https://github.com/llvmbot created https://github.com/llvm/llvm-project/pull/234248
Backport ee216ff7ed6326989dfbd8f206342657aa1fda84 Requested by: @folkertdev >From dd55564731330ab452334b6078debcef0d7ed465 Mon Sep 17 00:00:00 2001 From: Folkert de Vries <[email protected]> Date: Wed, 21 Oct 2026 17:19:41 +0200 Subject: [PATCH] [LoongArch] fix more `sitofp` missing instruction issues (#234143) Turns out https://github.com/llvm/llvm-project/issues/228894 / https://github.com/llvm/llvm-project/pull/228896 was not sufficient, there are still isel issues with this conversion: - https://godbolt.org/z/arTaEhKsT - https://godbolt.org/z/nGh63Kvo6 (cherry picked from commit ee216ff7ed6326989dfbd8f206342657aa1fda84) --- .../LoongArch/LoongArchISelLowering.cpp | 4 + .../LoongArch/lsx/ir-instruction/sitofp.ll | 525 ++++++++++++++---- 2 files changed, 411 insertions(+), 118 deletions(-) diff --git a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp index 675fa046a7011b..5f4a50f1d4d35d 100644 --- a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp +++ b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp @@ -8288,6 +8288,10 @@ static SDValue performSINT_TO_FPCombine(SDNode *N, SelectionDAG &DAG, if (!SrcVT.isSimple() || !VT.isSimple()) return SDValue(); + // [x]vffint.s.l needs LSX and at least one full 128-bit source register. + if (!Subtarget.hasExtLSX() || SrcVT.getSizeInBits() < 128) + return SDValue(); + // Combine [x]vffint.s.l for vector si64 to float conversion. return MergeBlocksConvert(N, DAG, LoongArchISD::VFFINT, BlockBits); } diff --git a/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/sitofp.ll b/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/sitofp.ll index d6177d5ca0969b..b86299cccf9640 100644 --- a/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/sitofp.ll +++ b/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/sitofp.ll @@ -1,14 +1,35 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 3 -; RUN: llc --mtriple=loongarch32 --mattr=+32s,+lsx < %s | FileCheck %s --check-prefixes=CHECK,LA32 -; RUN: llc --mtriple=loongarch64 --mattr=+lsx < %s | FileCheck %s --check-prefixes=CHECK,LA64 +; RUN: llc --mtriple=loongarch32 --mattr=+32s,+lsx < %s | FileCheck %s --check-prefixes=LSX,LA32 +; RUN: llc --mtriple=loongarch64 --mattr=+lsx < %s | FileCheck %s --check-prefixes=LSX,LA64 +; RUN: llc --mtriple=loongarch64 --mattr=-lsx < %s | FileCheck %s --check-prefix=LA64-NO-LSX define void @sitofp_v4i32_v4f32(ptr %res, ptr %in){ -; CHECK-LABEL: sitofp_v4i32_v4f32: -; CHECK: # %bb.0: -; CHECK-NEXT: vld $vr0, $a1, 0 -; CHECK-NEXT: vffint.s.w $vr0, $vr0 -; CHECK-NEXT: vst $vr0, $a0, 0 -; CHECK-NEXT: ret +; LSX-LABEL: sitofp_v4i32_v4f32: +; LSX: # %bb.0: +; LSX-NEXT: vld $vr0, $a1, 0 +; LSX-NEXT: vffint.s.w $vr0, $vr0 +; LSX-NEXT: vst $vr0, $a0, 0 +; LSX-NEXT: ret +; +; LA64-NO-LSX-LABEL: sitofp_v4i32_v4f32: +; LA64-NO-LSX: # %bb.0: +; LA64-NO-LSX-NEXT: ld.w $a2, $a1, 0 +; LA64-NO-LSX-NEXT: ld.w $a3, $a1, 12 +; LA64-NO-LSX-NEXT: ld.w $a4, $a1, 4 +; LA64-NO-LSX-NEXT: ld.w $a1, $a1, 8 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa0, $a2 +; LA64-NO-LSX-NEXT: ffint.s.w $fa0, $fa0 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa1, $a4 +; LA64-NO-LSX-NEXT: ffint.s.w $fa1, $fa1 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa2, $a1 +; LA64-NO-LSX-NEXT: ffint.s.w $fa2, $fa2 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa3, $a3 +; LA64-NO-LSX-NEXT: ffint.s.w $fa3, $fa3 +; LA64-NO-LSX-NEXT: fst.s $fa3, $a0, 12 +; LA64-NO-LSX-NEXT: fst.s $fa2, $a0, 8 +; LA64-NO-LSX-NEXT: fst.s $fa1, $a0, 4 +; LA64-NO-LSX-NEXT: fst.s $fa0, $a0, 0 +; LA64-NO-LSX-NEXT: ret %v0 = load <4 x i32>, ptr %in %v1 = sitofp <4 x i32> %v0 to <4 x float> store <4 x float> %v1, ptr %res @@ -16,12 +37,22 @@ define void @sitofp_v4i32_v4f32(ptr %res, ptr %in){ } define void @sitofp_v2i64_v2f64(ptr %res, ptr %in){ -; CHECK-LABEL: sitofp_v2i64_v2f64: -; CHECK: # %bb.0: -; CHECK-NEXT: vld $vr0, $a1, 0 -; CHECK-NEXT: vffint.d.l $vr0, $vr0 -; CHECK-NEXT: vst $vr0, $a0, 0 -; CHECK-NEXT: ret +; LSX-LABEL: sitofp_v2i64_v2f64: +; LSX: # %bb.0: +; LSX-NEXT: vld $vr0, $a1, 0 +; LSX-NEXT: vffint.d.l $vr0, $vr0 +; LSX-NEXT: vst $vr0, $a0, 0 +; LSX-NEXT: ret +; +; LA64-NO-LSX-LABEL: sitofp_v2i64_v2f64: +; LA64-NO-LSX: # %bb.0: +; LA64-NO-LSX-NEXT: fld.d $fa0, $a1, 0 +; LA64-NO-LSX-NEXT: fld.d $fa1, $a1, 8 +; LA64-NO-LSX-NEXT: ffint.d.l $fa0, $fa0 +; LA64-NO-LSX-NEXT: ffint.d.l $fa1, $fa1 +; LA64-NO-LSX-NEXT: fst.d $fa1, $a0, 8 +; LA64-NO-LSX-NEXT: fst.d $fa0, $a0, 0 +; LA64-NO-LSX-NEXT: ret %v0 = load <2 x i64>, ptr %in %v1 = sitofp <2 x i64> %v0 to <2 x double> store <2 x double> %v1, ptr %res @@ -43,185 +74,419 @@ define void @sitofp_v2i64_v2f32(ptr %res, ptr %in){ ; LA64-NEXT: vffint.s.l $vr0, $vr0, $vr0 ; LA64-NEXT: vstelm.d $vr0, $a0, 0, 0 ; LA64-NEXT: ret +; +; LA64-NO-LSX-LABEL: sitofp_v2i64_v2f32: +; LA64-NO-LSX: # %bb.0: +; LA64-NO-LSX-NEXT: ld.d $a2, $a1, 0 +; LA64-NO-LSX-NEXT: ld.d $a1, $a1, 8 +; LA64-NO-LSX-NEXT: movgr2fr.d $fa0, $a2 +; LA64-NO-LSX-NEXT: ffint.s.l $fa0, $fa0 +; LA64-NO-LSX-NEXT: movgr2fr.d $fa1, $a1 +; LA64-NO-LSX-NEXT: ffint.s.l $fa1, $fa1 +; LA64-NO-LSX-NEXT: fst.s $fa1, $a0, 4 +; LA64-NO-LSX-NEXT: fst.s $fa0, $a0, 0 +; LA64-NO-LSX-NEXT: ret %v0 = load <2 x i64>, ptr %in %v1 = sitofp <2 x i64> %v0 to <2 x float> store <2 x float> %v1, ptr %res ret void } +define <1 x float> @sitofp_v1i64_v1f32(<1 x i64> %a) { +; LA32-LABEL: sitofp_v1i64_v1f32: +; LA32: # %bb.0: # %entry +; LA32-NEXT: addi.w $sp, $sp, -16 +; LA32-NEXT: .cfi_def_cfa_offset 16 +; LA32-NEXT: st.w $ra, $sp, 12 # 4-byte Folded Spill +; LA32-NEXT: .cfi_offset 1, -4 +; LA32-NEXT: bl __floatdisf +; LA32-NEXT: ld.w $ra, $sp, 12 # 4-byte Folded Reload +; LA32-NEXT: addi.w $sp, $sp, 16 +; LA32-NEXT: ret +; +; LA64-LABEL: sitofp_v1i64_v1f32: +; LA64: # %bb.0: # %entry +; LA64-NEXT: movgr2fr.d $fa0, $a0 +; LA64-NEXT: ffint.s.l $fa0, $fa0 +; LA64-NEXT: ret +; +; LA64-NO-LSX-LABEL: sitofp_v1i64_v1f32: +; LA64-NO-LSX: # %bb.0: # %entry +; LA64-NO-LSX-NEXT: movgr2fr.d $fa0, $a0 +; LA64-NO-LSX-NEXT: ffint.s.l $fa0, $fa0 +; LA64-NO-LSX-NEXT: ret +entry: + %v = sitofp <1 x i64> %a to <1 x float> + ret <1 x float> %v +} + define <4 x double> @sitofp_v4i64_v4f64(<4 x i64> %a) { -; CHECK-LABEL: sitofp_v4i64_v4f64: -; CHECK: # %bb.0: -; CHECK-NEXT: vffint.d.l $vr0, $vr0 -; CHECK-NEXT: vffint.d.l $vr1, $vr1 -; CHECK-NEXT: ret +; LSX-LABEL: sitofp_v4i64_v4f64: +; LSX: # %bb.0: +; LSX-NEXT: vffint.d.l $vr0, $vr0 +; LSX-NEXT: vffint.d.l $vr1, $vr1 +; LSX-NEXT: ret +; +; LA64-NO-LSX-LABEL: sitofp_v4i64_v4f64: +; LA64-NO-LSX: # %bb.0: +; LA64-NO-LSX-NEXT: fld.d $fa0, $a1, 0 +; LA64-NO-LSX-NEXT: fld.d $fa1, $a1, 8 +; LA64-NO-LSX-NEXT: fld.d $fa2, $a1, 16 +; LA64-NO-LSX-NEXT: fld.d $fa3, $a1, 24 +; LA64-NO-LSX-NEXT: ffint.d.l $fa0, $fa0 +; LA64-NO-LSX-NEXT: ffint.d.l $fa1, $fa1 +; LA64-NO-LSX-NEXT: ffint.d.l $fa2, $fa2 +; LA64-NO-LSX-NEXT: ffint.d.l $fa3, $fa3 +; LA64-NO-LSX-NEXT: fst.d $fa3, $a0, 24 +; LA64-NO-LSX-NEXT: fst.d $fa2, $a0, 16 +; LA64-NO-LSX-NEXT: fst.d $fa1, $a0, 8 +; LA64-NO-LSX-NEXT: fst.d $fa0, $a0, 0 +; LA64-NO-LSX-NEXT: ret %cvt = sitofp <4 x i64> %a to <4 x double> ret <4 x double> %cvt } define <4 x float> @sitofp_v4i64_v4f32(<4 x i64> %a) { -; CHECK-LABEL: sitofp_v4i64_v4f32: -; CHECK: # %bb.0: -; CHECK-NEXT: vffint.s.l $vr0, $vr1, $vr0 -; CHECK-NEXT: ret +; LSX-LABEL: sitofp_v4i64_v4f32: +; LSX: # %bb.0: +; LSX-NEXT: vffint.s.l $vr0, $vr1, $vr0 +; LSX-NEXT: ret +; +; LA64-NO-LSX-LABEL: sitofp_v4i64_v4f32: +; LA64-NO-LSX: # %bb.0: +; LA64-NO-LSX-NEXT: ld.d $a2, $a1, 0 +; LA64-NO-LSX-NEXT: ld.d $a3, $a1, 24 +; LA64-NO-LSX-NEXT: ld.d $a4, $a1, 8 +; LA64-NO-LSX-NEXT: ld.d $a1, $a1, 16 +; LA64-NO-LSX-NEXT: movgr2fr.d $fa0, $a2 +; LA64-NO-LSX-NEXT: ffint.s.l $fa0, $fa0 +; LA64-NO-LSX-NEXT: movgr2fr.d $fa1, $a4 +; LA64-NO-LSX-NEXT: ffint.s.l $fa1, $fa1 +; LA64-NO-LSX-NEXT: movgr2fr.d $fa2, $a1 +; LA64-NO-LSX-NEXT: ffint.s.l $fa2, $fa2 +; LA64-NO-LSX-NEXT: movgr2fr.d $fa3, $a3 +; LA64-NO-LSX-NEXT: ffint.s.l $fa3, $fa3 +; LA64-NO-LSX-NEXT: fst.s $fa3, $a0, 12 +; LA64-NO-LSX-NEXT: fst.s $fa2, $a0, 8 +; LA64-NO-LSX-NEXT: fst.s $fa1, $a0, 4 +; LA64-NO-LSX-NEXT: fst.s $fa0, $a0, 0 +; LA64-NO-LSX-NEXT: ret %cvt = sitofp <4 x i64> %a to <4 x float> ret <4 x float> %cvt } define <4 x double> @sitofp_v4i32_v4f64(<4 x i32> %a) { -; CHECK-LABEL: sitofp_v4i32_v4f64: -; CHECK: # %bb.0: -; CHECK-NEXT: vslti.w $vr1, $vr0, 0 -; CHECK-NEXT: vilvl.w $vr2, $vr1, $vr0 -; CHECK-NEXT: vffint.d.l $vr2, $vr2 -; CHECK-NEXT: vilvh.w $vr0, $vr1, $vr0 -; CHECK-NEXT: vffint.d.l $vr1, $vr0 -; CHECK-NEXT: vori.b $vr0, $vr2, 0 -; CHECK-NEXT: ret +; LSX-LABEL: sitofp_v4i32_v4f64: +; LSX: # %bb.0: +; LSX-NEXT: vslti.w $vr1, $vr0, 0 +; LSX-NEXT: vilvl.w $vr2, $vr1, $vr0 +; LSX-NEXT: vffint.d.l $vr2, $vr2 +; LSX-NEXT: vilvh.w $vr0, $vr1, $vr0 +; LSX-NEXT: vffint.d.l $vr1, $vr0 +; LSX-NEXT: vori.b $vr0, $vr2, 0 +; LSX-NEXT: ret +; +; LA64-NO-LSX-LABEL: sitofp_v4i32_v4f64: +; LA64-NO-LSX: # %bb.0: +; LA64-NO-LSX-NEXT: ld.w $a2, $a1, 0 +; LA64-NO-LSX-NEXT: ld.w $a3, $a1, 24 +; LA64-NO-LSX-NEXT: ld.w $a4, $a1, 8 +; LA64-NO-LSX-NEXT: ld.w $a1, $a1, 16 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa0, $a2 +; LA64-NO-LSX-NEXT: ffint.d.w $fa0, $fa0 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa1, $a4 +; LA64-NO-LSX-NEXT: ffint.d.w $fa1, $fa1 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa2, $a1 +; LA64-NO-LSX-NEXT: ffint.d.w $fa2, $fa2 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa3, $a3 +; LA64-NO-LSX-NEXT: ffint.d.w $fa3, $fa3 +; LA64-NO-LSX-NEXT: fst.d $fa3, $a0, 24 +; LA64-NO-LSX-NEXT: fst.d $fa2, $a0, 16 +; LA64-NO-LSX-NEXT: fst.d $fa1, $a0, 8 +; LA64-NO-LSX-NEXT: fst.d $fa0, $a0, 0 +; LA64-NO-LSX-NEXT: ret %cvt = sitofp <4 x i32> %a to <4 x double> ret <4 x double> %cvt } define <2 x float> @sitofp_v2i32_v2f32(<2 x i32> %a) { -; CHECK-LABEL: sitofp_v2i32_v2f32: -; CHECK: # %bb.0: -; CHECK-NEXT: vffint.s.w $vr0, $vr0 -; CHECK-NEXT: ret +; LSX-LABEL: sitofp_v2i32_v2f32: +; LSX: # %bb.0: +; LSX-NEXT: vffint.s.w $vr0, $vr0 +; LSX-NEXT: ret +; +; LA64-NO-LSX-LABEL: sitofp_v2i32_v2f32: +; LA64-NO-LSX: # %bb.0: +; LA64-NO-LSX-NEXT: movgr2fr.w $fa0, $a0 +; LA64-NO-LSX-NEXT: ffint.s.w $fa0, $fa0 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa1, $a1 +; LA64-NO-LSX-NEXT: ffint.s.w $fa1, $fa1 +; LA64-NO-LSX-NEXT: ret %cvt = sitofp <2 x i32> %a to <2 x float> ret <2 x float> %cvt } define <2 x double> @sitofp_v4i32_v2f64(<4 x i32> %a) { -; CHECK-LABEL: sitofp_v4i32_v2f64: -; CHECK: # %bb.0: -; CHECK-NEXT: vslti.w $vr1, $vr0, 0 -; CHECK-NEXT: vilvl.w $vr0, $vr1, $vr0 -; CHECK-NEXT: vffint.d.l $vr0, $vr0 -; CHECK-NEXT: ret +; LSX-LABEL: sitofp_v4i32_v2f64: +; LSX: # %bb.0: +; LSX-NEXT: vslti.w $vr1, $vr0, 0 +; LSX-NEXT: vilvl.w $vr0, $vr1, $vr0 +; LSX-NEXT: vffint.d.l $vr0, $vr0 +; LSX-NEXT: ret +; +; LA64-NO-LSX-LABEL: sitofp_v4i32_v2f64: +; LA64-NO-LSX: # %bb.0: +; LA64-NO-LSX-NEXT: ld.w $a1, $a0, 0 +; LA64-NO-LSX-NEXT: ld.w $a0, $a0, 8 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa0, $a1 +; LA64-NO-LSX-NEXT: ffint.d.w $fa0, $fa0 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa1, $a0 +; LA64-NO-LSX-NEXT: ffint.d.w $fa1, $fa1 +; LA64-NO-LSX-NEXT: ret %cvt = sitofp <4 x i32> %a to <4 x double> %shuf = shufflevector <4 x double> %cvt, <4 x double> poison, <2 x i32> <i32 0, i32 1> ret <2 x double> %shuf } define <2 x double> @sitofp_v2i16_v2f64(<8 x i16> %a) { -; CHECK-LABEL: sitofp_v2i16_v2f64: -; CHECK: # %bb.0: -; CHECK-NEXT: vextrins.h $vr0, $vr0, 65 -; CHECK-NEXT: vslli.d $vr0, $vr0, 48 -; CHECK-NEXT: vsrai.d $vr0, $vr0, 48 -; CHECK-NEXT: vffint.d.l $vr0, $vr0 -; CHECK-NEXT: ret +; LSX-LABEL: sitofp_v2i16_v2f64: +; LSX: # %bb.0: +; LSX-NEXT: vextrins.h $vr0, $vr0, 65 +; LSX-NEXT: vslli.d $vr0, $vr0, 48 +; LSX-NEXT: vsrai.d $vr0, $vr0, 48 +; LSX-NEXT: vffint.d.l $vr0, $vr0 +; LSX-NEXT: ret +; +; LA64-NO-LSX-LABEL: sitofp_v2i16_v2f64: +; LA64-NO-LSX: # %bb.0: +; LA64-NO-LSX-NEXT: ld.h $a1, $a0, 0 +; LA64-NO-LSX-NEXT: ld.h $a0, $a0, 8 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa0, $a1 +; LA64-NO-LSX-NEXT: ffint.d.w $fa0, $fa0 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa1, $a0 +; LA64-NO-LSX-NEXT: ffint.d.w $fa1, $fa1 +; LA64-NO-LSX-NEXT: ret %shuf = shufflevector <8 x i16> %a, <8 x i16> poison, <2 x i32> <i32 0, i32 1> %cvt = sitofp <2 x i16> %shuf to <2 x double> ret <2 x double> %cvt } define <2 x double> @sitofp_v8i16_v2f64(<8 x i16> %a) { -; CHECK-LABEL: sitofp_v8i16_v2f64: -; CHECK: # %bb.0: -; CHECK-NEXT: vslti.h $vr1, $vr0, 0 -; CHECK-NEXT: vilvl.h $vr0, $vr1, $vr0 -; CHECK-NEXT: vslti.w $vr1, $vr0, 0 -; CHECK-NEXT: vilvl.w $vr0, $vr1, $vr0 -; CHECK-NEXT: vffint.d.l $vr0, $vr0 -; CHECK-NEXT: ret +; LSX-LABEL: sitofp_v8i16_v2f64: +; LSX: # %bb.0: +; LSX-NEXT: vslti.h $vr1, $vr0, 0 +; LSX-NEXT: vilvl.h $vr0, $vr1, $vr0 +; LSX-NEXT: vslti.w $vr1, $vr0, 0 +; LSX-NEXT: vilvl.w $vr0, $vr1, $vr0 +; LSX-NEXT: vffint.d.l $vr0, $vr0 +; LSX-NEXT: ret +; +; LA64-NO-LSX-LABEL: sitofp_v8i16_v2f64: +; LA64-NO-LSX: # %bb.0: +; LA64-NO-LSX-NEXT: ld.h $a1, $a0, 0 +; LA64-NO-LSX-NEXT: ld.h $a0, $a0, 8 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa0, $a1 +; LA64-NO-LSX-NEXT: ffint.d.w $fa0, $fa0 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa1, $a0 +; LA64-NO-LSX-NEXT: ffint.d.w $fa1, $fa1 +; LA64-NO-LSX-NEXT: ret %cvt = sitofp <8 x i16> %a to <8 x double> %shuf = shufflevector <8 x double> %cvt, <8 x double> poison, <2 x i32> <i32 0, i32 1> ret <2 x double> %shuf } define <2 x double> @sitofp_v2i8_v2f64(<16 x i8> %a) { -; CHECK-LABEL: sitofp_v2i8_v2f64: -; CHECK: # %bb.0: -; CHECK-NEXT: vextrins.b $vr0, $vr0, 129 -; CHECK-NEXT: vslli.d $vr0, $vr0, 56 -; CHECK-NEXT: vsrai.d $vr0, $vr0, 56 -; CHECK-NEXT: vffint.d.l $vr0, $vr0 -; CHECK-NEXT: ret +; LSX-LABEL: sitofp_v2i8_v2f64: +; LSX: # %bb.0: +; LSX-NEXT: vextrins.b $vr0, $vr0, 129 +; LSX-NEXT: vslli.d $vr0, $vr0, 56 +; LSX-NEXT: vsrai.d $vr0, $vr0, 56 +; LSX-NEXT: vffint.d.l $vr0, $vr0 +; LSX-NEXT: ret +; +; LA64-NO-LSX-LABEL: sitofp_v2i8_v2f64: +; LA64-NO-LSX: # %bb.0: +; LA64-NO-LSX-NEXT: ld.b $a1, $a0, 0 +; LA64-NO-LSX-NEXT: ld.b $a0, $a0, 8 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa0, $a1 +; LA64-NO-LSX-NEXT: ffint.d.w $fa0, $fa0 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa1, $a0 +; LA64-NO-LSX-NEXT: ffint.d.w $fa1, $fa1 +; LA64-NO-LSX-NEXT: ret %shuf = shufflevector <16 x i8> %a, <16 x i8> poison, <2 x i32> <i32 0, i32 1> %cvt = sitofp <2 x i8> %shuf to <2 x double> ret <2 x double> %cvt } define <2 x double> @sitofp_v16i8_v2f64(<16 x i8> %a) { -; CHECK-LABEL: sitofp_v16i8_v2f64: -; CHECK: # %bb.0: -; CHECK-NEXT: vslti.b $vr1, $vr0, 0 -; CHECK-NEXT: vilvl.b $vr0, $vr1, $vr0 -; CHECK-NEXT: vslti.h $vr1, $vr0, 0 -; CHECK-NEXT: vilvl.h $vr0, $vr1, $vr0 -; CHECK-NEXT: vslti.w $vr1, $vr0, 0 -; CHECK-NEXT: vilvl.w $vr0, $vr1, $vr0 -; CHECK-NEXT: vffint.d.l $vr0, $vr0 -; CHECK-NEXT: ret +; LSX-LABEL: sitofp_v16i8_v2f64: +; LSX: # %bb.0: +; LSX-NEXT: vslti.b $vr1, $vr0, 0 +; LSX-NEXT: vilvl.b $vr0, $vr1, $vr0 +; LSX-NEXT: vslti.h $vr1, $vr0, 0 +; LSX-NEXT: vilvl.h $vr0, $vr1, $vr0 +; LSX-NEXT: vslti.w $vr1, $vr0, 0 +; LSX-NEXT: vilvl.w $vr0, $vr1, $vr0 +; LSX-NEXT: vffint.d.l $vr0, $vr0 +; LSX-NEXT: ret +; +; LA64-NO-LSX-LABEL: sitofp_v16i8_v2f64: +; LA64-NO-LSX: # %bb.0: +; LA64-NO-LSX-NEXT: ld.b $a1, $a0, 0 +; LA64-NO-LSX-NEXT: ld.b $a0, $a0, 8 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa0, $a1 +; LA64-NO-LSX-NEXT: ffint.d.w $fa0, $fa0 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa1, $a0 +; LA64-NO-LSX-NEXT: ffint.d.w $fa1, $fa1 +; LA64-NO-LSX-NEXT: ret %cvt = sitofp <16 x i8> %a to <16 x double> %shuf = shufflevector <16 x double> %cvt, <16 x double> poison, <2 x i32> <i32 0, i32 1> ret <2 x double> %shuf } define <4 x double> @sitofp_v4i16_v4f64(<8 x i16> %a) { -; CHECK-LABEL: sitofp_v4i16_v4f64: -; CHECK: # %bb.0: -; CHECK-NEXT: vslti.h $vr1, $vr0, 0 -; CHECK-NEXT: vilvl.h $vr1, $vr1, $vr0 -; CHECK-NEXT: vslti.w $vr2, $vr1, 0 -; CHECK-NEXT: vilvl.w $vr0, $vr2, $vr1 -; CHECK-NEXT: vffint.d.l $vr0, $vr0 -; CHECK-NEXT: vilvh.w $vr1, $vr2, $vr1 -; CHECK-NEXT: vffint.d.l $vr1, $vr1 -; CHECK-NEXT: ret +; LSX-LABEL: sitofp_v4i16_v4f64: +; LSX: # %bb.0: +; LSX-NEXT: vslti.h $vr1, $vr0, 0 +; LSX-NEXT: vilvl.h $vr1, $vr1, $vr0 +; LSX-NEXT: vslti.w $vr2, $vr1, 0 +; LSX-NEXT: vilvl.w $vr0, $vr2, $vr1 +; LSX-NEXT: vffint.d.l $vr0, $vr0 +; LSX-NEXT: vilvh.w $vr1, $vr2, $vr1 +; LSX-NEXT: vffint.d.l $vr1, $vr1 +; LSX-NEXT: ret +; +; LA64-NO-LSX-LABEL: sitofp_v4i16_v4f64: +; LA64-NO-LSX: # %bb.0: +; LA64-NO-LSX-NEXT: ld.h $a2, $a1, 0 +; LA64-NO-LSX-NEXT: ld.h $a3, $a1, 24 +; LA64-NO-LSX-NEXT: ld.h $a4, $a1, 8 +; LA64-NO-LSX-NEXT: ld.h $a1, $a1, 16 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa0, $a2 +; LA64-NO-LSX-NEXT: ffint.d.w $fa0, $fa0 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa1, $a4 +; LA64-NO-LSX-NEXT: ffint.d.w $fa1, $fa1 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa2, $a1 +; LA64-NO-LSX-NEXT: ffint.d.w $fa2, $fa2 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa3, $a3 +; LA64-NO-LSX-NEXT: ffint.d.w $fa3, $fa3 +; LA64-NO-LSX-NEXT: fst.d $fa3, $a0, 24 +; LA64-NO-LSX-NEXT: fst.d $fa2, $a0, 16 +; LA64-NO-LSX-NEXT: fst.d $fa1, $a0, 8 +; LA64-NO-LSX-NEXT: fst.d $fa0, $a0, 0 +; LA64-NO-LSX-NEXT: ret %shuf = shufflevector <8 x i16> %a, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> %cvt = sitofp <4 x i16> %shuf to <4 x double> ret <4 x double> %cvt } define <4 x double> @sitofp_v8i16_v4f64(<8 x i16> %a) { -; CHECK-LABEL: sitofp_v8i16_v4f64: -; CHECK: # %bb.0: -; CHECK-NEXT: vslti.h $vr1, $vr0, 0 -; CHECK-NEXT: vilvl.h $vr1, $vr1, $vr0 -; CHECK-NEXT: vslti.w $vr2, $vr1, 0 -; CHECK-NEXT: vilvl.w $vr0, $vr2, $vr1 -; CHECK-NEXT: vffint.d.l $vr0, $vr0 -; CHECK-NEXT: vilvh.w $vr1, $vr2, $vr1 -; CHECK-NEXT: vffint.d.l $vr1, $vr1 -; CHECK-NEXT: ret +; LSX-LABEL: sitofp_v8i16_v4f64: +; LSX: # %bb.0: +; LSX-NEXT: vslti.h $vr1, $vr0, 0 +; LSX-NEXT: vilvl.h $vr1, $vr1, $vr0 +; LSX-NEXT: vslti.w $vr2, $vr1, 0 +; LSX-NEXT: vilvl.w $vr0, $vr2, $vr1 +; LSX-NEXT: vffint.d.l $vr0, $vr0 +; LSX-NEXT: vilvh.w $vr1, $vr2, $vr1 +; LSX-NEXT: vffint.d.l $vr1, $vr1 +; LSX-NEXT: ret +; +; LA64-NO-LSX-LABEL: sitofp_v8i16_v4f64: +; LA64-NO-LSX: # %bb.0: +; LA64-NO-LSX-NEXT: ld.h $a2, $a1, 0 +; LA64-NO-LSX-NEXT: ld.h $a3, $a1, 24 +; LA64-NO-LSX-NEXT: ld.h $a4, $a1, 8 +; LA64-NO-LSX-NEXT: ld.h $a1, $a1, 16 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa0, $a2 +; LA64-NO-LSX-NEXT: ffint.d.w $fa0, $fa0 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa1, $a4 +; LA64-NO-LSX-NEXT: ffint.d.w $fa1, $fa1 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa2, $a1 +; LA64-NO-LSX-NEXT: ffint.d.w $fa2, $fa2 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa3, $a3 +; LA64-NO-LSX-NEXT: ffint.d.w $fa3, $fa3 +; LA64-NO-LSX-NEXT: fst.d $fa3, $a0, 24 +; LA64-NO-LSX-NEXT: fst.d $fa2, $a0, 16 +; LA64-NO-LSX-NEXT: fst.d $fa1, $a0, 8 +; LA64-NO-LSX-NEXT: fst.d $fa0, $a0, 0 +; LA64-NO-LSX-NEXT: ret %cvt = sitofp <8 x i16> %a to <8 x double> %shuf = shufflevector <8 x double> %cvt, <8 x double> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> ret <4 x double> %shuf } define <4 x double> @sitofp_v4i8_v4f64(<16 x i8> %a) { -; CHECK-LABEL: sitofp_v4i8_v4f64: -; CHECK: # %bb.0: -; CHECK-NEXT: vilvl.b $vr0, $vr0, $vr0 -; CHECK-NEXT: vilvl.h $vr0, $vr0, $vr0 -; CHECK-NEXT: vslli.w $vr0, $vr0, 24 -; CHECK-NEXT: vsrai.w $vr1, $vr0, 24 -; CHECK-NEXT: vslti.w $vr2, $vr1, 0 -; CHECK-NEXT: vilvl.w $vr0, $vr2, $vr1 -; CHECK-NEXT: vffint.d.l $vr0, $vr0 -; CHECK-NEXT: vilvh.w $vr1, $vr2, $vr1 -; CHECK-NEXT: vffint.d.l $vr1, $vr1 -; CHECK-NEXT: ret +; LSX-LABEL: sitofp_v4i8_v4f64: +; LSX: # %bb.0: +; LSX-NEXT: vilvl.b $vr0, $vr0, $vr0 +; LSX-NEXT: vilvl.h $vr0, $vr0, $vr0 +; LSX-NEXT: vslli.w $vr0, $vr0, 24 +; LSX-NEXT: vsrai.w $vr1, $vr0, 24 +; LSX-NEXT: vslti.w $vr2, $vr1, 0 +; LSX-NEXT: vilvl.w $vr0, $vr2, $vr1 +; LSX-NEXT: vffint.d.l $vr0, $vr0 +; LSX-NEXT: vilvh.w $vr1, $vr2, $vr1 +; LSX-NEXT: vffint.d.l $vr1, $vr1 +; LSX-NEXT: ret +; +; LA64-NO-LSX-LABEL: sitofp_v4i8_v4f64: +; LA64-NO-LSX: # %bb.0: +; LA64-NO-LSX-NEXT: ld.b $a2, $a1, 0 +; LA64-NO-LSX-NEXT: ld.b $a3, $a1, 24 +; LA64-NO-LSX-NEXT: ld.b $a4, $a1, 8 +; LA64-NO-LSX-NEXT: ld.b $a1, $a1, 16 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa0, $a2 +; LA64-NO-LSX-NEXT: ffint.d.w $fa0, $fa0 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa1, $a4 +; LA64-NO-LSX-NEXT: ffint.d.w $fa1, $fa1 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa2, $a1 +; LA64-NO-LSX-NEXT: ffint.d.w $fa2, $fa2 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa3, $a3 +; LA64-NO-LSX-NEXT: ffint.d.w $fa3, $fa3 +; LA64-NO-LSX-NEXT: fst.d $fa3, $a0, 24 +; LA64-NO-LSX-NEXT: fst.d $fa2, $a0, 16 +; LA64-NO-LSX-NEXT: fst.d $fa1, $a0, 8 +; LA64-NO-LSX-NEXT: fst.d $fa0, $a0, 0 +; LA64-NO-LSX-NEXT: ret %shuf = shufflevector <16 x i8> %a, <16 x i8> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> %cvt = sitofp <4 x i8> %shuf to <4 x double> ret <4 x double> %cvt } define <4 x double> @sitofp_v16i8_v4f64(<16 x i8> %a) { -; CHECK-LABEL: sitofp_v16i8_v4f64: -; CHECK: # %bb.0: -; CHECK-NEXT: vslti.b $vr1, $vr0, 0 -; CHECK-NEXT: vilvl.b $vr0, $vr1, $vr0 -; CHECK-NEXT: vslti.h $vr1, $vr0, 0 -; CHECK-NEXT: vilvl.h $vr1, $vr1, $vr0 -; CHECK-NEXT: vslti.w $vr2, $vr1, 0 -; CHECK-NEXT: vilvl.w $vr0, $vr2, $vr1 -; CHECK-NEXT: vffint.d.l $vr0, $vr0 -; CHECK-NEXT: vilvh.w $vr1, $vr2, $vr1 -; CHECK-NEXT: vffint.d.l $vr1, $vr1 -; CHECK-NEXT: ret +; LSX-LABEL: sitofp_v16i8_v4f64: +; LSX: # %bb.0: +; LSX-NEXT: vslti.b $vr1, $vr0, 0 +; LSX-NEXT: vilvl.b $vr0, $vr1, $vr0 +; LSX-NEXT: vslti.h $vr1, $vr0, 0 +; LSX-NEXT: vilvl.h $vr1, $vr1, $vr0 +; LSX-NEXT: vslti.w $vr2, $vr1, 0 +; LSX-NEXT: vilvl.w $vr0, $vr2, $vr1 +; LSX-NEXT: vffint.d.l $vr0, $vr0 +; LSX-NEXT: vilvh.w $vr1, $vr2, $vr1 +; LSX-NEXT: vffint.d.l $vr1, $vr1 +; LSX-NEXT: ret +; +; LA64-NO-LSX-LABEL: sitofp_v16i8_v4f64: +; LA64-NO-LSX: # %bb.0: +; LA64-NO-LSX-NEXT: ld.b $a2, $a1, 0 +; LA64-NO-LSX-NEXT: ld.b $a3, $a1, 24 +; LA64-NO-LSX-NEXT: ld.b $a4, $a1, 8 +; LA64-NO-LSX-NEXT: ld.b $a1, $a1, 16 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa0, $a2 +; LA64-NO-LSX-NEXT: ffint.d.w $fa0, $fa0 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa1, $a4 +; LA64-NO-LSX-NEXT: ffint.d.w $fa1, $fa1 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa2, $a1 +; LA64-NO-LSX-NEXT: ffint.d.w $fa2, $fa2 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa3, $a3 +; LA64-NO-LSX-NEXT: ffint.d.w $fa3, $fa3 +; LA64-NO-LSX-NEXT: fst.d $fa3, $a0, 24 +; LA64-NO-LSX-NEXT: fst.d $fa2, $a0, 16 +; LA64-NO-LSX-NEXT: fst.d $fa1, $a0, 8 +; LA64-NO-LSX-NEXT: fst.d $fa0, $a0, 0 +; LA64-NO-LSX-NEXT: ret %cvt = sitofp <16 x i8> %a to <16 x double> %shuf = shufflevector <16 x double> %cvt, <16 x double> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> ret <4 x double> %shuf @@ -277,6 +542,30 @@ define <5 x double> @sitofp_v5i32_v5f64(<5 x i32> %a) { ; LA64-NEXT: vst $vr1, $a0, 16 ; LA64-NEXT: vst $vr3, $a0, 0 ; LA64-NEXT: ret +; +; LA64-NO-LSX-LABEL: sitofp_v5i32_v5f64: +; LA64-NO-LSX: # %bb.0: +; LA64-NO-LSX-NEXT: ld.w $a2, $a1, 32 +; LA64-NO-LSX-NEXT: ld.w $a3, $a1, 0 +; LA64-NO-LSX-NEXT: ld.w $a4, $a1, 24 +; LA64-NO-LSX-NEXT: ld.w $a5, $a1, 8 +; LA64-NO-LSX-NEXT: ld.w $a1, $a1, 16 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa0, $a3 +; LA64-NO-LSX-NEXT: ffint.d.w $fa0, $fa0 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa1, $a5 +; LA64-NO-LSX-NEXT: ffint.d.w $fa1, $fa1 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa2, $a1 +; LA64-NO-LSX-NEXT: ffint.d.w $fa2, $fa2 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa3, $a4 +; LA64-NO-LSX-NEXT: ffint.d.w $fa3, $fa3 +; LA64-NO-LSX-NEXT: movgr2fr.w $fa4, $a2 +; LA64-NO-LSX-NEXT: ffint.d.w $fa4, $fa4 +; LA64-NO-LSX-NEXT: fst.d $fa4, $a0, 32 +; LA64-NO-LSX-NEXT: fst.d $fa3, $a0, 24 +; LA64-NO-LSX-NEXT: fst.d $fa2, $a0, 16 +; LA64-NO-LSX-NEXT: fst.d $fa1, $a0, 8 +; LA64-NO-LSX-NEXT: fst.d $fa0, $a0, 0 +; LA64-NO-LSX-NEXT: ret %v = sitofp <5 x i32> %a to <5 x double> ret <5 x double> %v } _______________________________________________ llvm-branch-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
