[llvm][LoongArch] Introduce LASX and LSX conversion intrinsicsusers/hev/llvm-lasx-lsx-conversion

This patch introduces the LASX and LSX conversion intrinsics:

- <8 x float> @llvm.loongarch.lasx.cast.128.s(<4 x float>)
- <4 x double> @llvm.loongarch.lasx.cast.128.d(<2 x double>)
- <4 x i64> @llvm.loongarch.lasx.cast.128(<2 x i64>)
- <8 x float> @llvm.loongarch.lasx.concat.128.s(<4 x float>, <4 x float>)
- <4 x double> @llvm.loongarch.lasx.concat.128.d(<2 x double>, <2 x double>)
- <4 x i64> @llvm.loongarch.lasx.concat.128(<2 x i64>, <2 x i64>)
- <4 x float> @llvm.loongarch.lasx.extract.128.lo.s(<8 x float>)
- <2 x double> @llvm.loongarch.lasx.extract.128.lo.d(<4 x double>)
- <2 x i64> @llvm.loongarch.lasx.extract.128.lo(<4 x i64>)
- <4 x float> @llvm.loongarch.lasx.extract.128.hi.s(<8 x float>)
- <2 x double> @llvm.loongarch.lasx.extract.128.hi.d(<4 x double>)
- <2 x i64> @llvm.loongarch.lasx.extract.128.hi(<4 x i64>)
- <8 x float> @llvm.loongarch.lasx.insert.128.lo.s(<8 x float>, <4 x float>)
- <4 x double> @llvm.loongarch.lasx.insert.128.lo.d(<4 x double>, <2 x double>)
- <4 x i64> @llvm.loongarch.lasx.insert.128.lo(<4 x i64>, <2 x i64>)
- <8 x float> @llvm.loongarch.lasx.insert.128.hi.s(<8 x float>, <4 x float>)
- <4 x double> @llvm.loongarch.lasx.insert.128.hi.d(<4 x double>, <2 x double>)
- <4 x i64> @llvm.loongarch.lasx.insert.128.hi(<4 x i64>, <2 x i64>)

4 files changed, 308 insertions, 0 deletions

diff --git a/llvm/include/llvm/IR/IntrinsicsLoongArch.td b/llvm/include/llvm/IR/IntrinsicsLoongArch.td
index 84026aa..1c46965 100644
--- a/llvm/include/llvm/IR/IntrinsicsLoongArch.td
+++ b/llvm/include/llvm/IR/IntrinsicsLoongArch.td
@@ -1192,4 +1192,42 @@ def int_loongarch_lasx_xvstelm_w
 def int_loongarch_lasx_xvstelm_d
   : VecInt<[], [llvm_v4i64_ty, llvm_ptr_ty, llvm_i32_ty, llvm_i32_ty],
            [IntrWriteMem, IntrArgMemOnly, ImmArg<ArgIndex<2>>, ImmArg<ArgIndex<3>>]>;
+
+// LASX and LSX conversion
+def int_loongarch_lasx_cast_128_s
+  : VecInt<[llvm_v8f32_ty], [llvm_v4f32_ty], [IntrNoMem]>;
+def int_loongarch_lasx_cast_128_d
+  : VecInt<[llvm_v4f64_ty], [llvm_v2f64_ty], [IntrNoMem]>;
+def int_loongarch_lasx_cast_128
+  : VecInt<[llvm_v4i64_ty], [llvm_v2i64_ty], [IntrNoMem]>;
+def int_loongarch_lasx_concat_128_s
+  : VecInt<[llvm_v8f32_ty], [llvm_v4f32_ty, llvm_v4f32_ty], [IntrNoMem]>;
+def int_loongarch_lasx_concat_128_d
+  : VecInt<[llvm_v4f64_ty], [llvm_v2f64_ty, llvm_v2f64_ty], [IntrNoMem]>;
+def int_loongarch_lasx_concat_128
+  : VecInt<[llvm_v4i64_ty], [llvm_v2i64_ty, llvm_v2i64_ty], [IntrNoMem]>;
+def int_loongarch_lasx_extract_128_lo_s
+  : VecInt<[llvm_v4f32_ty], [llvm_v8f32_ty], [IntrNoMem]>;
+def int_loongarch_lasx_extract_128_lo_d
+  : VecInt<[llvm_v2f64_ty], [llvm_v4f64_ty], [IntrNoMem]>;
+def int_loongarch_lasx_extract_128_lo
+  : VecInt<[llvm_v2i64_ty], [llvm_v4i64_ty], [IntrNoMem]>;
+def int_loongarch_lasx_extract_128_hi_s
+  : VecInt<[llvm_v4f32_ty], [llvm_v8f32_ty], [IntrNoMem]>;
+def int_loongarch_lasx_extract_128_hi_d
+  : VecInt<[llvm_v2f64_ty], [llvm_v4f64_ty], [IntrNoMem]>;
+def int_loongarch_lasx_extract_128_hi
+  : VecInt<[llvm_v2i64_ty], [llvm_v4i64_ty], [IntrNoMem]>;
+def int_loongarch_lasx_insert_128_lo_s
+  : VecInt<[llvm_v8f32_ty], [llvm_v8f32_ty, llvm_v4f32_ty], [IntrNoMem]>;
+def int_loongarch_lasx_insert_128_lo_d
+  : VecInt<[llvm_v4f64_ty], [llvm_v4f64_ty, llvm_v2f64_ty], [IntrNoMem]>;
+def int_loongarch_lasx_insert_128_lo
+  : VecInt<[llvm_v4i64_ty], [llvm_v4i64_ty, llvm_v2i64_ty], [IntrNoMem]>;
+def int_loongarch_lasx_insert_128_hi_s
+  : VecInt<[llvm_v8f32_ty], [llvm_v8f32_ty, llvm_v4f32_ty], [IntrNoMem]>;
+def int_loongarch_lasx_insert_128_hi_d
+  : VecInt<[llvm_v4f64_ty], [llvm_v4f64_ty, llvm_v2f64_ty], [IntrNoMem]>;
+def int_loongarch_lasx_insert_128_hi
+  : VecInt<[llvm_v4i64_ty], [llvm_v4i64_ty, llvm_v2i64_ty], [IntrNoMem]>;
 } // TargetPrefix = "loongarch"
diff --git a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
index 634914d..1ae5fc8 100644
--- a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
+++ b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
@@ -6168,6 +6168,11 @@ performINTRINSIC_WO_CHAINCombine(SDNode *N, SelectionDAG &DAG,
                        N->getOperand(1),
                        DAG.getNode(ISD::ANY_EXTEND, DL, Subtarget.getGRLenVT(),
                                    N->getOperand(2)));
+  case Intrinsic::loongarch_lasx_concat_128_s:
+  case Intrinsic::loongarch_lasx_concat_128_d:
+  case Intrinsic::loongarch_lasx_concat_128:
+    return DAG.getNode(ISD::CONCAT_VECTORS, DL, N->getValueType(0),
+                       N->getOperand(1), N->getOperand(2));
   }
   return SDValue();
 }
diff --git a/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td b/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td
index a79c01c..5fed702 100644
--- a/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td
+++ b/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td
@@ -2028,6 +2028,37 @@ defm : subvector_subreg_lowering<LSX128, v2f64, LASX256, v4f64,  2,  sub_128>;
 defm : subvector_subreg_lowering<LSX128, v8i16, LASX256, v16i16, 8,  sub_128>;
 defm : subvector_subreg_lowering<LSX128, v16i8, LASX256, v32i8,  16, sub_128>;
 
+// LASX and LSX conversion
+def : Pat<(int_loongarch_lasx_cast_128_s (v4f32 LSX128:$src)),
+          (INSERT_SUBREG (IMPLICIT_DEF), LSX128:$src, sub_128)>;
+def : Pat<(int_loongarch_lasx_cast_128_d (v2f64 LSX128:$src)),
+          (INSERT_SUBREG (IMPLICIT_DEF), LSX128:$src, sub_128)>;
+def : Pat<(int_loongarch_lasx_cast_128 (v2i64 LSX128:$src)),
+          (INSERT_SUBREG (IMPLICIT_DEF), LSX128:$src, sub_128)>;
+def : Pat<(int_loongarch_lasx_extract_128_lo_s (v8f32 LASX256:$src)),
+          (EXTRACT_SUBREG LASX256:$src, sub_128)>;
+def : Pat<(int_loongarch_lasx_extract_128_lo_d (v4f64 LASX256:$src)),
+          (EXTRACT_SUBREG LASX256:$src, sub_128)>;
+def : Pat<(int_loongarch_lasx_extract_128_lo (v4i64 LASX256:$src)),
+          (EXTRACT_SUBREG LASX256:$src, sub_128)>;
+def : Pat<(int_loongarch_lasx_extract_128_hi_s (v8f32 LASX256:$src)),
+          (EXTRACT_SUBREG (XVPERMI_Q (IMPLICIT_DEF), LASX256:$src, 1), sub_128)>;
+def : Pat<(int_loongarch_lasx_extract_128_hi_d (v4f64 LASX256:$src)),
+          (EXTRACT_SUBREG (XVPERMI_Q (IMPLICIT_DEF), LASX256:$src, 1), sub_128)>;
+def : Pat<(int_loongarch_lasx_extract_128_hi (v4i64 LASX256:$src)),
+          (EXTRACT_SUBREG (XVPERMI_Q (IMPLICIT_DEF), LASX256:$src, 1), sub_128)>;
+def : Pat<(int_loongarch_lasx_insert_128_lo_s (v8f32 LASX256:$src), (v4f32 LSX128:$lo)),
+          (XVPERMI_Q LASX256:$src, (INSERT_SUBREG (IMPLICIT_DEF), LSX128:$lo, sub_128), 48)>;
+def : Pat<(int_loongarch_lasx_insert_128_lo_d (v4f64 LASX256:$src), (v2f64 LSX128:$lo)),
+          (XVPERMI_Q LASX256:$src, (INSERT_SUBREG (IMPLICIT_DEF), LSX128:$lo, sub_128), 48)>;
+def : Pat<(int_loongarch_lasx_insert_128_lo (v4i64 LASX256:$src), (v2i64 LSX128:$lo)),
+          (XVPERMI_Q LASX256:$src, (INSERT_SUBREG (IMPLICIT_DEF), LSX128:$lo, sub_128), 48)>;
+def : Pat<(int_loongarch_lasx_insert_128_hi_s (v8f32 LASX256:$src), (v4f32 LSX128:$lo)),
+          (XVPERMI_Q LASX256:$src, (INSERT_SUBREG (IMPLICIT_DEF), LSX128:$lo, sub_128), 2)>;
+def : Pat<(int_loongarch_lasx_insert_128_hi_d (v4f64 LASX256:$src), (v2f64 LSX128:$lo)),
+          (XVPERMI_Q LASX256:$src, (INSERT_SUBREG (IMPLICIT_DEF), LSX128:$lo, sub_128), 2)>;
+def : Pat<(int_loongarch_lasx_insert_128_hi (v4i64 LASX256:$src), (v2i64 LSX128:$lo)),
+          (XVPERMI_Q LASX256:$src, (INSERT_SUBREG (IMPLICIT_DEF), LSX128:$lo, sub_128), 2)>;
 } // Predicates = [HasExtLASX]
 
 /// Intrinsic pattern
diff --git a/llvm/test/CodeGen/LoongArch/lasx/intrinsic-conversion.ll b/llvm/test/CodeGen/LoongArch/lasx/intrinsic-conversion.ll
new file mode 100644
index 0000000..7b418cf
--- /dev/null
+++ b/llvm/test/CodeGen/LoongArch/lasx/intrinsic-conversion.ll
@@ -0,0 +1,234 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc --mtriple=loongarch32 --mattr=+32s,+lasx < %s | FileCheck %s
+; RUN: llc --mtriple=loongarch64 --mattr=+lasx < %s | FileCheck %s
+
+declare <8 x float> @llvm.loongarch.lasx.cast.128.s(<4 x float>)
+
+define <8 x float> @lasx_cast_128_s(<4 x float> %va) {
+; CHECK-LABEL: lasx_cast_128_s:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    # kill: def $vr0 killed $vr0 def $xr0
+; CHECK-NEXT:    ret
+entry:
+  %res = call <8 x float> @llvm.loongarch.lasx.cast.128.s(<4 x float> %va)
+  ret <8 x float> %res
+}
+
+declare <4 x double> @llvm.loongarch.lasx.cast.128.d(<2 x double>)
+
+define <4 x double> @lasx_cast_128_d(<2 x double> %va) {
+; CHECK-LABEL: lasx_cast_128_d:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    # kill: def $vr0 killed $vr0 def $xr0
+; CHECK-NEXT:    ret
+entry:
+  %res = call <4 x double> @llvm.loongarch.lasx.cast.128.d(<2 x double> %va)
+  ret <4 x double> %res
+}
+
+declare <4 x i64> @llvm.loongarch.lasx.cast.128(<2 x i64>)
+
+define <4 x i64> @lasx_cast_128(<2 x i64> %va) {
+; CHECK-LABEL: lasx_cast_128:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    # kill: def $vr0 killed $vr0 def $xr0
+; CHECK-NEXT:    ret
+entry:
+  %res = call <4 x i64> @llvm.loongarch.lasx.cast.128(<2 x i64> %va)
+  ret <4 x i64> %res
+}
+
+declare <8 x float> @llvm.loongarch.lasx.concat.128.s(<4 x float>, <4 x float>)
+
+define <8 x float> @lasx_concat_128_s(<4 x float> %va, <4 x float> %vb) {
+; CHECK-LABEL: lasx_concat_128_s:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    # kill: def $vr1 killed $vr1 def $xr1
+; CHECK-NEXT:    # kill: def $vr0 killed $vr0 def $xr0
+; CHECK-NEXT:    xvpermi.q $xr0, $xr1, 2
+; CHECK-NEXT:    ret
+entry:
+  %res = call <8 x float> @llvm.loongarch.lasx.concat.128.s(<4 x float> %va, <4 x float> %vb)
+  ret <8 x float> %res
+}
+
+declare <4 x double> @llvm.loongarch.lasx.concat.128.d(<2 x double>, <2 x double>)
+
+define <4 x double> @lasx_concat_128_d(<2 x double> %va, <2 x double> %vb) {
+; CHECK-LABEL: lasx_concat_128_d:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    # kill: def $vr1 killed $vr1 def $xr1
+; CHECK-NEXT:    # kill: def $vr0 killed $vr0 def $xr0
+; CHECK-NEXT:    xvpermi.q $xr0, $xr1, 2
+; CHECK-NEXT:    ret
+entry:
+  %res = call <4 x double> @llvm.loongarch.lasx.concat.128.d(<2 x double> %va, <2 x double> %vb)
+  ret <4 x double> %res
+}
+
+declare <4 x i64> @llvm.loongarch.lasx.concat.128(<2 x i64>, <2 x i64>)
+
+define <4 x i64> @lasx_concat_128(<2 x i64> %va, <2 x i64> %vb) {
+; CHECK-LABEL: lasx_concat_128:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    # kill: def $vr1 killed $vr1 def $xr1
+; CHECK-NEXT:    # kill: def $vr0 killed $vr0 def $xr0
+; CHECK-NEXT:    xvpermi.q $xr0, $xr1, 2
+; CHECK-NEXT:    ret
+entry:
+  %res = call <4 x i64> @llvm.loongarch.lasx.concat.128(<2 x i64> %va, <2 x i64> %vb)
+  ret <4 x i64> %res
+}
+
+declare <4 x float> @llvm.loongarch.lasx.extract.128.lo.s(<8 x float>)
+
+define <4 x float> @lasx_extract_128_lo_s(<8 x float> %va) {
+; CHECK-LABEL: lasx_extract_128_lo_s:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    # kill: def $vr0 killed $vr0 killed $xr0
+; CHECK-NEXT:    ret
+entry:
+  %res = call <4 x float> @llvm.loongarch.lasx.extract.128.lo.s(<8 x float> %va)
+  ret <4 x float> %res
+}
+
+declare <2 x double> @llvm.loongarch.lasx.extract.128.lo.d(<4 x double>)
+
+define <2 x double> @lasx_extract_128_lo_d(<4 x double> %va) {
+; CHECK-LABEL: lasx_extract_128_lo_d:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    # kill: def $vr0 killed $vr0 killed $xr0
+; CHECK-NEXT:    ret
+entry:
+  %res = call <2 x double> @llvm.loongarch.lasx.extract.128.lo.d(<4 x double> %va)
+  ret <2 x double> %res
+}
+
+declare <2 x i64> @llvm.loongarch.lasx.extract.128.lo(<4 x i64>)
+
+define <2 x i64> @lasx_extract_128_lo(<4 x i64> %va) {
+; CHECK-LABEL: lasx_extract_128_lo:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    # kill: def $vr0 killed $vr0 killed $xr0
+; CHECK-NEXT:    ret
+entry:
+  %res = call <2 x i64> @llvm.loongarch.lasx.extract.128.lo(<4 x i64> %va)
+  ret <2 x i64> %res
+}
+
+declare <4 x float> @llvm.loongarch.lasx.extract.128.hi.s(<8 x float>)
+
+define <4 x float> @lasx_extract_128_hi_s(<8 x float> %va) {
+; CHECK-LABEL: lasx_extract_128_hi_s:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xvpermi.q $xr0, $xr0, 1
+; CHECK-NEXT:    # kill: def $vr0 killed $vr0 killed $xr0
+; CHECK-NEXT:    ret
+entry:
+  %res = call <4 x float> @llvm.loongarch.lasx.extract.128.hi.s(<8 x float> %va)
+  ret <4 x float> %res
+}
+
+declare <2 x double> @llvm.loongarch.lasx.extract.128.hi.d(<4 x double>)
+
+define <2 x double> @lasx_extract_128_hi_d(<4 x double> %va) {
+; CHECK-LABEL: lasx_extract_128_hi_d:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xvpermi.q $xr0, $xr0, 1
+; CHECK-NEXT:    # kill: def $vr0 killed $vr0 killed $xr0
+; CHECK-NEXT:    ret
+entry:
+  %res = call <2 x double> @llvm.loongarch.lasx.extract.128.hi.d(<4 x double> %va)
+  ret <2 x double> %res
+}
+
+declare <2 x i64> @llvm.loongarch.lasx.extract.128.hi(<4 x i64>)
+
+define <2 x i64> @lasx_extract_128_hi(<4 x i64> %va) {
+; CHECK-LABEL: lasx_extract_128_hi:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xvpermi.q $xr0, $xr0, 1
+; CHECK-NEXT:    # kill: def $vr0 killed $vr0 killed $xr0
+; CHECK-NEXT:    ret
+entry:
+  %res = call <2 x i64> @llvm.loongarch.lasx.extract.128.hi(<4 x i64> %va)
+  ret <2 x i64> %res
+}
+
+declare <8 x float> @llvm.loongarch.lasx.insert.128.lo.s(<8 x float>, <4 x float>)
+
+define <8 x float> @lasx_insert_128_lo_s(<8 x float> %va, <4 x float> %vb) {
+; CHECK-LABEL: lasx_insert_128_lo_s:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    # kill: def $vr1 killed $vr1 def $xr1
+; CHECK-NEXT:    xvpermi.q $xr0, $xr1, 48
+; CHECK-NEXT:    ret
+entry:
+  %res = call <8 x float> @llvm.loongarch.lasx.insert.128.lo.s(<8 x float> %va, <4 x float> %vb)
+  ret <8 x float> %res
+}
+
+declare <4 x double> @llvm.loongarch.lasx.insert.128.lo.d(<4 x double>, <2 x double>)
+
+define <4 x double> @lasx_insert_128_lo_d(<4 x double> %va, <2 x double> %vb) {
+; CHECK-LABEL: lasx_insert_128_lo_d:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    # kill: def $vr1 killed $vr1 def $xr1
+; CHECK-NEXT:    xvpermi.q $xr0, $xr1, 48
+; CHECK-NEXT:    ret
+entry:
+  %res = call <4 x double> @llvm.loongarch.lasx.insert.128.lo.d(<4 x double> %va, <2 x double> %vb)
+  ret <4 x double> %res
+}
+
+declare <4 x i64> @llvm.loongarch.lasx.insert.128.lo(<4 x i64>, <2 x i64>)
+
+define <4 x i64> @lasx_insert_128_lo(<4 x i64> %va, <2 x i64> %vb) {
+; CHECK-LABEL: lasx_insert_128_lo:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    # kill: def $vr1 killed $vr1 def $xr1
+; CHECK-NEXT:    xvpermi.q $xr0, $xr1, 48
+; CHECK-NEXT:    ret
+entry:
+  %res = call <4 x i64> @llvm.loongarch.lasx.insert.128.lo(<4 x i64> %va, <2 x i64> %vb)
+  ret <4 x i64> %res
+}
+
+declare <8 x float> @llvm.loongarch.lasx.insert.128.hi.s(<8 x float>, <4 x float>)
+
+define <8 x float> @lasx_insert_128_hi_s(<8 x float> %va, <4 x float> %vb) {
+; CHECK-LABEL: lasx_insert_128_hi_s:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    # kill: def $vr1 killed $vr1 def $xr1
+; CHECK-NEXT:    xvpermi.q $xr0, $xr1, 2
+; CHECK-NEXT:    ret
+entry:
+  %res = call <8 x float> @llvm.loongarch.lasx.insert.128.hi.s(<8 x float> %va, <4 x float> %vb)
+  ret <8 x float> %res
+}
+
+declare <4 x double> @llvm.loongarch.lasx.insert.128.hi.d(<4 x double>, <2 x double>)
+
+define <4 x double> @lasx_insert_128_hi_d(<4 x double> %va, <2 x double> %vb) {
+; CHECK-LABEL: lasx_insert_128_hi_d:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    # kill: def $vr1 killed $vr1 def $xr1
+; CHECK-NEXT:    xvpermi.q $xr0, $xr1, 2
+; CHECK-NEXT:    ret
+entry:
+  %res = call <4 x double> @llvm.loongarch.lasx.insert.128.hi.d(<4 x double> %va, <2 x double> %vb)
+  ret <4 x double> %res
+}
+
+declare <4 x i64> @llvm.loongarch.lasx.insert.128.hi(<4 x i64>, <2 x i64>)
+
+define <4 x i64> @lasx_insert_128_hi(<4 x i64> %va, <2 x i64> %vb) {
+; CHECK-LABEL: lasx_insert_128_hi:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    # kill: def $vr1 killed $vr1 def $xr1
+; CHECK-NEXT:    xvpermi.q $xr0, $xr1, 2
+; CHECK-NEXT:    ret
+entry:
+  %res = call <4 x i64> @llvm.loongarch.lasx.insert.128.hi(<4 x i64> %va, <2 x i64> %vb)
+  ret <4 x i64> %res
+}