Merging r359834:

------------------------------------------------------------------------
r359834 | evandro | 2019-05-02 15:01:39 -0700 (Thu, 02 May 2019) | 3 lines

[AArch64] Update for Exynos

Fix the forwarding of multiplication results for Exynos M4.
------------------------------------------------------------------------

git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/release_80@359946 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 18 insertions, 82 deletions

diff --git a/lib/Target/AArch64/AArch64SchedExynosM4.td b/lib/Target/AArch64/AArch64SchedExynosM4.td
index 4d892465b3f2..61652b1d8e3d 100644
--- a/lib/Target/AArch64/AArch64SchedExynosM4.td
+++ b/lib/Target/AArch64/AArch64SchedExynosM4.td
@@ -239,7 +239,6 @@ def M4WriteNEONK   : SchedWriteRes<[M4UnitNSHF,
                                     M4UnitS0]>    { let Latency = 5;
                                                     let NumMicroOps = 2; }
 def M4WriteNEONL   : SchedWriteRes<[M4UnitNMUL]>  { let Latency = 3; }
-def M4WriteNEONM   : SchedWriteRes<[M4UnitNMUL]>  { let Latency = 3; }
 def M4WriteNEONN   : SchedWriteRes<[M4UnitNMSC,
                                     M4UnitNMSC]>  { let Latency = 5;
                                                     let NumMicroOps = 2; }
@@ -480,8 +479,6 @@ def M4WriteCOPY    : SchedWriteVariant<[SchedVar<ExynosFPPred, [M4WriteNALU1]>,
                                         SchedVar<NoSchedPred,  [M4WriteZ0]>]>;
 def M4WriteMOVI    : SchedWriteVariant<[SchedVar<IsZeroFPIdiomPred, [M4WriteZ0]>,
                                         SchedVar<NoSchedPred,       [M4WriteNALU1]>]>;
-def M4WriteMULL    : SchedWriteVariant<[SchedVar<ExynosLongVectorUpperPred, [M4WriteNEONM]>,
-                                        SchedVar<NoSchedPred,               [M4WriteNMUL3]>]>;
 
 // Fast forwarding.
 def M4ReadAESM1    : SchedReadAdvance<+1, [M4WriteNCRY1]>;
@@ -489,7 +486,8 @@ def M4ReadFMACM1   : SchedReadAdvance<+1, [M4WriteFMAC4,
                                            M4WriteFMAC4H,
                                            M4WriteFMAC5]>;
 def M4ReadNMULM1   : SchedReadAdvance<+1, [M4WriteNMUL3]>;
-def M4ReadMULLP2   : SchedReadAdvance<-2, [M4WriteNEONM]>;
+def M4ReadNMULP2   : SchedReadAdvance<-2, [M4WriteNMUL3]>;
+
 
 //===----------------------------------------------------------------------===//
 // Coarse scheduling model.
@@ -662,10 +660,8 @@ def : InstRW<[M4WriteNEONK],  (instregex "^FMOVDXHighr")>;
 def : InstRW<[M4WriteFCVT3H], (instregex "^F(RECP|RSQRT)Ev1f16")>;
 def : InstRW<[M4WriteFCVT3],  (instregex "^F(RECP|RSQRT)Ev1i(32|64)")>;
 def : InstRW<[M4WriteNMSC1],  (instregex "^FRECPXv1")>;
-def : InstRW<[M4WriteFMAC4H,
-              M4ReadFMACM1],  (instregex "^F(RECP|RSQRT)S16")>;
-def : InstRW<[M4WriteFMAC4,
-              M4ReadFMACM1],  (instregex "^F(RECP|RSQRT)S(32|64)")>;
+def : InstRW<[M4WriteFMAC4H], (instregex "^F(RECP|RSQRT)S16")>;
+def : InstRW<[M4WriteFMAC4],  (instregex "^F(RECP|RSQRT)S(32|64)")>;
 
 // FP load instructions.
 def : InstRW<[WriteVLD],    (instregex "^LDR[SDQ]l")>;
@@ -736,14 +732,20 @@ def : InstRW<[M4WriteNALU1],  (instregex "^(AND|BIC|EOR|NOT|ORN|ORR)v")>;
 def : InstRW<[M4WriteNMSC1],  (instregex "^[SU](MIN|MAX)v")>;
 def : InstRW<[M4WriteNMSC2],  (instregex "^[SU](MIN|MAX)Pv")>;
 def : InstRW<[M4WriteNHAD3],  (instregex "^[SU](MIN|MAX)Vv")>;
-def : InstRW<[M4WriteNMUL3],  (instregex "^(SQR?D)?MULH?v")>;
 def : InstRW<[M4WriteNMUL3,
               M4ReadNMULM1],  (instregex "^ML[AS]v")>;
-def : InstRW<[M4WriteNMUL3],  (instregex "^SQRDML[AS]H")>;
-def : InstRW<[M4WriteMULL,
-              M4ReadMULLP2],  (instregex "^(S|U|SQD)ML[AS]Lv")>;
-def : InstRW<[M4WriteMULL,
-              M4ReadMULLP2],  (instregex "^(S|U|SQD)MULLv")>;
+def : InstRW<[M4WriteNMUL3,
+              M4ReadNMULM1],  (instregex "^(SQR?D)?MULH?v")>;
+def : InstRW<[M4WriteNMUL3,
+              M4ReadNMULM1],  (instregex "^SQRDML[AS]H")>;
+def : InstRW<[M4WriteNMUL3,
+              M4ReadNMULM1],  (instregex "^(S|U|SQD)ML[AS]L(v1(i32|i64)|v2i32|v4i16|v8i8)")>;
+def : InstRW<[M4WriteNMUL3,
+              M4ReadNMULP2],  (instregex "^(S|U|SQD)ML[AS]L(v4i32|v8i16|v16i8)")>;
+def : InstRW<[M4WriteNMUL3,
+              M4ReadNMULM1],  (instregex "^(S|U|SQD)MULL(v1(i32|i64)|v2i32|v4i16|v8i8)")>;
+def : InstRW<[M4WriteNMUL3,
+              M4ReadNMULP2],  (instregex "^(S|U|SQD)MULL(v4i32|v8i16|v16i8)")>;
 def : InstRW<[M4WriteNMUL3],  (instregex "^[SU]DOT(lane)?v")>;
 def : InstRW<[M4WriteNHAD3],  (instregex "^[SU]ADALPv")>;
 def : InstRW<[M4WriteNSHT4A], (instregex "^[SU]R?SRA[dv]")>;
@@ -808,10 +810,8 @@ def : InstRW<[M4WriteNALU1],  (instregex "^FMOVv.f(32|64)")>;
 def : InstRW<[M4WriteFCVT3H], (instregex "^F(RECP|RSQRT)Ev[248]f16")>;
 def : InstRW<[M4WriteFCVT3],  (instregex "^F(RECP|RSQRT)Ev[248]f(32|64)")>;
 def : InstRW<[M4WriteFCVT3],  (instregex "^U(RECP|RSQRT)Ev[24]i32")>;
-def : InstRW<[M4WriteFMAC4H,
-              M4ReadFMACM1],  (instregex "^F(RECP|RSQRT)Sv.f16")>;
-def : InstRW<[M4WriteFMAC4,
-              M4ReadFMACM1],  (instregex "^F(RECP|RSQRT)Sv.f(32|64)")>;
+def : InstRW<[M4WriteFMAC4H], (instregex "^F(RECP|RSQRT)Sv.f16")>;
+def : InstRW<[M4WriteFMAC4],  (instregex "^F(RECP|RSQRT)Sv.f(32|64)")>;
 def : InstRW<[M4WriteNSHF1],  (instregex "^REV(16|32|64)v")>;
 def : InstRW<[M4WriteNSHFA],  (instregex "^TB[LX]v(8|16)i8One")>;
 def : InstRW<[M4WriteNSHFB],  (instregex "^TB[LX]v(8|16)i8Two")>;
diff --git a/lib/Target/AArch64/AArch64SchedPredExynos.td b/lib/Target/AArch64/AArch64SchedPredExynos.td
index 48c54230e9d8..316036d89406 100644
--- a/lib/Target/AArch64/AArch64SchedPredExynos.td
+++ b/lib/Target/AArch64/AArch64SchedPredExynos.td
@@ -103,17 +103,6 @@ def ExynosScaledIdxPred : MCSchedPredicate<ExynosScaledIdxFn>;
 // Identify FP instructions.
 def ExynosFPPred : MCSchedPredicate<CheckAny<[CheckDForm, CheckQForm]>>;
 
-// Identify whether an instruction whose result is a long vector
-// operates on the upper half of the input registers.
-def ExynosLongVectorUpperFn   : TIIPredicate<
-                                  "isExynosLongVectorUpper",
-                                  MCOpcodeSwitchStatement<
-                                  [MCOpcodeSwitchCase<
-                                    IsLongVectorUpperOp.ValidOpcodes,
-                                    MCReturnStatement<TruePred>>],
-                                  MCReturnStatement<FalsePred>>>;
-def ExynosLongVectorUpperPred : MCSchedPredicate<ExynosLongVectorUpperFn>;
-
 // Identify 128-bit NEON instructions.
 def ExynosQFormPred : MCSchedPredicate<CheckQForm>;
 
diff --git a/lib/Target/AArch64/AArch64SchedPredicates.td b/lib/Target/AArch64/AArch64SchedPredicates.td
index dbaf11fc95dd..b23572b41b9c 100644
--- a/lib/Target/AArch64/AArch64SchedPredicates.td
+++ b/lib/Target/AArch64/AArch64SchedPredicates.td
@@ -268,59 +268,6 @@ def IsStoreRegOffsetOp     : CheckOpcode<[STRBBroW, STRBBroX,
 def IsLoadStoreRegOffsetOp : CheckOpcode<!listconcat(IsLoadRegOffsetOp.ValidOpcodes,
                                                      IsStoreRegOffsetOp.ValidOpcodes)>;
 
-// Identify whether an instruction whose result is a long vector
-// operates on the upper half of the input registers.
-def IsLongVectorUpperOp    : CheckOpcode<[FCVTLv8i16, FCVTLv4i32,
-                                          FCVTNv8i16, FCVTNv4i32,
-                                          FCVTXNv4f32,
-                                          PMULLv16i8, PMULLv2i64,
-                                          RADDHNv8i16_v16i8, RADDHNv4i32_v8i16, RADDHNv2i64_v4i32,
-                                          RSHRNv16i8_shift, RSHRNv8i16_shift, RSHRNv4i32_shift,
-                                          RSUBHNv8i16_v16i8, RSUBHNv4i32_v8i16, RSUBHNv2i64_v4i32,
-                                          SABALv16i8_v8i16, SABALv8i16_v4i32, SABALv4i32_v2i64,
-                                          SABDLv16i8_v8i16, SABDLv8i16_v4i32, SABDLv4i32_v2i64,
-                                          SADDLv16i8_v8i16, SADDLv8i16_v4i32, SADDLv4i32_v2i64,
-                                          SADDWv16i8_v8i16, SADDWv8i16_v4i32, SADDWv4i32_v2i64,
-                                          SHLLv16i8, SHLLv8i16, SHLLv4i32,
-                                          SHRNv16i8_shift, SHRNv8i16_shift, SHRNv4i32_shift,
-                                          SMLALv16i8_v8i16, SMLALv8i16_v4i32, SMLALv4i32_v2i64,
-                                          SMLALv8i16_indexed, SMLALv4i32_indexed,
-                                          SMLSLv16i8_v8i16, SMLSLv8i16_v4i32, SMLSLv4i32_v2i64,
-                                          SMLSLv8i16_indexed, SMLSLv4i32_indexed,
-                                          SMULLv16i8_v8i16, SMULLv8i16_v4i32, SMULLv4i32_v2i64,
-                                          SMULLv8i16_indexed, SMULLv4i32_indexed,
-                                          SQDMLALv8i16_v4i32, SQDMLALv4i32_v2i64,
-                                          SQDMLALv8i16_indexed, SQDMLALv4i32_indexed,
-                                          SQDMLSLv8i16_v4i32, SQDMLSLv4i32_v2i64,
-                                          SQDMLSLv8i16_indexed, SQDMLSLv4i32_indexed,
-                                          SQDMULLv8i16_v4i32, SQDMULLv4i32_v2i64,
-                                          SQDMULLv8i16_indexed, SQDMULLv4i32_indexed,
-                                          SQRSHRNv16i8_shift, SQRSHRNv8i16_shift, SQRSHRNv4i32_shift,
-                                          SQRSHRUNv16i8_shift, SQRSHRUNv8i16_shift, SQRSHRUNv4i32_shift,
-                                          SQSHRNv16i8_shift, SQSHRNv8i16_shift, SQSHRNv4i32_shift,
-                                          SQSHRUNv16i8_shift, SQSHRUNv8i16_shift, SQSHRUNv4i32_shift,
-                                          SQXTNv16i8, SQXTNv8i16, SQXTNv4i32,
-                                          SQXTUNv16i8, SQXTUNv8i16, SQXTUNv4i32,
-                                          SSHLLv16i8_shift, SSHLLv8i16_shift, SSHLLv4i32_shift,
-                                          SSUBLv16i8_v8i16, SSUBLv8i16_v4i32, SSUBLv4i32_v2i64,
-                                          SSUBWv16i8_v8i16, SSUBWv8i16_v4i32, SSUBWv4i32_v2i64,
-                                          UABALv16i8_v8i16, UABALv8i16_v4i32, UABALv4i32_v2i64,
-                                          UABDLv16i8_v8i16, UABDLv8i16_v4i32, UABDLv4i32_v2i64,
-                                          UADDLv16i8_v8i16, UADDLv8i16_v4i32, UADDLv4i32_v2i64,
-                                          UADDWv16i8_v8i16, UADDWv8i16_v4i32, UADDWv4i32_v2i64,
-                                          UMLALv16i8_v8i16, UMLALv8i16_v4i32, UMLALv4i32_v2i64,
-                                          UMLALv8i16_indexed, UMLALv4i32_indexed,
-                                          UMLSLv16i8_v8i16, UMLSLv8i16_v4i32, UMLSLv4i32_v2i64,
-                                          UMLSLv8i16_indexed, UMLSLv4i32_indexed,
-                                          UMULLv16i8_v8i16, UMULLv8i16_v4i32, UMULLv4i32_v2i64,
-                                          UMULLv8i16_indexed, UMULLv4i32_indexed,
-                                          UQSHRNv16i8_shift, UQSHRNv8i16_shift, UQSHRNv4i32_shift,
-                                          UQXTNv16i8, UQXTNv8i16, UQXTNv4i32,
-                                          USHLLv16i8_shift, USHLLv8i16_shift, USHLLv4i32_shift,
-                                          USUBLv16i8_v8i16, USUBLv8i16_v4i32, USUBLv4i32_v2i64,
-                                          USUBWv16i8_v8i16, USUBWv8i16_v4i32, USUBWv4i32_v2i64,
-                                          XTNv16i8, XTNv8i16, XTNv4i32]>;
-
 // Target predicates.
 
 // Identify an instruction that effectively transfers a register to another.