[mips][msa] Implemented insert.d intrinsic.

This intrinsic is lowered into an equivalent INSERT_VECTOR_ELT which is
further lowered into a sequence of insert.w's on MIPS32.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@191521 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/llvm/IR/IntrinsicsMips.td b/include/llvm/IR/IntrinsicsMips.td
index 53e21428bf4ea2311b7189b429ff0b66b76185a7..5d8a7be7947bb1da641f4d9ba31ffb379c410bb3 100644 (file)
--- a/include/llvm/IR/IntrinsicsMips.td
+++ b/include/llvm/IR/IntrinsicsMips.td
@@ -1208,6 +1208,9 @@ def int_mips_insert_h : GCCBuiltin<"__builtin_msa_insert_h">,
 def int_mips_insert_w : GCCBuiltin<"__builtin_msa_insert_w">,
   Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_i32_ty, llvm_i32_ty],
   [IntrNoMem]>;
+def int_mips_insert_d : GCCBuiltin<"__builtin_msa_insert_d">,
+  Intrinsic<[llvm_v2i64_ty], [llvm_v2i64_ty, llvm_i32_ty, llvm_i64_ty],
+  [IntrNoMem]>;
 
 def int_mips_insve_b : GCCBuiltin<"__builtin_msa_insve_b">,
   Intrinsic<[llvm_v16i8_ty],

diff --git a/lib/Target/Mips/MipsSEISelLowering.cpp b/lib/Target/Mips/MipsSEISelLowering.cpp
index b05ab343413f9a2f4bea85d8cc538af058829c28..52509b72d1b3ad86f447751c8de9a9eaef7bc77a 100644 (file)
--- a/lib/Target/Mips/MipsSEISelLowering.cpp
+++ b/lib/Target/Mips/MipsSEISelLowering.cpp
@@ -1048,19 +1048,6 @@ static SDValue lowerMSACopyIntr(SDValue Op, SelectionDAG &DAG, unsigned Opc) {
   return Result;
 }
 
-// Lower an MSA insert intrinsic into the specified SelectionDAG node
-static SDValue lowerMSAInsertIntr(SDValue Op, SelectionDAG &DAG, unsigned Opc) {
-  SDLoc DL(Op);
-  SDValue Op0 = Op->getOperand(1);
-  SDValue Op1 = Op->getOperand(2);
-  SDValue Op2 = Op->getOperand(3);
-  EVT ResTy = Op->getValueType(0);
-
-  SDValue Result = DAG.getNode(Opc, DL, ResTy, Op0, Op2, Op1);
-
-  return Result;
-}
-
 static SDValue
 lowerMSASplatImm(SDLoc DL, EVT ResTy, SDValue ImmOp, SelectionDAG &DAG) {
   EVT ViaVecTy = ResTy;
@@ -1381,7 +1368,9 @@ SDValue MipsSETargetLowering::lowerINTRINSIC_WO_CHAIN(SDValue Op,
   case Intrinsic::mips_insert_b:
   case Intrinsic::mips_insert_h:
   case Intrinsic::mips_insert_w:
-    return lowerMSAInsertIntr(Op, DAG, ISD::INSERT_VECTOR_ELT);
+  case Intrinsic::mips_insert_d:
+    return DAG.getNode(ISD::INSERT_VECTOR_ELT, SDLoc(Op), Op->getValueType(0),
+                       Op->getOperand(1), Op->getOperand(3), Op->getOperand(2));
   case Intrinsic::mips_ldi_b:
   case Intrinsic::mips_ldi_h:
   case Intrinsic::mips_ldi_w:

diff --git a/test/CodeGen/Mips/msa/elm_insv.ll b/test/CodeGen/Mips/msa/elm_insv.ll
index 409503eb7cd6eb191c721f35276554b95f188296..a34002a375c5490b97af5c3cdd29dfef24001078 100644 (file)
--- a/test/CodeGen/Mips/msa/elm_insv.ll
+++ b/test/CodeGen/Mips/msa/elm_insv.ll
@@ -19,10 +19,10 @@ entry:
 declare <16 x i8> @llvm.mips.insert.b(<16 x i8>, i32, i32) nounwind
 
 ; CHECK: llvm_mips_insert_b_test:
-; CHECK: lw
-; CHECK: ld.b
-; CHECK: insert.b
-; CHECK: st.b
+; CHECK-DAG: lw [[R1:\$[0-9]+]], 0(
+; CHECK-DAG: ld.b [[R2:\$w[0-9]+]], 0(
+; CHECK-DAG: insert.b [[R2]][1], [[R1]]
+; CHECK-DAG: st.b [[R2]], 0(
 ; CHECK: .size llvm_mips_insert_b_test
 ;
 @llvm_mips_insert_h_ARG1 = global <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>, align 16
@@ -41,10 +41,10 @@ entry:
 declare <8 x i16> @llvm.mips.insert.h(<8 x i16>, i32, i32) nounwind
 
 ; CHECK: llvm_mips_insert_h_test:
-; CHECK: lw
-; CHECK: ld.h
-; CHECK: insert.h
-; CHECK: st.h
+; CHECK-DAG: lw [[R1:\$[0-9]+]], 0(
+; CHECK-DAG: ld.h [[R2:\$w[0-9]+]], 0(
+; CHECK-DAG: insert.h [[R2]][1], [[R1]]
+; CHECK-DAG: st.h [[R2]], 0(
 ; CHECK: .size llvm_mips_insert_h_test
 ;
 @llvm_mips_insert_w_ARG1 = global <4 x i32> <i32 0, i32 1, i32 2, i32 3>, align 16
@@ -63,12 +63,36 @@ entry:
 declare <4 x i32> @llvm.mips.insert.w(<4 x i32>, i32, i32) nounwind
 
 ; CHECK: llvm_mips_insert_w_test:
-; CHECK: lw
-; CHECK: ld.w
-; CHECK: insert.w
-; CHECK: st.w
+; CHECK-DAG: lw [[R1:\$[0-9]+]], 0(
+; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0(
+; CHECK-DAG: insert.w [[R2]][1], [[R1]]
+; CHECK-DAG: st.w [[R2]], 0(
 ; CHECK: .size llvm_mips_insert_w_test
 ;
+@llvm_mips_insert_d_ARG1 = global <2 x i64> <i64 0, i64 1>, align 16
+@llvm_mips_insert_d_ARG3 = global i64 27, align 16
+@llvm_mips_insert_d_RES  = global <2 x i64> <i64 0, i64 0>, align 16
+
+define void @llvm_mips_insert_d_test() nounwind {
+entry:
+  %0 = load <2 x i64>* @llvm_mips_insert_d_ARG1
+  %1 = load i64* @llvm_mips_insert_d_ARG3
+  %2 = tail call <2 x i64> @llvm.mips.insert.d(<2 x i64> %0, i32 1, i64 %1)
+  store <2 x i64> %2, <2 x i64>* @llvm_mips_insert_d_RES
+  ret void
+}
+
+declare <2 x i64> @llvm.mips.insert.d(<2 x i64>, i32, i64) nounwind
+
+; CHECK: llvm_mips_insert_d_test:
+; CHECK-DAG: lw [[R1:\$[0-9]+]], 0(
+; CHECK-DAG: lw [[R2:\$[0-9]+]], 4(
+; CHECK-DAG: ld.w [[R3:\$w[0-9]+]],
+; CHECK-DAG: insert.w [[R3]][2], [[R1]]
+; CHECK-DAG: insert.w [[R3]][3], [[R2]]
+; CHECK-DAG: st.w [[R3]],
+; CHECK: .size llvm_mips_insert_d_test
+;
 @llvm_mips_insve_b_ARG1 = global <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, align 16
 @llvm_mips_insve_b_ARG3 = global <16 x i8> <i8 16, i8 17, i8 18, i8 19, i8 20, i8 21, i8 22, i8 23, i8 24, i8 25, i8 26, i8 27, i8 28, i8 29, i8 30, i8 31>, align 16
 @llvm_mips_insve_b_RES  = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
@@ -85,10 +109,12 @@ entry:
 declare <16 x i8> @llvm.mips.insve.b(<16 x i8>, i32, <16 x i8>) nounwind
 
 ; CHECK: llvm_mips_insve_b_test:
-; CHECK: ld.b
-; CHECK: ld.b
-; CHECK: insve.b
-; CHECK: st.b
+; CHECK-DAG: lw [[R1:\$[0-9]+]], %got(llvm_mips_insve_b_ARG1)(
+; CHECK-DAG: lw [[R2:\$[0-9]+]], %got(llvm_mips_insve_b_ARG3)(
+; CHECK-DAG: ld.b [[R3:\$w[0-9]+]], 0([[R1]])
+; CHECK-DAG: ld.b [[R4:\$w[0-9]+]], 0([[R2]])
+; CHECK-DAG: insve.b [[R3]][1], [[R4]][0]
+; CHECK-DAG: st.b [[R3]],
 ; CHECK: .size llvm_mips_insve_b_test
 ;
 @llvm_mips_insve_h_ARG1 = global <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>, align 16
@@ -107,10 +133,12 @@ entry:
 declare <8 x i16> @llvm.mips.insve.h(<8 x i16>, i32, <8 x i16>) nounwind
 
 ; CHECK: llvm_mips_insve_h_test:
-; CHECK: ld.h
-; CHECK: ld.h
-; CHECK: insve.h
-; CHECK: st.h
+; CHECK-DAG: lw [[R1:\$[0-9]+]], %got(llvm_mips_insve_h_ARG1)(
+; CHECK-DAG: lw [[R2:\$[0-9]+]], %got(llvm_mips_insve_h_ARG3)(
+; CHECK-DAG: ld.h [[R3:\$w[0-9]+]], 0([[R1]])
+; CHECK-DAG: ld.h [[R4:\$w[0-9]+]], 0([[R2]])
+; CHECK-DAG: insve.h [[R3]][1], [[R4]][0]
+; CHECK-DAG: st.h [[R3]],
 ; CHECK: .size llvm_mips_insve_h_test
 ;
 @llvm_mips_insve_w_ARG1 = global <4 x i32> <i32 0, i32 1, i32 2, i32 3>, align 16
@@ -129,10 +157,12 @@ entry:
 declare <4 x i32> @llvm.mips.insve.w(<4 x i32>, i32, <4 x i32>) nounwind
 
 ; CHECK: llvm_mips_insve_w_test:
-; CHECK: ld.w
-; CHECK: ld.w
-; CHECK: insve.w
-; CHECK: st.w
+; CHECK-DAG: lw [[R1:\$[0-9]+]], %got(llvm_mips_insve_w_ARG1)(
+; CHECK-DAG: lw [[R2:\$[0-9]+]], %got(llvm_mips_insve_w_ARG3)(
+; CHECK-DAG: ld.w [[R3:\$w[0-9]+]], 0([[R1]])
+; CHECK-DAG: ld.w [[R4:\$w[0-9]+]], 0([[R2]])
+; CHECK-DAG: insve.w [[R3]][1], [[R4]][0]
+; CHECK-DAG: st.w [[R3]],
 ; CHECK: .size llvm_mips_insve_w_test
 ;
 @llvm_mips_insve_d_ARG1 = global <2 x i64> <i64 0, i64 1>, align 16
@@ -151,9 +181,11 @@ entry:
 declare <2 x i64> @llvm.mips.insve.d(<2 x i64>, i32, <2 x i64>) nounwind
 
 ; CHECK: llvm_mips_insve_d_test:
-; CHECK: ld.d
-; CHECK: ld.d
-; CHECK: insve.d
-; CHECK: st.d
+; CHECK-DAG: lw [[R1:\$[0-9]+]], %got(llvm_mips_insve_d_ARG1)(
+; CHECK-DAG: lw [[R2:\$[0-9]+]], %got(llvm_mips_insve_d_ARG3)(
+; CHECK-DAG: ld.d [[R3:\$w[0-9]+]], 0([[R1]])
+; CHECK-DAG: ld.d [[R4:\$w[0-9]+]], 0([[R2]])
+; CHECK-DAG: insve.d [[R3]][1], [[R4]][0]
+; CHECK-DAG: st.d [[R3]],
 ; CHECK: .size llvm_mips_insve_d_test
 ;