[LoongArch][NFC] Pre-commit tests for vector fptrunc from vxf64 to vxf32

[ylzsx]

No description provided.

[llvmbot]

@llvm/pr-subscribers-backend-loongarch

Author: Zhaoxin Yang (ylzsx)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/164058.diff

2 Files Affected:

• (added) llvm/test/CodeGen/LoongArch/lasx/ir-instruction/fptrunc.ll (+65)
• (added) llvm/test/CodeGen/LoongArch/lsx/ir-instruction/fptrunc.ll (+72)

diff --git a/llvm/test/CodeGen/LoongArch/lasx/ir-instruction/fptrunc.ll b/llvm/test/CodeGen/LoongArch/lasx/ir-instruction/fptrunc.ll
new file mode 100644
index 0000000000000..6ade53d9ef531
--- /dev/null
+++ b/llvm/test/CodeGen/LoongArch/lasx/ir-instruction/fptrunc.ll
@@ -0,0 +1,65 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc --mtriple=loongarch32 --mattr=+32s,+lasx < %s | FileCheck %s --check-prefixes=CHECK
+; RUN: llc --mtriple=loongarch64 --mattr=+lasx < %s | FileCheck %s --check-prefixes=CHECK
+
+;; fptrunc
+define void @fptrunc_v4f64_to_v4f32(ptr %res, ptr %a0) nounwind {
+; CHECK-LABEL: fptrunc_v4f64_to_v4f32:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xvld $xr0, $a1, 0
+; CHECK-NEXT:    xvpickve.d $xr1, $xr0, 1
+; CHECK-NEXT:    fcvt.s.d $fa1, $fa1
+; CHECK-NEXT:    xvpickve.d $xr2, $xr0, 0
+; CHECK-NEXT:    fcvt.s.d $fa2, $fa2
+; CHECK-NEXT:    vextrins.w $vr2, $vr1, 16
+; CHECK-NEXT:    xvpickve.d $xr1, $xr0, 2
+; CHECK-NEXT:    fcvt.s.d $fa1, $fa1
+; CHECK-NEXT:    vextrins.w $vr2, $vr1, 32
+; CHECK-NEXT:    xvpickve.d $xr0, $xr0, 3
+; CHECK-NEXT:    fcvt.s.d $fa0, $fa0
+; CHECK-NEXT:    vextrins.w $vr2, $vr0, 48
+; CHECK-NEXT:    vst $vr2, $a0, 0
+; CHECK-NEXT:    ret
+entry:
+  %v0 = load <4 x double>, ptr %a0
+  %trunc = fptrunc <4 x double> %v0 to <4 x float>
+  store <4 x float> %trunc, ptr %res
+  ret void
+}
+
+define void @fptrunc_v8f64_to_v8f32(ptr %res, ptr %a0) nounwind {
+; CHECK-LABEL: fptrunc_v8f64_to_v8f32:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xvld $xr0, $a1, 32
+; CHECK-NEXT:    xvld $xr1, $a1, 0
+; CHECK-NEXT:    xvpickve.d $xr2, $xr0, 1
+; CHECK-NEXT:    fcvt.s.d $fa2, $fa2
+; CHECK-NEXT:    xvpickve.d $xr3, $xr0, 0
+; CHECK-NEXT:    fcvt.s.d $fa3, $fa3
+; CHECK-NEXT:    vextrins.w $vr3, $vr2, 16
+; CHECK-NEXT:    xvpickve.d $xr2, $xr0, 2
+; CHECK-NEXT:    fcvt.s.d $fa2, $fa2
+; CHECK-NEXT:    vextrins.w $vr3, $vr2, 32
+; CHECK-NEXT:    xvpickve.d $xr0, $xr0, 3
+; CHECK-NEXT:    fcvt.s.d $fa0, $fa0
+; CHECK-NEXT:    vextrins.w $vr3, $vr0, 48
+; CHECK-NEXT:    xvpickve.d $xr0, $xr1, 1
+; CHECK-NEXT:    fcvt.s.d $fa0, $fa0
+; CHECK-NEXT:    xvpickve.d $xr2, $xr1, 0
+; CHECK-NEXT:    fcvt.s.d $fa2, $fa2
+; CHECK-NEXT:    vextrins.w $vr2, $vr0, 16
+; CHECK-NEXT:    xvpickve.d $xr0, $xr1, 2
+; CHECK-NEXT:    fcvt.s.d $fa0, $fa0
+; CHECK-NEXT:    vextrins.w $vr2, $vr0, 32
+; CHECK-NEXT:    xvpickve.d $xr0, $xr1, 3
+; CHECK-NEXT:    fcvt.s.d $fa0, $fa0
+; CHECK-NEXT:    vextrins.w $vr2, $vr0, 48
+; CHECK-NEXT:    xvpermi.q $xr2, $xr3, 2
+; CHECK-NEXT:    xvst $xr2, $a0, 0
+; CHECK-NEXT:    ret
+entry:
+  %v0 = load <8 x double>, ptr %a0
+  %trunc = fptrunc <8 x double> %v0 to <8 x float>
+  store <8 x float> %trunc, ptr %res
+  ret void
+}
diff --git a/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/fptrunc.ll b/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/fptrunc.ll
new file mode 100644
index 0000000000000..acd487a889c4e
--- /dev/null
+++ b/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/fptrunc.ll
@@ -0,0 +1,72 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; 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
+
+;; fptrunc
+define void @fptrunc_v1f64_to_v1f32(ptr %res, ptr %a0) nounwind {
+; CHECK-LABEL: fptrunc_v1f64_to_v1f32:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    fld.d $fa0, $a1, 0
+; CHECK-NEXT:    fcvt.s.d $fa0, $fa0
+; CHECK-NEXT:    fst.s $fa0, $a0, 0
+; CHECK-NEXT:    ret
+entry:
+  %v0 = load <1 x double>, ptr %a0
+  %trunc = fptrunc <1 x double> %v0 to <1 x float>
+  store <1 x float> %trunc, ptr %res
+  ret void
+}
+
+define void @fptrunc_v2f64_to_v2f32(ptr %res, ptr %a0) nounwind {
+; LA32-LABEL: fptrunc_v2f64_to_v2f32:
+; LA32:       # %bb.0: # %entry
+; LA32-NEXT:    vld $vr0, $a1, 0
+; LA32-NEXT:    vreplvei.d $vr1, $vr0, 0
+; LA32-NEXT:    fcvt.s.d $fa1, $fa1
+; LA32-NEXT:    vreplvei.d $vr0, $vr0, 1
+; LA32-NEXT:    fcvt.s.d $fa0, $fa0
+; LA32-NEXT:    fst.s $fa0, $a0, 4
+; LA32-NEXT:    fst.s $fa1, $a0, 0
+; LA32-NEXT:    ret
+;
+; LA64-LABEL: fptrunc_v2f64_to_v2f32:
+; LA64:       # %bb.0: # %entry
+; LA64-NEXT:    vld $vr0, $a1, 0
+; LA64-NEXT:    vreplvei.d $vr1, $vr0, 1
+; LA64-NEXT:    fcvt.s.d $fa1, $fa1
+; LA64-NEXT:    vreplvei.d $vr0, $vr0, 0
+; LA64-NEXT:    fcvt.s.d $fa0, $fa0
+; LA64-NEXT:    vextrins.w $vr0, $vr1, 16
+; LA64-NEXT:    vstelm.d $vr0, $a0, 0, 0
+; LA64-NEXT:    ret
+entry:
+  %v0 = load <2 x double>, ptr %a0
+  %trunc = fptrunc <2 x double> %v0 to <2 x float>
+  store <2 x float> %trunc, ptr %res
+  ret void
+}
+
+define void @fptrunc_v4f64_to_v4f32(ptr %res, ptr %a0) nounwind {
+; CHECK-LABEL: fptrunc_v4f64_to_v4f32:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a1, 16
+; CHECK-NEXT:    vreplvei.d $vr2, $vr0, 1
+; CHECK-NEXT:    fcvt.s.d $fa2, $fa2
+; CHECK-NEXT:    vreplvei.d $vr0, $vr0, 0
+; CHECK-NEXT:    fcvt.s.d $fa0, $fa0
+; CHECK-NEXT:    vextrins.w $vr0, $vr2, 16
+; CHECK-NEXT:    vreplvei.d $vr2, $vr1, 0
+; CHECK-NEXT:    fcvt.s.d $fa2, $fa2
+; CHECK-NEXT:    vextrins.w $vr0, $vr2, 32
+; CHECK-NEXT:    vreplvei.d $vr1, $vr1, 1
+; CHECK-NEXT:    fcvt.s.d $fa1, $fa1
+; CHECK-NEXT:    vextrins.w $vr0, $vr1, 48
+; CHECK-NEXT:    vst $vr0, $a0, 0
+; CHECK-NEXT:    ret
+entry:
+  %v0 = load <4 x double>, ptr %a0
+  %trunc = fptrunc <4 x double> %v0 to <4 x float>
+  store <4 x float> %trunc, ptr %res
+  ret void
+}