[X86][Btver2] Add correct mul/imul schedule costs

Integer multiply is performed on the JMul function unit and i64 requires double pumping

llvm-svn: 327707

1 files changed, 14 insertions, 1 deletions

diff --git a/llvm/lib/Target/X86/X86ScheduleBtVer2.td b/llvm/lib/Target/X86/X86ScheduleBtVer2.td
index 88a85609fbd..58c67e63180 100644
--- a/llvm/lib/Target/X86/X86ScheduleBtVer2.td
+++ b/llvm/lib/Target/X86/X86ScheduleBtVer2.td
@@ -117,7 +117,7 @@ def : WriteRes<WriteRMW, [JSAGU]>;
 ////////////////////////////////////////////////////////////////////////////////
 
 defm : JWriteResIntPair<WriteALU,   [JALU01], 1>;
-defm : JWriteResIntPair<WriteIMul,  [JALU1], 3>;
+defm : JWriteResIntPair<WriteIMul,  [JALU1, JMul], 3, [1, 1], 2>; // i8/i16/i32 multiplication
 defm : JWriteResIntPair<WriteIDiv,  [JALU1, JDiv], 41, [1, 41], 2>; // Worst case (i64 division)
 
 def  : WriteRes<WriteIMulH, [JALU1]> {
@@ -152,6 +152,19 @@ def JWriteTZCNTLd : SchedWriteRes<[JLAGU, JALU01]> {
 def : InstRW<[JWriteTZCNT], (instrs TZCNT16rr, TZCNT32rr, TZCNT64rr)>;
 def : InstRW<[JWriteTZCNTLd], (instrs TZCNT16rm, TZCNT32rm, TZCNT64rm)>;
 
+def JWriteIMul64 : SchedWriteRes<[JALU1, JMul]> {
+  let Latency = 6;
+  let ResourceCycles = [1, 4];
+  let NumMicroOps = 2;
+}
+def JWriteIMul64Ld : SchedWriteRes<[JLAGU, JALU1, JMul]> {
+  let Latency = 9;
+  let ResourceCycles = [1, 1, 4];
+  let NumMicroOps = 2;
+}
+def : InstRW<[JWriteIMul64], (instrs MUL64r, IMUL64r)>;
+def : InstRW<[JWriteIMul64Ld], (instrs MUL64m, IMUL64m)>;
+
 def JWriteIDiv8 : SchedWriteRes<[JALU1, JDiv]> {
   let Latency = 12;
   let ResourceCycles = [1, 12];