[mips][msa] Add DLSA instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201081 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/llvm/IR/IntrinsicsMips.td b/include/llvm/IR/IntrinsicsMips.td
index 50c0c278f46959a5cdabd7da375928d65c364a8b..34557612cb96edbe806c0c8e74a04128c2b1e5df 100644 (file)
--- a/include/llvm/IR/IntrinsicsMips.td
+++ b/include/llvm/IR/IntrinsicsMips.td
@@ -837,6 +837,12 @@ def int_mips_div_u_w : GCCBuiltin<"__builtin_msa_div_u_w">,
 def int_mips_div_u_d : GCCBuiltin<"__builtin_msa_div_u_d">,
   Intrinsic<[llvm_v2i64_ty], [llvm_v2i64_ty, llvm_v2i64_ty], [IntrNoMem]>;
 
+// This instruction is part of the MSA spec but it does not share the
+// __builtin_msa prefix because it operates on GP registers.
+def int_mips_dlsa : GCCBuiltin<"__builtin_mips_dlsa">,
+  Intrinsic<[llvm_i64_ty], [llvm_i64_ty, llvm_i64_ty, llvm_i32_ty],
+            [IntrNoMem]>;
+
 def int_mips_dotp_s_h : GCCBuiltin<"__builtin_msa_dotp_s_h">,
   Intrinsic<[llvm_v8i16_ty], [llvm_v16i8_ty, llvm_v16i8_ty], [IntrNoMem]>;
 def int_mips_dotp_s_w : GCCBuiltin<"__builtin_msa_dotp_s_w">,

diff --git a/lib/Target/Mips/MipsMSAInstrFormats.td b/lib/Target/Mips/MipsMSAInstrFormats.td
index 3942d05914f80c8868259300fffb76204397c36a..6bd0366b52e7b0c8a5ff654a667e379a7cdbe230 100644 (file)
--- a/lib/Target/Mips/MipsMSAInstrFormats.td
+++ b/lib/Target/Mips/MipsMSAInstrFormats.td
@@ -27,6 +27,10 @@ class MSASpecial : MSAInst {
   let Inst{31-26} = 0b000000;
 }
 
+class MSA64Special : MSA64Inst {
+  let Inst{31-26} = 0b000000;
+}
+
 class MSAPseudo<dag outs, dag ins, list<dag> pattern,
                 InstrItinClass itin = IIPseudo>:
   MipsPseudo<outs, ins, pattern, itin> {
@@ -445,3 +449,17 @@ class SPECIAL_LSA_FMT<bits<6> minor>: MSASpecial {
   let Inst{7-6} = sa;
   let Inst{5-0} = minor;
 }
+
+class SPECIAL_DLSA_FMT<bits<6> minor>: MSA64Special {
+  bits<5> rs;
+  bits<5> rt;
+  bits<5> rd;
+  bits<2> sa;
+
+  let Inst{25-21} = rs;
+  let Inst{20-16} = rt;
+  let Inst{15-11} = rd;
+  let Inst{10-8} = 0b000;
+  let Inst{7-6} = sa;
+  let Inst{5-0} = minor;
+}

diff --git a/lib/Target/Mips/MipsMSAInstrInfo.td b/lib/Target/Mips/MipsMSAInstrInfo.td
index 545bfccb88d421120c7e02652797f6af94a5640c..b23cfe600567b5bfc66a727595a59c8914a3bdfb 100644 (file)
--- a/lib/Target/Mips/MipsMSAInstrInfo.td
+++ b/lib/Target/Mips/MipsMSAInstrInfo.td
@@ -878,6 +878,7 @@ class LDI_W_ENC  : MSA_I10_FMT<0b110, 0b10, 0b000111>;
 class LDI_D_ENC  : MSA_I10_FMT<0b110, 0b11, 0b000111>;
 
 class LSA_ENC : SPECIAL_LSA_FMT<0b000101>;
+class DLSA_ENC : SPECIAL_DLSA_FMT<0b010101>;
 
 class MADD_Q_H_ENC : MSA_3RF_FMT<0b0101, 0b0, 0b011100>;
 class MADD_Q_W_ENC : MSA_3RF_FMT<0b0101, 0b1, 0b011100>;
@@ -2324,6 +2325,7 @@ class LSA_DESC_BASE<string instr_asm, RegisterOperand RORD,
 }
 
 class LSA_DESC : LSA_DESC_BASE<"lsa", GPR32Opnd>;
+class DLSA_DESC : LSA_DESC_BASE<"dlsa", GPR64Opnd>;
 
 class MADD_Q_H_DESC : MSA_3RF_4RF_DESC_BASE<"madd_q.h", int_mips_madd_q_h,
                                             MSA128HOpnd>;
@@ -3191,6 +3193,7 @@ def LDI_W : LDI_W_ENC, LDI_W_DESC;
 def LDI_D : LDI_D_ENC, LDI_D_DESC;
 
 def LSA : LSA_ENC, LSA_DESC;
+def DLSA : DLSA_ENC, DLSA_DESC;
 
 def MADD_Q_H : MADD_Q_H_ENC, MADD_Q_H_DESC;
 def MADD_Q_W : MADD_Q_W_ENC, MADD_Q_W_DESC;

diff --git a/lib/Target/Mips/MipsSEISelLowering.cpp b/lib/Target/Mips/MipsSEISelLowering.cpp
index aa33fab720f88eb5d8289e91bc0477ce6c9d7220..3d3ce0f572166e7e4d961596886aaa3e33b8065f 100644 (file)
--- a/lib/Target/Mips/MipsSEISelLowering.cpp
+++ b/lib/Target/Mips/MipsSEISelLowering.cpp
@@ -1807,7 +1807,8 @@ SDValue MipsSETargetLowering::lowerINTRINSIC_WO_CHAIN(SDValue Op,
   case Intrinsic::mips_ldi_w:
   case Intrinsic::mips_ldi_d:
     return lowerMSASplatImm(Op, 1, DAG);
-  case Intrinsic::mips_lsa: {
+  case Intrinsic::mips_lsa:
+  case Intrinsic::mips_dlsa: {
     EVT ResTy = Op->getValueType(0);
     return DAG.getNode(ISD::ADD, SDLoc(Op), ResTy, Op->getOperand(1),
                        DAG.getNode(ISD::SHL, SDLoc(Op), ResTy,

diff --git a/test/CodeGen/Mips/msa/special.ll b/test/CodeGen/Mips/msa/special.ll
index b325c68b6a173b29fbc1a9676cd79991876f5573..f65a14f7bb1fb934eeb7b3e714ec4d11ffff025f 100644 (file)
--- a/test/CodeGen/Mips/msa/special.ll
+++ b/test/CodeGen/Mips/msa/special.ll
@@ -2,6 +2,8 @@
 
 ; RUN: llc -march=mips -mattr=+msa,+fp64 < %s | \
 ; RUN:   FileCheck %s --check-prefix=MIPS32
+; RUN: llc -march=mips64 -mcpu=mips64r2 -mattr=+msa,+fp64 < %s | \
+; RUN:   FileCheck %s --check-prefix=MIPS64
 
 define i32 @llvm_mips_lsa_test(i32 %a, i32 %b) nounwind {
 entry:
@@ -25,3 +27,26 @@ entry:
 ; MIPS32: lsa_test:
 ; MIPS32: lsa {{\$[0-9]+}}, $5, $4, 2
 ; MIPS32: .size lsa_test
+
+define i64 @llvm_mips_dlsa_test(i64 %a, i64 %b) nounwind {
+entry:
+  %0 = tail call i64 @llvm.mips.dlsa(i64 %a, i64 %b, i32 2)
+  ret i64 %0
+}
+
+declare i64 @llvm.mips.dlsa(i64, i64, i32) nounwind
+
+; MIPS64: llvm_mips_dlsa_test:
+; MIPS64: dlsa {{\$[0-9]+}}, $5, $4, 2
+; MIPS64: .size llvm_mips_dlsa_test
+
+define i64 @dlsa_test(i64 %a, i64 %b) nounwind {
+entry:
+  %0 = shl i64 %b, 2
+  %1 = add i64 %a, %0
+  ret i64 %1
+}
+
+; MIPS64: dlsa_test:
+; MIPS64: dlsa {{\$[0-9]+}}, $5, $4, 2
+; MIPS64: .size dlsa_test

diff --git a/test/MC/Mips/msa/test_dlsa.s b/test/MC/Mips/msa/test_dlsa.s
new file mode 100644 (file)
index 0000000..a70999d
--- /dev/null
+++ b/test/MC/Mips/msa/test_dlsa.s
@@ -0,0 +1,21 @@
+# RUN: llvm-mc %s -arch=mips64 -mcpu=mips64r2 -mattr=+msa -show-encoding | \
+# RUN:   FileCheck %s
+#
+# RUN: llvm-mc %s -arch=mips -mcpu=mips64r2 -mattr=+msa -filetype=obj -o - | \
+# RUN:   llvm-objdump -d -arch=mips64 -mattr=+msa - | \
+# RUN:     FileCheck %s -check-prefix=CHECKOBJDUMP
+#
+# CHECK:        dlsa        $8, $9, $10, 1              # encoding: [0x01,0x2a,0x40,0x15]
+# CHECK:        dlsa        $8, $9, $10, 2              # encoding: [0x01,0x2a,0x40,0x55]
+# CHECK:        dlsa        $8, $9, $10, 3              # encoding: [0x01,0x2a,0x40,0x95]
+# CHECK:        dlsa        $8, $9, $10, 4              # encoding: [0x01,0x2a,0x40,0xd5]
+
+# CHECKOBJDUMP: dlsa        $8, $9, $10, 1
+# CHECKOBJDUMP: dlsa        $8, $9, $10, 2
+# CHECKOBJDUMP: dlsa        $8, $9, $10, 3
+# CHECKOBJDUMP: dlsa        $8, $9, $10, 4
+
+                dlsa        $8, $9, $10, 1
+                dlsa        $8, $9, $10, 2
+                dlsa        $8, $9, $10, 3
+                dlsa        $8, $9, $10, 4