16-RV32-NEXT: # %bb.1: # %start
+; CHECK-FP16-RV32-NEXT: fmv.d fa1, fa0
+; CHECK-FP16-RV32-NEXT: .LBB18_2: # %start
+; CHECK-FP16-RV32-NEXT: fmin.d fa0, fa1, fa1
+; CHECK-FP16-RV32-NEXT: ret
+;
+; CHECK-NOFP16-RV32-LABEL: fcanonicalize_softfloat:
+; CHECK-NOFP16-RV32: # %bb.0: # %start
+; CHECK-NOFP16-RV32-NEXT: flt.d a0, fa0, fa1
+; CHECK-NOFP16-RV32-NEXT: feq.d a1, fa0, fa0
+; CHECK-NOFP16-RV32-NEXT: xori a1, a1, 1
+; CHECK-NOFP16-RV32-NEXT: or a0, a1, a0
+; CHECK-NOFP16-RV32-NEXT: bnez a0, .LBB18_2
+; CHECK-NOFP16-RV32-NEXT: # %bb.1: # %start
+; CHECK-NOFP16-RV32-NEXT: fmv.d fa1, fa0
+; CHECK-NOFP16-RV32-NEXT: .LBB18_2: # %start
+; CHECK-NOFP16-RV32-NEXT: fmin.d fa0, fa1, fa1
+; CHECK-NOFP16-RV32-NEXT: ret
+; RV32-SOFT-LABEL: fcanonicalize_softfloat:
+; RV32-SOFT: # %bb.0: # %start
+; RV32-SOFT-NEXT: addi sp, sp, -32
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 32
+; RV32-SOFT-NEXT: sw ra, 28(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s0, 24(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s1, 20(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s2, 16(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s3, 12(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: sw s4, 8(sp) # 4-byte Folded Spill
+; RV32-SOFT-NEXT: .cfi_offset ra, -4
+; RV32-SOFT-NEXT: .cfi_offset s0, -8
+; RV32-SOFT-NEXT: .cfi_offset s1, -12
+; RV32-SOFT-NEXT: .cfi_offset s2, -16
+; RV32-SOFT-NEXT: .cfi_offset s3, -20
+; RV32-SOFT-NEXT: .cfi_offset s4, -24
+; RV32-SOFT-NEXT: mv s0, a3
+; RV32-SOFT-NEXT: mv s1, a2
+; RV32-SOFT-NEXT: mv s2, a1
+; RV32-SOFT-NEXT: mv s3, a0
+; RV32-SOFT-NEXT: call __ltdf2
+; RV32-SOFT-NEXT: srli s4, a0, 31
+; RV32-SOFT-NEXT: mv a0, s3
+; RV32-SOFT-NEXT: mv a1, s2
+; RV32-SOFT-NEXT: mv a2, s3
+; RV32-SOFT-NEXT: mv a3, s2
+; RV32-SOFT-NEXT: call __unorddf2
+; RV32-SOFT-NEXT: snez a0, a0
+; RV32-SOFT-NEXT: or a0, a0, s4
+; RV32-SOFT-NEXT: bnez a0, .LBB18_2
+; RV32-SOFT-NEXT: # %bb.1: # %start
+; RV32-SOFT-NEXT: mv s1, s3
+; RV32-SOFT-NEXT: mv s0, s2
+; RV32-SOFT-NEXT: .LBB18_2: # %start
+; RV32-SOFT-NEXT: mv a0, s1
+; RV32-SOFT-NEXT: mv a1, s0
+; RV32-SOFT-NEXT: mv a2, s1
+; RV32-SOFT-NEXT: mv a3, s0
+; RV32-SOFT-NEXT: call fminimum_num
+; RV32-SOFT-NEXT: lw ra, 28(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s0, 24(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s1, 20(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s2, 16(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s3, 12(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: lw s4, 8(sp) # 4-byte Folded Reload
+; RV32-SOFT-NEXT: .cfi_restore ra
+; RV32-SOFT-NEXT: .cfi_restore s0
+; RV32-SOFT-NEXT: .cfi_restore s1
+; RV32-SOFT-NEXT: .cfi_restore s2
+; RV32-SOFT-NEXT: .cfi_restore s3
+; RV32-SOFT-NEXT: .cfi_restore s4
+; RV32-SOFT-NEXT: addi sp, sp, 32
+; RV32-SOFT-NEXT: .cfi_def_cfa_offset 0
+; RV32-SOFT-NEXT: ret
+start:
+ %2 = fcmp olt double %0, %1
+ %3 = fcmp uno double %0, 0.000000e+00
+ %or.cond.i.i = or i1 %3, %2
+ %4 = select i1 %or.cond.i.i, double %1, double %0
+ %5 = tail call double @llvm.canonicalize.f64(double %4) #2
+ ret double %5
+}
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; CHECK: {{.*}}
More information about the llvm-commits
mailing list