[X86][AVX] Split WriteFLogic into XMM and YMM/ZMM scheduler classes

This removes all the AND/ANDN/OR/XOR PS/PD InstRW overrides.

llvm-svn: 331051

diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td
index 6046541..c39429e 100644 (file)
--- a/llvm/lib/Target/X86/X86InstrAVX512.td
+++ b/llvm/lib/Target/X86/X86InstrAVX512.td
@@ -4998,13 +4998,14 @@ multiclass avx512_fp_sae_packed<bits<8> opc, string OpcodeStr,
 
 multiclass avx512_fp_binop_p<bits<8> opc, string OpcodeStr, SDPatternOperator OpNode,
                              Predicate prd, X86FoldableSchedWrite sched,
+                             X86FoldableSchedWrite schedY,
                              bit IsCommutable = 0> {
   let Predicates = [prd] in {
   defm PSZ : avx512_fp_packed<opc, OpcodeStr, OpNode, v16f32_info,
-                              sched, IsCommutable>, EVEX_V512, PS,
+                              schedY, IsCommutable>, EVEX_V512, PS,
                               EVEX_CD8<32, CD8VF>;
   defm PDZ : avx512_fp_packed<opc, OpcodeStr, OpNode, v8f64_info,
-                              sched, IsCommutable>, EVEX_V512, PD, VEX_W,
+                              schedY, IsCommutable>, EVEX_V512, PD, VEX_W,
                               EVEX_CD8<64, CD8VF>;
   }
 
@@ -5014,13 +5015,13 @@ multiclass avx512_fp_binop_p<bits<8> opc, string OpcodeStr, SDPatternOperator Op
                                    sched, IsCommutable>, EVEX_V128, PS,
                                    EVEX_CD8<32, CD8VF>;
     defm PSZ256 : avx512_fp_packed<opc, OpcodeStr, OpNode, v8f32x_info,
-                                   sched, IsCommutable>, EVEX_V256, PS,
+                                   schedY, IsCommutable>, EVEX_V256, PS,
                                    EVEX_CD8<32, CD8VF>;
     defm PDZ128 : avx512_fp_packed<opc, OpcodeStr, OpNode, v2f64x_info,
                                    sched, IsCommutable>, EVEX_V128, PD, VEX_W,
                                    EVEX_CD8<64, CD8VF>;
     defm PDZ256 : avx512_fp_packed<opc, OpcodeStr, OpNode, v4f64x_info,
-                                   sched, IsCommutable>, EVEX_V256, PD, VEX_W,
+                                   schedY, IsCommutable>, EVEX_V256, PD, VEX_W,
                                    EVEX_CD8<64, CD8VF>;
   }
 }
@@ -5042,26 +5043,37 @@ multiclass avx512_fp_binop_p_sae<bits<8> opc, string OpcodeStr, SDNode OpNodeRnd
 }
 
 defm VADD : avx512_fp_binop_p<0x58, "vadd", fadd, HasAVX512,
-                              WriteFAdd, 1>,
+                              WriteFAdd, WriteFAdd, 1>,
             avx512_fp_binop_p_round<0x58, "vadd", X86faddRnd, WriteFAdd>;
-defm VMUL : avx512_fp_binop_p<0x59, "vmul", fmul, HasAVX512, WriteFMul, 1>,
+defm VMUL : avx512_fp_binop_p<0x59, "vmul", fmul, HasAVX512,
+                              WriteFMul, WriteFMul, 1>,
             avx512_fp_binop_p_round<0x59, "vmul", X86fmulRnd, WriteFMul>;
-defm VSUB : avx512_fp_binop_p<0x5C, "vsub", fsub, HasAVX512, WriteFAdd>,
+defm VSUB : avx512_fp_binop_p<0x5C, "vsub", fsub, HasAVX512,
+                              WriteFAdd, WriteFAdd>,
             avx512_fp_binop_p_round<0x5C, "vsub", X86fsubRnd, WriteFAdd>;
-defm VDIV : avx512_fp_binop_p<0x5E, "vdiv", fdiv, HasAVX512, WriteFDiv>,
+defm VDIV : avx512_fp_binop_p<0x5E, "vdiv", fdiv, HasAVX512,
+                              WriteFDiv, WriteFDiv>,
             avx512_fp_binop_p_round<0x5E, "vdiv", X86fdivRnd, WriteFDiv>;
-defm VMIN : avx512_fp_binop_p<0x5D, "vmin", X86fmin, HasAVX512, WriteFCmp, 0>,
+defm VMIN : avx512_fp_binop_p<0x5D, "vmin", X86fmin, HasAVX512,
+                              WriteFCmp, WriteFCmp, 0>,
             avx512_fp_binop_p_sae<0x5D, "vmin", X86fminRnd, WriteFCmp>;
-defm VMAX : avx512_fp_binop_p<0x5F, "vmax", X86fmax, HasAVX512, WriteFCmp, 0>,
+defm VMAX : avx512_fp_binop_p<0x5F, "vmax", X86fmax, HasAVX512,
+                              WriteFCmp, WriteFCmp, 0>,
             avx512_fp_binop_p_sae<0x5F, "vmax", X86fmaxRnd, WriteFCmp>;
 let isCodeGenOnly = 1 in {
-  defm VMINC : avx512_fp_binop_p<0x5D, "vmin", X86fminc, HasAVX512, WriteFCmp, 1>;
-  defm VMAXC : avx512_fp_binop_p<0x5F, "vmax", X86fmaxc, HasAVX512, WriteFCmp, 1>;
-}
-defm VAND  : avx512_fp_binop_p<0x54, "vand", null_frag, HasDQI,  WriteFLogic, 1>;
-defm VANDN : avx512_fp_binop_p<0x55, "vandn", null_frag, HasDQI, WriteFLogic, 0>;
-defm VOR   : avx512_fp_binop_p<0x56, "vor", null_frag, HasDQI,  WriteFLogic, 1>;
-defm VXOR  : avx512_fp_binop_p<0x57, "vxor", null_frag, HasDQI, WriteFLogic, 1>;
+  defm VMINC : avx512_fp_binop_p<0x5D, "vmin", X86fminc, HasAVX512,
+                                 WriteFCmp, WriteFCmp, 1>;
+  defm VMAXC : avx512_fp_binop_p<0x5F, "vmax", X86fmaxc, HasAVX512,
+                                 WriteFCmp, WriteFCmp, 1>;
+}
+defm VAND  : avx512_fp_binop_p<0x54, "vand", null_frag, HasDQI,
+                               WriteFLogic, WriteFLogicY, 1>;
+defm VANDN : avx512_fp_binop_p<0x55, "vandn", null_frag, HasDQI,
+                               WriteFLogic, WriteFLogicY, 0>;
+defm VOR   : avx512_fp_binop_p<0x56, "vor", null_frag, HasDQI,
+                               WriteFLogic, WriteFLogicY, 1>;
+defm VXOR  : avx512_fp_binop_p<0x57, "vxor", null_frag, HasDQI,
+                               WriteFLogic, WriteFLogicY, 1>;
 
 // Patterns catch floating point selects with bitcasted integer logic ops.
 multiclass avx512_fp_logical_lowering<string InstrStr, SDNode OpNode,
@@ -9860,9 +9872,9 @@ def : Pat<(vselect (v2i1 VK2WM:$mask), (v2f64 (X86VBroadcast (loadv2f64 addr:$sr
 //===----------------------------------------------------------------------===//
 
 defm VUNPCKH : avx512_fp_binop_p<0x15, "vunpckh", X86Unpckh, HasAVX512,
-                                 WriteFShuffle>;
+                                 WriteFShuffle, WriteFShuffle>;
 defm VUNPCKL : avx512_fp_binop_p<0x14, "vunpckl", X86Unpckl, HasAVX512,
-                                 WriteFShuffle>;
+                                 WriteFShuffle, WriteFShuffle>;
 
 defm VPUNPCKLBW : avx512_binop_rm_vl_b<0x60, "vpunpcklbw", X86Unpckl,
                                        WriteShuffle, HasBWI>;

diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td
index 68cc4c6..a3b75de 100644 (file)
--- a/llvm/lib/Target/X86/X86InstrSSE.td
+++ b/llvm/lib/Target/X86/X86InstrSSE.td
@@ -87,6 +87,7 @@ multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
                                       string OpcodeStr, X86MemOperand x86memop,
+                                      X86FoldableSchedWrite sched,
                                       list<dag> pat_rr, list<dag> pat_rm,
                                       bit Is2Addr = 1> {
   let isCommutable = 1, hasSideEffects = 0 in
@@ -95,14 +96,14 @@ multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
            !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
        pat_rr, d>,
-       Sched<[WriteFLogic]>;
+       Sched<[sched]>;
   let hasSideEffects = 0, mayLoad = 1 in
   def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
        !if(Is2Addr,
            !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
        pat_rm, d>,
-       Sched<[WriteFLogic.Folded, ReadAfterLd]>;
+       Sched<[sched.Folded, ReadAfterLd]>;
 }
 
 
@@ -2334,29 +2335,29 @@ multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
                                    SDNode OpNode> {
   let Predicates = [HasAVX, NoVLX] in {
   defm V#NAME#PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
-        !strconcat(OpcodeStr, "ps"), f256mem,
+        !strconcat(OpcodeStr, "ps"), f256mem, WriteFLogicY,
         [], [], 0>, PS, VEX_4V, VEX_L, VEX_WIG;
 
   defm V#NAME#PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
-        !strconcat(OpcodeStr, "pd"), f256mem,
+        !strconcat(OpcodeStr, "pd"), f256mem, WriteFLogicY,
         [], [], 0>, PD, VEX_4V, VEX_L, VEX_WIG;
 
   defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
-       !strconcat(OpcodeStr, "ps"), f128mem,
+       !strconcat(OpcodeStr, "ps"), f128mem, WriteFLogic,
        [], [], 0>, PS, VEX_4V, VEX_WIG;
 
   defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
-       !strconcat(OpcodeStr, "pd"), f128mem,
+       !strconcat(OpcodeStr, "pd"), f128mem, WriteFLogic,
        [], [], 0>, PD, VEX_4V, VEX_WIG;
   }
 
   let Constraints = "$src1 = $dst" in {
     defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
-         !strconcat(OpcodeStr, "ps"), f128mem,
+         !strconcat(OpcodeStr, "ps"), f128mem, WriteFLogic,
          [], []>, PS;
 
     defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
-         !strconcat(OpcodeStr, "pd"), f128mem,
+         !strconcat(OpcodeStr, "pd"), f128mem, WriteFLogic,
          [], []>, PD;
   }
 }

diff --git a/llvm/lib/Target/X86/X86SchedBroadwell.td b/llvm/lib/Target/X86/X86SchedBroadwell.td
index 72044bf..d21b9bd 100755 (executable)
--- a/llvm/lib/Target/X86/X86SchedBroadwell.td
+++ b/llvm/lib/Target/X86/X86SchedBroadwell.td
@@ -166,7 +166,8 @@ defm : BWWriteResPair<WriteFMA,    [BWPort01], 5, [1], 1, 5>; // Fused Multiply
 defm : BWWriteResPair<WriteFMAS,   [BWPort01], 5, [1], 1, 5>; // Fused Multiply Add (Scalar).
 defm : BWWriteResPair<WriteFMAY,   [BWPort01], 5, [1], 1, 6>; // Fused Multiply Add (YMM/ZMM).
 defm : BWWriteResPair<WriteFSign,     [BWPort5],  1>; // Floating point fabs/fchs.
-defm : BWWriteResPair<WriteFLogic,    [BWPort5],  1>; // Floating point and/or/xor logicals.
+defm : BWWriteResPair<WriteFLogic,    [BWPort5], 1, [1], 1, 5>; // Floating point and/or/xor logicals.
+defm : BWWriteResPair<WriteFLogicY,   [BWPort5], 1, [1], 1, 6>; // Floating point and/or/xor logicals (YMM/ZMM).
 defm : BWWriteResPair<WriteFShuffle,  [BWPort5],  1>; // Floating point vector shuffles.
 defm : BWWriteResPair<WriteFVarShuffle,  [BWPort5],  1>; // Floating point vector variable shuffles.
 defm : BWWriteResPair<WriteFBlend,  [BWPort015],  1>; // Floating point vector blends.
@@ -1090,13 +1091,7 @@ def BWWriteResGroup75 : SchedWriteRes<[BWPort5,BWPort23]> {
   let NumMicroOps = 2;
   let ResourceCycles = [1,1];
 }
-def: InstRW<[BWWriteResGroup75], (instregex "VANDNPDYrm",
-                                            "VANDNPSYrm",
-                                            "VANDPDYrm",
-                                            "VANDPSYrm",
-                                            "VORPDYrm",
-                                            "VORPSYrm",
-                                            "VPACKSSDWYrm",
+def: InstRW<[BWWriteResGroup75], (instregex "VPACKSSDWYrm",
                                             "VPACKSSWBYrm",
                                             "VPACKUSDWYrm",
                                             "VPACKUSWBYrm",
@@ -1123,9 +1118,7 @@ def: InstRW<[BWWriteResGroup75], (instregex "VANDNPDYrm",
                                             "VUNPCKHPDYrm",
                                             "VUNPCKHPSYrm",
                                             "VUNPCKLPDYrm",
-                                            "VUNPCKLPSYrm",
-                                            "VXORPDYrm",
-                                            "VXORPSYrm")>;
+                                            "VUNPCKLPSYrm")>;
 
 def BWWriteResGroup76 : SchedWriteRes<[BWPort23,BWPort15]> {
   let Latency = 7;

diff --git a/llvm/lib/Target/X86/X86SchedHaswell.td b/llvm/lib/Target/X86/X86SchedHaswell.td
index c2ea8b1..fb4d9b5 100644 (file)
--- a/llvm/lib/Target/X86/X86SchedHaswell.td
+++ b/llvm/lib/Target/X86/X86SchedHaswell.td
@@ -163,7 +163,8 @@ defm : HWWriteResPair<WriteFMA,   [HWPort01], 5, [1], 1, 6>;
 defm : HWWriteResPair<WriteFMAS,  [HWPort01], 5, [1], 1, 5>;
 defm : HWWriteResPair<WriteFMAY,  [HWPort01], 5, [1], 1, 7>;
 defm : HWWriteResPair<WriteFSign,  [HWPort0], 1>;
-defm : HWWriteResPair<WriteFLogic, [HWPort5], 1, [1], 1, 6>;
+defm : HWWriteResPair<WriteFLogic,  [HWPort5], 1, [1], 1, 6>;
+defm : HWWriteResPair<WriteFLogicY, [HWPort5], 1, [1], 1, 7>;
 defm : HWWriteResPair<WriteFShuffle,  [HWPort5],  1>;
 defm : HWWriteResPair<WriteFVarShuffle,  [HWPort5],  1>;
 defm : HWWriteResPair<WriteFBlend,  [HWPort015], 1, [1], 1, 6>;
@@ -910,13 +911,7 @@ def HWWriteResGroup13_1 : SchedWriteRes<[HWPort5,HWPort23]> {
   let NumMicroOps = 2;
   let ResourceCycles = [1,1];
 }
-def: InstRW<[HWWriteResGroup13_1], (instregex "VANDNPDYrm",
-                                              "VANDNPSYrm",
-                                              "VANDPDYrm",
-                                              "VANDPSYrm",
-                                              "VORPDYrm",
-                                              "VORPSYrm",
-                                              "VPACKSSDWYrm",
+def: InstRW<[HWWriteResGroup13_1], (instregex "VPACKSSDWYrm",
                                               "VPACKSSWBYrm",
                                               "VPACKUSDWYrm",
                                               "VPACKUSWBYrm",
@@ -946,9 +941,7 @@ def: InstRW<[HWWriteResGroup13_1], (instregex "VANDNPDYrm",
                                               "VUNPCKHPDYrm",
                                               "VUNPCKHPSYrm",
                                               "VUNPCKLPDYrm",
-                                              "VUNPCKLPSYrm",
-                                              "VXORPDYrm",
-                                              "VXORPSYrm")>;
+                                              "VUNPCKLPSYrm")>;
 
 def HWWriteResGroup14 : SchedWriteRes<[HWPort6,HWPort23]> {
   let Latency = 6;

diff --git a/llvm/lib/Target/X86/X86SchedSandyBridge.td b/llvm/lib/Target/X86/X86SchedSandyBridge.td
index 55475f7..478d886 100644 (file)
--- a/llvm/lib/Target/X86/X86SchedSandyBridge.td
+++ b/llvm/lib/Target/X86/X86SchedSandyBridge.td
@@ -151,6 +151,7 @@ defm : SBWriteResPair<WriteCvtI2F, [SBPort1], 4>;
 defm : SBWriteResPair<WriteCvtF2F, [SBPort1], 3>;
 defm : SBWriteResPair<WriteFSign,    [SBPort5], 1>;
 defm : SBWriteResPair<WriteFLogic,   [SBPort5], 1, [1], 1, 6>;
+defm : SBWriteResPair<WriteFLogicY,  [SBPort5], 1, [1], 1, 7>;
 defm : SBWriteResPair<WriteFShuffle, [SBPort5],  1>;
 defm : SBWriteResPair<WriteFVarShuffle, [SBPort5],  1>;
 defm : SBWriteResPair<WriteFBlend,    [SBPort05], 1, [1], 1, 6>;
@@ -1142,13 +1143,7 @@ def SBWriteResGroup73 : SchedWriteRes<[SBPort5,SBPort23]> {
   let NumMicroOps = 2;
   let ResourceCycles = [1,1];
 }
-def: InstRW<[SBWriteResGroup73], (instregex "VANDNPDYrm",
-                                            "VANDNPSYrm",
-                                            "VANDPDYrm",
-                                            "VANDPSYrm",
-                                            "VORPDYrm",
-                                            "VORPSYrm",
-                                            "VPERM2F128rm",
+def: InstRW<[SBWriteResGroup73], (instregex "VPERM2F128rm",
                                             "VPERMILPDYmi",
                                             "VPERMILPDYrm",
                                             "VPERMILPSYmi",
@@ -1158,9 +1153,7 @@ def: InstRW<[SBWriteResGroup73], (instregex "VANDNPDYrm",
                                             "VUNPCKHPDYrm",
                                             "VUNPCKHPSYrm",
                                             "VUNPCKLPDYrm",
-                                            "VUNPCKLPSYrm",
-                                            "VXORPDYrm",
-                                            "VXORPSYrm")>;
+                                            "VUNPCKLPSYrm")>;
 
 def SBWriteResGroup74 : SchedWriteRes<[SBPort23,SBPort05]> {
   let Latency = 8;

diff --git a/llvm/lib/Target/X86/X86SchedSkylakeClient.td b/llvm/lib/Target/X86/X86SchedSkylakeClient.td
index bd7c37d..4b51cc8 100644 (file)
--- a/llvm/lib/Target/X86/X86SchedSkylakeClient.td
+++ b/llvm/lib/Target/X86/X86SchedSkylakeClient.td
@@ -163,7 +163,8 @@ defm : SKLWriteResPair<WriteFMA,    [SKLPort01], 4, [1], 1, 6>; // Fused Multipl
 defm : SKLWriteResPair<WriteFMAS,   [SKLPort01], 4, [1], 1, 5>; // Fused Multiply Add (Scalar).
 defm : SKLWriteResPair<WriteFMAY,   [SKLPort01], 4, [1], 1, 7>; // Fused Multiply Add (YMM/ZMM).
 defm : SKLWriteResPair<WriteFSign,   [SKLPort0], 1>; // Floating point fabs/fchs.
-defm : SKLWriteResPair<WriteFLogic, [SKLPort015], 1, [1], 1, 6>; // Floating point and/or/xor logicals.
+defm : SKLWriteResPair<WriteFLogic,  [SKLPort015], 1, [1], 1, 6>; // Floating point and/or/xor logicals.
+defm : SKLWriteResPair<WriteFLogicY, [SKLPort015], 1, [1], 1, 7>; // Floating point and/or/xor logicals (YMM/ZMM).
 defm : SKLWriteResPair<WriteFShuffle,  [SKLPort5],  1>; // Floating point vector shuffles.
 defm : SKLWriteResPair<WriteFVarShuffle,  [SKLPort5],  1>; // Floating point vector shuffles.
 defm : SKLWriteResPair<WriteFBlend,  [SKLPort015], 1, [1], 1, 6>; // Floating point vector blends.
@@ -1624,16 +1625,10 @@ def SKLWriteResGroup110 : SchedWriteRes<[SKLPort23,SKLPort015]> {
   let NumMicroOps = 2;
   let ResourceCycles = [1,1];
 }
-def: InstRW<[SKLWriteResGroup110], (instregex "VANDNPDYrm",
-                                              "VANDNPSYrm",
-                                              "VANDPDYrm",
-                                              "VANDPSYrm",
-                                              "VBLENDPDYrmi",
+def: InstRW<[SKLWriteResGroup110], (instregex "VBLENDPDYrmi",
                                               "VBLENDPSYrmi",
                                               "VMASKMOVPDYrm",
                                               "VMASKMOVPSYrm",
-                                              "VORPDYrm",
-                                              "VORPSYrm",
                                               "VPADDBYrm",
                                               "VPADDDYrm",
                                               "VPADDQYrm",
@@ -1648,9 +1643,7 @@ def: InstRW<[SKLWriteResGroup110], (instregex "VANDNPDYrm",
                                               "VPSUBDYrm",
                                               "VPSUBQYrm",
                                               "VPSUBWYrm",
-                                              "VPXORYrm",
-                                              "VXORPDYrm",
-                                              "VXORPSYrm")>;
+                                              "VPXORYrm")>;
 
 def SKLWriteResGroup112 : SchedWriteRes<[SKLPort0,SKLPort5,SKLPort23]> {
   let Latency = 8;

diff --git a/llvm/lib/Target/X86/X86SchedSkylakeServer.td b/llvm/lib/Target/X86/X86SchedSkylakeServer.td
index b74b8f3..2c25e91 100755 (executable)
--- a/llvm/lib/Target/X86/X86SchedSkylakeServer.td
+++ b/llvm/lib/Target/X86/X86SchedSkylakeServer.td
@@ -164,6 +164,7 @@ defm : SKXWriteResPair<WriteFMAS, [SKXPort015], 4, [1], 1, 5>; // Fused Multiply
 defm : SKXWriteResPair<WriteFMAY, [SKXPort015], 4, [1], 1, 7>; // Fused Multiply Add (YMM/ZMM).
 defm : SKXWriteResPair<WriteFSign,  [SKXPort0], 1>; // Floating point fabs/fchs.
 defm : SKXWriteResPair<WriteFLogic, [SKXPort015], 1, [1], 1, 6>; // Floating point and/or/xor logicals.
+defm : SKXWriteResPair<WriteFLogicY, [SKXPort015], 1, [1], 1, 7>; // Floating point and/or/xor logicals (YMM/ZMM).
 defm : SKXWriteResPair<WriteFShuffle,  [SKXPort5],  1>; // Floating point vector shuffles.
 defm : SKXWriteResPair<WriteFVarShuffle,  [SKXPort5],  1>; // Floating point vector variable shuffles.
 defm : SKXWriteResPair<WriteFBlend, [SKXPort015], 1, [1], 1, 6>; // Floating point vector blends.
@@ -3165,19 +3166,7 @@ def SKXWriteResGroup121 : SchedWriteRes<[SKXPort23,SKXPort015]> {
   let NumMicroOps = 2;
   let ResourceCycles = [1,1];
 }
-def: InstRW<[SKXWriteResGroup121], (instregex "VANDNPDYrm",
-                                              "VANDNPDZ256rm(b?)",
-                                              "VANDNPDZrm(b?)",
-                                              "VANDNPSYrm",
-                                              "VANDNPSZ256rm(b?)",
-                                              "VANDNPSZrm(b?)",
-                                              "VANDPDYrm",
-                                              "VANDPDZ256rm(b?)",
-                                              "VANDPDZrm(b?)",
-                                              "VANDPSYrm",
-                                              "VANDPSZ256rm(b?)",
-                                              "VANDPSZrm(b?)",
-                                              "VBLENDMPDZ256rm(b?)",
+def: InstRW<[SKXWriteResGroup121], (instregex "VBLENDMPDZ256rm(b?)",
                                               "VBLENDMPDZrm(b?)",
                                               "VBLENDMPSZ256rm(b?)",
                                               "VBLENDMPSZrm(b?)",
@@ -3244,12 +3233,6 @@ def: InstRW<[SKXWriteResGroup121], (instregex "VANDNPDYrm",
                                               "VMOVUPDZrm(b?)",
                                               "VMOVUPSZ256rm(b?)",
                                               "VMOVUPSZrm(b?)",
-                                              "VORPDYrm",
-                                              "VORPDZ256rm(b?)",
-                                              "VORPDZrm(b?)",
-                                              "VORPSYrm",
-                                              "VORPSZ256rm(b?)",
-                                              "VORPSZrm(b?)",
                                               "VPADDBYrm",
                                               "VPADDBZ256rm(b?)",
                                               "VPADDBZrm(b?)",
@@ -3311,13 +3294,7 @@ def: InstRW<[SKXWriteResGroup121], (instregex "VANDNPDYrm",
                                               "VPXORDZrm(b?)",
                                               "VPXORQZ256rm(b?)",
                                               "VPXORQZrm(b?)",
-                                              "VPXORYrm",
-                                              "VXORPDYrm",
-                                              "VXORPDZ256rm(b?)",
-                                              "VXORPDZrm(b?)",
-                                              "VXORPSYrm",
-                                              "VXORPSZ256rm(b?)",
-                                              "VXORPSZrm(b?)")>;
+                                              "VPXORYrm")>;
 
 def SKXWriteResGroup123 : SchedWriteRes<[SKXPort0,SKXPort5,SKXPort23]> {
   let Latency = 8;

diff --git a/llvm/lib/Target/X86/X86Schedule.td b/llvm/lib/Target/X86/X86Schedule.td
index 31d14f6..e87475a 100644 (file)
--- a/llvm/lib/Target/X86/X86Schedule.td
+++ b/llvm/lib/Target/X86/X86Schedule.td
@@ -89,8 +89,9 @@ defm WriteFRsqrt : X86SchedWritePair; // Floating point reciprocal square root e
 defm WriteFMA    : X86SchedWritePair; // Fused Multiply Add.
 defm WriteFMAS   : X86SchedWritePair; // Fused Multiply Add (Scalar).
 defm WriteFMAY   : X86SchedWritePair; // Fused Multiply Add (YMM/ZMM).
-defm WriteFSign : X86SchedWritePair; // Floating point fabs/fchs.
-defm WriteFLogic : X86SchedWritePair; // Floating point and/or/xor logicals.
+defm WriteFSign  : X86SchedWritePair; // Floating point fabs/fchs.
+defm WriteFLogic  : X86SchedWritePair; // Floating point and/or/xor logicals.
+defm WriteFLogicY : X86SchedWritePair; // Floating point and/or/xor logicals (YMM/ZMM).
 defm WriteFShuffle  : X86SchedWritePair; // Floating point vector shuffles.
 defm WriteFVarShuffle  : X86SchedWritePair; // Floating point vector variable shuffles.
 defm WriteFBlend  : X86SchedWritePair; // Floating point vector blends.

diff --git a/llvm/lib/Target/X86/X86ScheduleAtom.td b/llvm/lib/Target/X86/X86ScheduleAtom.td
index 9dd23b6..23ede68 100644 (file)
--- a/llvm/lib/Target/X86/X86ScheduleAtom.td
+++ b/llvm/lib/Target/X86/X86ScheduleAtom.td
@@ -212,6 +212,7 @@ defm : AtomWriteResPair<WriteFDiv,          [AtomPort01], [AtomPort01], 34, 34,
 defm : AtomWriteResPair<WriteFSqrt,         [AtomPort01], [AtomPort01], 34, 34, [34], [34]>;
 defm : AtomWriteResPair<WriteFSign,          [AtomPort1],  [AtomPort1]>;
 defm : AtomWriteResPair<WriteFLogic,        [AtomPort01],  [AtomPort0]>;
+defm : AtomWriteResPair<WriteFLogicY,       [AtomPort01],  [AtomPort0]>; // NOTE: Doesn't exist on Atom.
 defm : AtomWriteResPair<WriteFShuffle,       [AtomPort0],  [AtomPort0]>;
 defm : AtomWriteResPair<WriteFVarShuffle,    [AtomPort0],  [AtomPort0]>; // NOTE: Doesn't exist on Atom.
 defm : AtomWriteResPair<WriteFMA,            [AtomPort0],  [AtomPort0]>; // NOTE: Doesn't exist on Atom.

diff --git a/llvm/lib/Target/X86/X86ScheduleBtVer2.td b/llvm/lib/Target/X86/X86ScheduleBtVer2.td
index 6f8eb9b..28d54b9 100644 (file)
--- a/llvm/lib/Target/X86/X86ScheduleBtVer2.td
+++ b/llvm/lib/Target/X86/X86ScheduleBtVer2.td
@@ -129,6 +129,25 @@ multiclass JWriteResFpuPair<X86FoldableSchedWrite SchedRW,
   }
 }
 
+multiclass JWriteResYMMPair<X86FoldableSchedWrite SchedRW,
+                            list<ProcResourceKind> ExePorts,
+                            int Lat, list<int> Res = [2], int UOps = 2> {
+  // Register variant is using a single cycle on ExePort.
+  def : WriteRes<SchedRW, ExePorts> {
+    let Latency = Lat;
+    let ResourceCycles = Res;
+    let NumMicroOps = UOps;
+  }
+
+  // Memory variant also uses 2 cycles on JLAGU and adds 5 cycles to the
+  // latency.
+  def : WriteRes<SchedRW.Folded, !listconcat([JLAGU], ExePorts)> {
+    let Latency = !add(Lat, 5);
+    let ResourceCycles = !listconcat([2], Res);
+    let NumMicroOps = UOps;
+  }
+}
+
 // A folded store needs a cycle on the SAGU for the store data.
 def : WriteRes<WriteRMW, [JSAGU]>;
 
@@ -309,6 +328,7 @@ defm : JWriteResFpuPair<WriteFDiv,         [JFPU1, JFPM], 19, [1, 19]>;
 defm : JWriteResFpuPair<WriteFSqrt,        [JFPU1, JFPM], 21, [1, 21]>;
 defm : JWriteResFpuPair<WriteFSign,        [JFPU1, JFPM],  2>;
 defm : JWriteResFpuPair<WriteFLogic,      [JFPU01, JFPX],  1>;
+defm : JWriteResYMMPair<WriteFLogicY,     [JFPU01, JFPX],  1, [2, 2], 2>;
 defm : JWriteResFpuPair<WriteFShuffle,    [JFPU01, JFPX],  1>;
 defm : JWriteResFpuPair<WriteFVarShuffle, [JFPU01, JFPX],  2, [1, 4], 3>;
 defm : JWriteResFpuPair<WriteFBlend,      [JFPU01, JFPX],  1>;
@@ -527,25 +547,6 @@ def : InstRW<[JWriteCVTPH2PSYLd], (instrs VCVTPH2PSYrm)>;
 // AVX instructions.
 ////////////////////////////////////////////////////////////////////////////////
 
-def JWriteFLogicY: SchedWriteRes<[JFPU01, JFPX]> {
-  let ResourceCycles = [2, 2];
-  let NumMicroOps = 2;
-}
-def : InstRW<[JWriteFLogicY], (instrs VORPDYrr, VORPSYrr,
-                                      VXORPDYrr, VXORPSYrr,
-                                      VANDPDYrr, VANDPSYrr,
-                                      VANDNPDYrr, VANDNPSYrr)>;
-
-def JWriteFLogicYLd: SchedWriteRes<[JLAGU, JFPU01, JFPX]> {
-  let Latency = 6;
-  let ResourceCycles = [2, 2, 2];
-  let NumMicroOps = 2;
-}
-def : InstRW<[JWriteFLogicYLd, ReadAfterLd], (instrs VORPDYrm, VORPSYrm,
-                                                     VXORPDYrm, VXORPSYrm,
-                                                     VANDPDYrm, VANDPSYrm,
-                                                     VANDNPDYrm, VANDNPSYrm)>;
-
 def JWriteVDPPSY: SchedWriteRes<[JFPU1, JFPM, JFPA]> {
   let Latency = 12;
   let ResourceCycles = [2, 6, 6];

diff --git a/llvm/lib/Target/X86/X86ScheduleSLM.td b/llvm/lib/Target/X86/X86ScheduleSLM.td
index 8663d2f..0504519 100644 (file)
--- a/llvm/lib/Target/X86/X86ScheduleSLM.td
+++ b/llvm/lib/Target/X86/X86ScheduleSLM.td
@@ -142,6 +142,7 @@ defm : SLMWriteResPair<WriteCvtI2F, [SLM_FPC_RSV01], 4>;
 defm : SLMWriteResPair<WriteCvtF2F, [SLM_FPC_RSV01], 4>;
 defm : SLMWriteResPair<WriteFSign,  [SLM_FPC_RSV01], 1>;
 defm : SLMWriteResPair<WriteFLogic, [SLM_FPC_RSV01], 1>;
+defm : SLMWriteResPair<WriteFLogicY, [SLM_FPC_RSV01], 1>;
 defm : SLMWriteResPair<WriteFShuffle, [SLM_FPC_RSV0],  1>;
 defm : SLMWriteResPair<WriteFVarShuffle, [SLM_FPC_RSV0],  1>;
 defm : SLMWriteResPair<WriteFBlend,  [SLM_FPC_RSV0],  1>;

diff --git a/llvm/lib/Target/X86/X86ScheduleZnver1.td b/llvm/lib/Target/X86/X86ScheduleZnver1.td
index 9836677..83358a7 100644 (file)
--- a/llvm/lib/Target/X86/X86ScheduleZnver1.td
+++ b/llvm/lib/Target/X86/X86ScheduleZnver1.td
@@ -204,6 +204,7 @@ defm : ZnWriteResFpuPair<WriteCvtF2I,    [ZnFPU3],  5>;
 defm : ZnWriteResFpuPair<WriteFDiv,      [ZnFPU3], 15>;
 defm : ZnWriteResFpuPair<WriteFSign,     [ZnFPU3],  2>;
 defm : ZnWriteResFpuPair<WriteFLogic,    [ZnFPU],   1>;
+defm : ZnWriteResFpuPair<WriteFLogicY,   [ZnFPU],   1>;
 defm : ZnWriteResFpuPair<WriteFShuffle,  [ZnFPU12], 1>;
 defm : ZnWriteResFpuPair<WriteFVarShuffle, [ZnFPU12], 1>;
 defm : ZnWriteResFpuPair<WriteFMul,      [ZnFPU0],  5>;

diff --git a/llvm/test/CodeGen/X86/avx512-schedule.ll b/llvm/test/CodeGen/X86/avx512-schedule.ll
index c39827c..7fbc517 100755 (executable)
--- a/llvm/test/CodeGen/X86/avx512-schedule.ll
+++ b/llvm/test/CodeGen/X86/avx512-schedule.ll
@@ -657,7 +657,7 @@ define <8 x i64> @addq_broadcast(<8 x i64> %a) nounwind {
 define <8 x i64> @orq_broadcast(<8 x i64> %a) nounwind {
 ; GENERIC-LABEL: orq_broadcast:
 ; GENERIC:       # %bb.0:
-; GENERIC-NEXT:    vorpd {{.*}}(%rip){1to8}, %zmm0, %zmm0 # sched: [7:1.00]
+; GENERIC-NEXT:    vorpd {{.*}}(%rip){1to8}, %zmm0, %zmm0 # sched: [8:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: orq_broadcast:
@@ -671,7 +671,7 @@ define <8 x i64> @orq_broadcast(<8 x i64> %a) nounwind {
 define <16 x i32> @andd512fold(<16 x i32> %y, <16 x i32>* %x) {
 ; GENERIC-LABEL: andd512fold:
 ; GENERIC:       # %bb.0: # %entry
-; GENERIC-NEXT:    vandps (%rdi), %zmm0, %zmm0 # sched: [7:1.00]
+; GENERIC-NEXT:    vandps (%rdi), %zmm0, %zmm0 # sched: [8:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: andd512fold:
@@ -687,7 +687,7 @@ entry:
 define <8 x i64> @andqbrst(<8 x i64> %p1, i64* %ap) {
 ; GENERIC-LABEL: andqbrst:
 ; GENERIC:       # %bb.0: # %entry
-; GENERIC-NEXT:    vandpd (%rdi){1to8}, %zmm0, %zmm0 # sched: [7:1.00]
+; GENERIC-NEXT:    vandpd (%rdi){1to8}, %zmm0, %zmm0 # sched: [8:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: andqbrst:
@@ -994,7 +994,7 @@ define <8 x double> @test_maskz_broadcast_vaddpd(<8 x double> %i, double* %j,
 define <16 x float>  @test_fxor(<16 x float> %a) {
 ; GENERIC-LABEL: test_fxor:
 ; GENERIC:       # %bb.0:
-; GENERIC-NEXT:    vxorps {{.*}}(%rip){1to16}, %zmm0, %zmm0 # sched: [7:1.00]
+; GENERIC-NEXT:    vxorps {{.*}}(%rip){1to16}, %zmm0, %zmm0 # sched: [8:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: test_fxor:
@@ -1009,7 +1009,7 @@ define <16 x float>  @test_fxor(<16 x float> %a) {
 define <8 x float>  @test_fxor_8f32(<8 x float> %a) {
 ; GENERIC-LABEL: test_fxor_8f32:
 ; GENERIC:       # %bb.0:
-; GENERIC-NEXT:    vxorps {{.*}}(%rip){1to8}, %ymm0, %ymm0 # sched: [7:1.00]
+; GENERIC-NEXT:    vxorps {{.*}}(%rip){1to8}, %ymm0, %ymm0 # sched: [8:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: test_fxor_8f32:
@@ -1023,7 +1023,7 @@ define <8 x float>  @test_fxor_8f32(<8 x float> %a) {
 define <8 x double> @fabs_v8f64(<8 x double> %p)
 ; GENERIC-LABEL: fabs_v8f64:
 ; GENERIC:       # %bb.0:
-; GENERIC-NEXT:    vandpd {{.*}}(%rip){1to8}, %zmm0, %zmm0 # sched: [7:1.00]
+; GENERIC-NEXT:    vandpd {{.*}}(%rip){1to8}, %zmm0, %zmm0 # sched: [8:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: fabs_v8f64:
@@ -1039,7 +1039,7 @@ declare <8 x double> @llvm.fabs.v8f64(<8 x double> %p)
 define <16 x float> @fabs_v16f32(<16 x float> %p)
 ; GENERIC-LABEL: fabs_v16f32:
 ; GENERIC:       # %bb.0:
-; GENERIC-NEXT:    vandps {{.*}}(%rip){1to16}, %zmm0, %zmm0 # sched: [7:1.00]
+; GENERIC-NEXT:    vandps {{.*}}(%rip){1to16}, %zmm0, %zmm0 # sched: [8:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: fabs_v16f32:
@@ -4809,7 +4809,7 @@ define <16 x float> @test_x86_fnmsub_ps_z(<16 x float> %a0, <16 x float> %a1, <1
 ; GENERIC-LABEL: test_x86_fnmsub_ps_z:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vmulps %zmm1, %zmm0, %zmm0 # sched: [5:1.00]
-; GENERIC-NEXT:    vxorps {{.*}}(%rip){1to16}, %zmm0, %zmm0 # sched: [7:1.00]
+; GENERIC-NEXT:    vxorps {{.*}}(%rip){1to16}, %zmm0, %zmm0 # sched: [8:1.00]
 ; GENERIC-NEXT:    vsubps %zmm2, %zmm0, %zmm0 # sched: [3:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;