[X86] Split int2double and float2double scheduler classes on Haswell/Broadwell to...

Haswell/Broadwell have numerous conversion instructions that use different scheduler pipes for the reg-reg and reg-mem variants (and not an additional Port23 uop for memory folding) - so declare the classes separately instead of using the HWWriteResPair/BWWriteResPair helpers

diff --git a/llvm/lib/Target/X86/X86SchedBroadwell.td b/llvm/lib/Target/X86/X86SchedBroadwell.td
index 0040d0e..9ffc4d1 100644 (file)
--- a/llvm/lib/Target/X86/X86SchedBroadwell.td
+++ b/llvm/lib/Target/X86/X86SchedBroadwell.td
@@ -366,17 +366,21 @@ defm : BWWriteResPair<WriteCvtPD2I,   [BWPort1,BWPort5], 4, [1,1], 2, 5>;
 defm : BWWriteResPair<WriteCvtPD2IY,  [BWPort1,BWPort5], 6, [1,1], 2, 6>;
 defm : X86WriteResPairUnsupported<WriteCvtPD2IZ>;
 
-defm : BWWriteResPair<WriteCvtI2SS,   [BWPort1], 4>;
+defm : X86WriteRes<WriteCvtI2SS,      [BWPort1,BWPort5], 4, [1,1], 2>;
+defm : X86WriteRes<WriteCvtI2SSLd,   [BWPort1,BWPort23], 9, [1,1], 2>;
 defm : BWWriteResPair<WriteCvtI2PS,   [BWPort1], 3>;
 defm : BWWriteResPair<WriteCvtI2PSY,  [BWPort1], 3, [1], 1, 6>;
 defm : X86WriteResPairUnsupported<WriteCvtI2PSZ>;
-defm : BWWriteResPair<WriteCvtI2SD,   [BWPort1], 4>;
+defm : X86WriteRes<WriteCvtI2SD,      [BWPort1,BWPort5], 4, [1,1], 2>;
+defm : X86WriteRes<WriteCvtI2SDLd,   [BWPort1,BWPort23], 9, [1,1], 2>;
 defm : BWWriteResPair<WriteCvtI2PD,   [BWPort1,BWPort5], 4, [1,1], 2, 5>;
 defm : BWWriteResPair<WriteCvtI2PDY,  [BWPort1,BWPort5], 6, [1,1], 2, 5>;
 defm : X86WriteResPairUnsupported<WriteCvtI2PDZ>;
 
-defm : BWWriteResPair<WriteCvtSS2SD,  [BWPort0,BWPort5], 2, [1,1], 2, 5>;
-defm : BWWriteResPair<WriteCvtPS2PD,  [BWPort0,BWPort5], 2, [1,1], 2, 5>;
+defm : X86WriteRes<WriteCvtSS2SD,     [BWPort0,BWPort5], 2, [1,1], 2>;
+defm : X86WriteRes<WriteCvtSS2SDLd,  [BWPort0,BWPort23], 6, [1,1], 2>;
+defm : X86WriteRes<WriteCvtPS2PD,     [BWPort0,BWPort5], 2, [1,1], 2>;
+defm : X86WriteRes<WriteCvtPS2PDLd,  [BWPort0,BWPort23], 6, [1,1], 2>;
 defm : BWWriteResPair<WriteCvtPS2PDY, [BWPort0,BWPort5], 4, [1,1], 2, 5>;
 defm : X86WriteResPairUnsupported<WriteCvtPS2PDZ>;
 defm : BWWriteResPair<WriteCvtSD2SS,  [BWPort1,BWPort5], 4, [1,1], 2, 5>;
@@ -851,10 +855,7 @@ def BWWriteResGroup42 : SchedWriteRes<[BWPort1,BWPort5]> {
   let NumMicroOps = 2;
   let ResourceCycles = [1,1];
 }
-def: InstRW<[BWWriteResGroup42], (instregex "MMX_CVT(T?)PS2PIrr",
-                                            "(V?)CVTSI642SDrr",
-                                            "(V?)CVTSI2SDrr",
-                                            "(V?)CVTSI2SSrr")>;
+def: InstRW<[BWWriteResGroup42], (instregex "MMX_CVT(T?)PS2PIrr")>;
 
 def BWWriteResGroup43 : SchedWriteRes<[BWPort0,BWPort4,BWPort237]> {
   let Latency = 4;
@@ -960,11 +961,7 @@ def BWWriteResGroup59 : SchedWriteRes<[BWPort0,BWPort23]> {
   let NumMicroOps = 2;
   let ResourceCycles = [1,1];
 }
-def: InstRW<[BWWriteResGroup59], (instrs CVTPS2PDrm, VCVTPS2PDrm,
-                                         CVTSS2SDrm, VCVTSS2SDrm,
-                                         CVTSS2SDrm_Int, VCVTSS2SDrm_Int,
-                                         VPSLLVQrm,
-                                         VPSRLVQrm)>;
+def: InstRW<[BWWriteResGroup59], (instrs VPSLLVQrm, VPSRLVQrm)>;
 
 def BWWriteResGroup62 : SchedWriteRes<[BWPort6,BWPort23]> {
   let Latency = 6;

diff --git a/llvm/lib/Target/X86/X86SchedHaswell.td b/llvm/lib/Target/X86/X86SchedHaswell.td
index c0c3826..5f47a7d 100644 (file)
--- a/llvm/lib/Target/X86/X86SchedHaswell.td
+++ b/llvm/lib/Target/X86/X86SchedHaswell.td
@@ -364,17 +364,21 @@ defm : HWWriteResPair<WriteCvtPS2I,   [HWPort1], 3, [1], 1, 6>;
 defm : HWWriteResPair<WriteCvtPS2IY,  [HWPort1], 3, [1], 1, 7>;
 defm : HWWriteResPair<WriteCvtPS2IZ,  [HWPort1], 3, [1], 1, 7>; // Unsupported = 1
 
-defm : HWWriteResPair<WriteCvtI2SD,   [HWPort1], 4>;
+defm : X86WriteRes<WriteCvtI2SD,      [HWPort1,HWPort5], 4, [1,1], 2>;
+defm : X86WriteRes<WriteCvtI2SDLd,   [HWPort1,HWPort23], 9, [1,1], 2>;
 defm : HWWriteResPair<WriteCvtI2PD,   [HWPort1,HWPort5], 4, [1,1], 2, 6>;
 defm : HWWriteResPair<WriteCvtI2PDY,  [HWPort1,HWPort5], 6, [1,1], 2, 6>;
 defm : HWWriteResPair<WriteCvtI2PDZ,  [HWPort1,HWPort5], 6, [1,1], 2, 6>; // Unsupported = 1
-defm : HWWriteResPair<WriteCvtI2SS,   [HWPort1], 4>;
+defm : X86WriteRes<WriteCvtI2SS,      [HWPort1,HWPort5], 4, [1,1], 2>;
+defm : X86WriteRes<WriteCvtI2SSLd,   [HWPort1,HWPort23], 9, [1,1], 2>;
 defm : HWWriteResPair<WriteCvtI2PS,   [HWPort1], 3, [1], 1, 6>;
 defm : HWWriteResPair<WriteCvtI2PSY,  [HWPort1], 3, [1], 1, 7>;
 defm : HWWriteResPair<WriteCvtI2PSZ,  [HWPort1], 3, [1], 1, 7>; // Unsupported = 1
 
-defm : HWWriteResPair<WriteCvtSS2SD,  [HWPort0,HWPort5], 2, [1,1], 2, 5>;
-defm : HWWriteResPair<WriteCvtPS2PD,  [HWPort0,HWPort5], 2, [1,1], 2, 5>;
+defm : X86WriteRes<WriteCvtSS2SD,     [HWPort0,HWPort5], 2, [1,1], 2>;
+defm : X86WriteRes<WriteCvtSS2SDLd,  [HWPort0,HWPort23], 7, [1,1], 2>;
+defm : X86WriteRes<WriteCvtPS2PD,     [HWPort0,HWPort5], 2, [1,1], 2>;
+defm : X86WriteRes<WriteCvtPS2PDLd,  [HWPort0,HWPort23], 6, [1,1], 2>;
 defm : HWWriteResPair<WriteCvtPS2PDY, [HWPort0,HWPort5], 4, [1,1], 2, 6>;
 defm : HWWriteResPair<WriteCvtPS2PDZ, [HWPort0,HWPort5], 4, [1,1], 2, 6>; // Unsupported = 1
 defm : HWWriteResPair<WriteCvtSD2SS,  [HWPort1,HWPort5], 4, [1,1], 2, 5>;
@@ -961,20 +965,12 @@ def: InstRW<[HWWriteResGroup10], (instrs CBW, CWDE, CDQE,
                                          STRm,
                                          SYSCALL)>;
 
-def HWWriteResGroup11 : SchedWriteRes<[HWPort0,HWPort23]> {
-  let Latency = 6;
-  let NumMicroOps = 2;
-  let ResourceCycles = [1,1];
-}
-def: InstRW<[HWWriteResGroup11], (instregex "(V?)CVTPS2PDrm")>;
-
 def HWWriteResGroup11_1 : SchedWriteRes<[HWPort0,HWPort23]> {
   let Latency = 7;
   let NumMicroOps = 2;
   let ResourceCycles = [1,1];
 }
 def: InstRW<[HWWriteResGroup11_1], (instrs VPSLLVQrm, VPSRLVQrm)>;
-def: InstRW<[HWWriteResGroup11_1], (instregex "(V?)CVTSS2SDrm")>;
 
 def HWWriteResGroup11_2 : SchedWriteRes<[HWPort0,HWPort23]> {
   let Latency = 8;
@@ -1346,8 +1342,6 @@ def HWWriteResGroup73 : SchedWriteRes<[HWPort1,HWPort5]> {
 }
 def: InstRW<[HWWriteResGroup73], (instrs MMX_CVTPS2PIrr,
                                          MMX_CVTTPS2PIrr)>;
-def: InstRW<[HWWriteResGroup73], (instregex "(V?)CVTSI(64)?2SDrr",
-                                            "(V?)CVTSI2SSrr")>;
 
 def HWWriteResGroup75 : SchedWriteRes<[HWPort1,HWPort23]> {
   let Latency = 11;