Intel Atom instruction itineraries for mov sign extension and mov zero extension.

Patch by Tyler Nowicki!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151743 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 68 insertions, 32 deletions

diff --git a/lib/Target/X86/X86InstrExtension.td b/lib/Target/X86/X86InstrExtension.td
index 8835d0aaa1..0d5490ad9c 100644
--- a/lib/Target/X86/X86InstrExtension.td
+++ b/lib/Target/X86/X86InstrExtension.td
@@ -37,40 +37,47 @@ let neverHasSideEffects = 1 in {
 }
 
 
+
 // Sign/Zero extenders
 def MOVSX16rr8 : I<0xBE, MRMSrcReg, (outs GR16:$dst), (ins GR8:$src),
-                   "movs{bw|x}\t{$src, $dst|$dst, $src}", []>, TB, OpSize;
+                   "movs{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVSX_R16_R8>,
+                   TB, OpSize;
 def MOVSX16rm8 : I<0xBE, MRMSrcMem, (outs GR16:$dst), (ins i8mem:$src),
-                   "movs{bw|x}\t{$src, $dst|$dst, $src}", []>, TB, OpSize;
+                   "movs{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVSX_R16_M8>,
+                   TB, OpSize;
 def MOVSX32rr8 : I<0xBE, MRMSrcReg, (outs GR32:$dst), (ins GR8:$src),
                    "movs{bl|x}\t{$src, $dst|$dst, $src}",
-                   [(set GR32:$dst, (sext GR8:$src))]>, TB;
+                   [(set GR32:$dst, (sext GR8:$src))], IIC_MOVSX>, TB;
 def MOVSX32rm8 : I<0xBE, MRMSrcMem, (outs GR32:$dst), (ins i8mem :$src),
                    "movs{bl|x}\t{$src, $dst|$dst, $src}",
-                   [(set GR32:$dst, (sextloadi32i8 addr:$src))]>, TB;
+                   [(set GR32:$dst, (sextloadi32i8 addr:$src))], IIC_MOVSX>, TB;
 def MOVSX32rr16: I<0xBF, MRMSrcReg, (outs GR32:$dst), (ins GR16:$src),
                    "movs{wl|x}\t{$src, $dst|$dst, $src}",
-                   [(set GR32:$dst, (sext GR16:$src))]>, TB;
+                   [(set GR32:$dst, (sext GR16:$src))], IIC_MOVSX>, TB;
 def MOVSX32rm16: I<0xBF, MRMSrcMem, (outs GR32:$dst), (ins i16mem:$src),
                    "movs{wl|x}\t{$src, $dst|$dst, $src}",
-                   [(set GR32:$dst, (sextloadi32i16 addr:$src))]>, TB;
+                   [(set GR32:$dst, (sextloadi32i16 addr:$src))], IIC_MOVSX>,
+                   TB;
 
 def MOVZX16rr8 : I<0xB6, MRMSrcReg, (outs GR16:$dst), (ins GR8:$src),
-                   "movz{bw|x}\t{$src, $dst|$dst, $src}", []>, TB, OpSize;
+                   "movz{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX_R16_R8>,
+                   TB, OpSize;
 def MOVZX16rm8 : I<0xB6, MRMSrcMem, (outs GR16:$dst), (ins i8mem:$src),
-                   "movz{bw|x}\t{$src, $dst|$dst, $src}", []>, TB, OpSize;
+                   "movz{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX_R16_M8>,
+                   TB, OpSize;
 def MOVZX32rr8 : I<0xB6, MRMSrcReg, (outs GR32:$dst), (ins GR8 :$src),
                    "movz{bl|x}\t{$src, $dst|$dst, $src}",
-                   [(set GR32:$dst, (zext GR8:$src))]>, TB;
+                   [(set GR32:$dst, (zext GR8:$src))], IIC_MOVZX>, TB;
 def MOVZX32rm8 : I<0xB6, MRMSrcMem, (outs GR32:$dst), (ins i8mem :$src),
                    "movz{bl|x}\t{$src, $dst|$dst, $src}",
-                   [(set GR32:$dst, (zextloadi32i8 addr:$src))]>, TB;
+                   [(set GR32:$dst, (zextloadi32i8 addr:$src))], IIC_MOVZX>, TB;
 def MOVZX32rr16: I<0xB7, MRMSrcReg, (outs GR32:$dst), (ins GR16:$src),
                    "movz{wl|x}\t{$src, $dst|$dst, $src}",
-                   [(set GR32:$dst, (zext GR16:$src))]>, TB;
+                   [(set GR32:$dst, (zext GR16:$src))], IIC_MOVZX>, TB;
 def MOVZX32rm16: I<0xB7, MRMSrcMem, (outs GR32:$dst), (ins i16mem:$src),
                    "movz{wl|x}\t{$src, $dst|$dst, $src}",
-                   [(set GR32:$dst, (zextloadi32i16 addr:$src))]>, TB;
+                   [(set GR32:$dst, (zextloadi32i16 addr:$src))], IIC_MOVZX>,
+                   TB;
 
 // These are the same as the regular MOVZX32rr8 and MOVZX32rm8
 // except that they use GR32_NOREX for the output operand register class
@@ -78,12 +85,12 @@ def MOVZX32rm16: I<0xB7, MRMSrcMem, (outs GR32:$dst), (ins i16mem:$src),
 def MOVZX32_NOREXrr8 : I<0xB6, MRMSrcReg,
                          (outs GR32_NOREX:$dst), (ins GR8_NOREX:$src),
                          "movz{bl|x}\t{$src, $dst|$dst, $src}",
-                         []>, TB;
+                         [], IIC_MOVZX>, TB;
 let mayLoad = 1 in
 def MOVZX32_NOREXrm8 : I<0xB6, MRMSrcMem,
                          (outs GR32_NOREX:$dst), (ins i8mem_NOREX:$src),
                          "movz{bl|x}\t{$src, $dst|$dst, $src}",
-                         []>, TB;
+                         [], IIC_MOVZX>, TB;
 
 // MOVSX64rr8 always has a REX prefix and it has an 8-bit register
 // operand, which makes it a rare instruction with an 8-bit register
@@ -91,32 +98,38 @@ def MOVZX32_NOREXrm8 : I<0xB6, MRMSrcMem,
 // were generalized, this would require a special register class.
 def MOVSX64rr8 : RI<0xBE, MRMSrcReg, (outs GR64:$dst), (ins GR8 :$src),
                     "movs{bq|x}\t{$src, $dst|$dst, $src}",
-                    [(set GR64:$dst, (sext GR8:$src))]>, TB;
+                    [(set GR64:$dst, (sext GR8:$src))], IIC_MOVSX>, TB;
 def MOVSX64rm8 : RI<0xBE, MRMSrcMem, (outs GR64:$dst), (ins i8mem :$src),
                     "movs{bq|x}\t{$src, $dst|$dst, $src}",
-                    [(set GR64:$dst, (sextloadi64i8 addr:$src))]>, TB;
+                    [(set GR64:$dst, (sextloadi64i8 addr:$src))], IIC_MOVSX>,
+                    TB;
 def MOVSX64rr16: RI<0xBF, MRMSrcReg, (outs GR64:$dst), (ins GR16:$src),
                     "movs{wq|x}\t{$src, $dst|$dst, $src}",
-                    [(set GR64:$dst, (sext GR16:$src))]>, TB;
+                    [(set GR64:$dst, (sext GR16:$src))], IIC_MOVSX>, TB;
 def MOVSX64rm16: RI<0xBF, MRMSrcMem, (outs GR64:$dst), (ins i16mem:$src),
                     "movs{wq|x}\t{$src, $dst|$dst, $src}",
-                    [(set GR64:$dst, (sextloadi64i16 addr:$src))]>, TB;
+                    [(set GR64:$dst, (sextloadi64i16 addr:$src))], IIC_MOVSX>,
+                    TB;
 def MOVSX64rr32: RI<0x63, MRMSrcReg, (outs GR64:$dst), (ins GR32:$src),
                     "movs{lq|xd}\t{$src, $dst|$dst, $src}",
-                    [(set GR64:$dst, (sext GR32:$src))]>;
+                    [(set GR64:$dst, (sext GR32:$src))], IIC_MOVSX>;
 def MOVSX64rm32: RI<0x63, MRMSrcMem, (outs GR64:$dst), (ins i32mem:$src),
                     "movs{lq|xd}\t{$src, $dst|$dst, $src}",
-                    [(set GR64:$dst, (sextloadi64i32 addr:$src))]>;
+                    [(set GR64:$dst, (sextloadi64i32 addr:$src))], IIC_MOVSX>;
 
 // movzbq and movzwq encodings for the disassembler
 def MOVZX64rr8_Q : RI<0xB6, MRMSrcReg, (outs GR64:$dst), (ins GR8:$src),
-                       "movz{bq|x}\t{$src, $dst|$dst, $src}", []>, TB;
+                       "movz{bq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>,
+                       TB;
 def MOVZX64rm8_Q : RI<0xB6, MRMSrcMem, (outs GR64:$dst), (ins i8mem:$src),
-                       "movz{bq|x}\t{$src, $dst|$dst, $src}", []>, TB;
+                       "movz{bq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>,
+                       TB;
 def MOVZX64rr16_Q : RI<0xB7, MRMSrcReg, (outs GR64:$dst), (ins GR16:$src),
-                       "movz{wq|x}\t{$src, $dst|$dst, $src}", []>, TB;
+                       "movz{wq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>,
+                       TB;
 def MOVZX64rm16_Q : RI<0xB7, MRMSrcMem, (outs GR64:$dst), (ins i16mem:$src),
-                       "movz{wq|x}\t{$src, $dst|$dst, $src}", []>, TB;
+                       "movz{wq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>,
+                       TB;
 
 // FIXME: These should be Pat patterns.
 let isCodeGenOnly = 1 in {
@@ -124,15 +137,17 @@ let isCodeGenOnly = 1 in {
 // Use movzbl instead of movzbq when the destination is a register; it's
 // equivalent due to implicit zero-extending, and it has a smaller encoding.
 def MOVZX64rr8 : I<0xB6, MRMSrcReg, (outs GR64:$dst), (ins GR8 :$src),
-                   "", [(set GR64:$dst, (zext GR8:$src))]>, TB;
+                   "", [(set GR64:$dst, (zext GR8:$src))], IIC_MOVZX>, TB;
 def MOVZX64rm8 : I<0xB6, MRMSrcMem, (outs GR64:$dst), (ins i8mem :$src),
-                   "", [(set GR64:$dst, (zextloadi64i8 addr:$src))]>, TB;
+                   "", [(set GR64:$dst, (zextloadi64i8 addr:$src))], IIC_MOVZX>,
+                   TB;
 // Use movzwl instead of movzwq when the destination is a register; it's
 // equivalent due to implicit zero-extending, and it has a smaller encoding.
 def MOVZX64rr16: I<0xB7, MRMSrcReg, (outs GR64:$dst), (ins GR16:$src),
-                   "", [(set GR64:$dst, (zext GR16:$src))]>, TB;
+                   "", [(set GR64:$dst, (zext GR16:$src))], IIC_MOVZX>, TB;
 def MOVZX64rm16: I<0xB7, MRMSrcMem, (outs GR64:$dst), (ins i16mem:$src),
-                   "", [(set GR64:$dst, (zextloadi64i16 addr:$src))]>, TB;
+                   "", [(set GR64:$dst, (zextloadi64i16 addr:$src))],
+                   IIC_MOVZX>, TB;
 
 // There's no movzlq instruction, but movl can be used for this purpose, using
 // implicit zero-extension. The preferred way to do 32-bit-to-64-bit zero
@@ -142,10 +157,9 @@ def MOVZX64rm16: I<0xB7, MRMSrcMem, (outs GR64:$dst), (ins i16mem:$src),
 // necessarily all zero. In such cases, we fall back to these explicit zext
 // instructions.
 def MOVZX64rr32 : I<0x89, MRMDestReg, (outs GR64:$dst), (ins GR32:$src),
-                    "", [(set GR64:$dst, (zext GR32:$src))]>;
+                    "", [(set GR64:$dst, (zext GR32:$src))], IIC_MOVZX>;
 def MOVZX64rm32 : I<0x8B, MRMSrcMem, (outs GR64:$dst), (ins i32mem:$src),
-                    "", [(set GR64:$dst, (zextloadi64i32 addr:$src))]>;
-
-
+                    "", [(set GR64:$dst, (zextloadi64i32 addr:$src))],
+                    IIC_MOVZX>;
 }
 
diff --git a/lib/Target/X86/X86Schedule.td b/lib/Target/X86/X86Schedule.td
index 9692833486..d6d0149566 100644
--- a/lib/Target/X86/X86Schedule.td
+++ b/lib/Target/X86/X86Schedule.td
@@ -103,6 +103,17 @@ def IIC_CALL_FAR_PTR : InstrItinClass;
 // ret
 def IIC_RET : InstrItinClass;
 def IIC_RET_IMM : InstrItinClass;
+//sign extension movs
+def IIC_MOVSX : InstrItinClass;
+def IIC_MOVSX_R16_R8 : InstrItinClass;
+def IIC_MOVSX_R16_M8 : InstrItinClass;
+def IIC_MOVSX_R16_R16 : InstrItinClass;
+def IIC_MOVSX_R32_R32 : InstrItinClass;
+//zero extension movs
+def IIC_MOVZX : InstrItinClass;
+def IIC_MOVZX_R16_R8 : InstrItinClass;
+def IIC_MOVZX_R16_M8 : InstrItinClass;
+
 // SSE scalar/parallel binary operations
 def IIC_SSE_ALU_F32S_RR : InstrItinClass;
 def IIC_SSE_ALU_F32S_RM : InstrItinClass;
diff --git a/lib/Target/X86/X86ScheduleAtom.td b/lib/Target/X86/X86ScheduleAtom.td
index 67cd0f2766..e8cf72a293 100644
--- a/lib/Target/X86/X86ScheduleAtom.td
+++ b/lib/Target/X86/X86ScheduleAtom.td
@@ -133,6 +133,17 @@ def AtomItineraries : ProcessorItineraries<
   //ret
   InstrItinData<IIC_RET, [InstrStage<79, [Port0, Port1]>] >,
   InstrItinData<IIC_RET_IMM, [InstrStage<1, [Port0], 0>,  InstrStage<1, [Port1]>] >,
+  //sign extension movs
+  InstrItinData<IIC_MOVSX,[InstrStage<1, [Port0] >] >,
+  InstrItinData<IIC_MOVSX_R16_R8, [InstrStage<2, [Port0, Port1]>] >,
+  InstrItinData<IIC_MOVSX_R16_M8, [InstrStage<3, [Port0, Port1]>] >,
+  InstrItinData<IIC_MOVSX_R16_R16, [InstrStage<1, [Port0, Port1]>] >,
+  InstrItinData<IIC_MOVSX_R32_R32, [InstrStage<1, [Port0, Port1]>] >,
+  //zero extension movs
+  InstrItinData<IIC_MOVZX,[InstrStage<1, [Port0]>] >,
+  InstrItinData<IIC_MOVZX_R16_R8, [InstrStage<2, [Port0, Port1]>] >,
+  InstrItinData<IIC_MOVZX_R16_M8, [InstrStage<3, [Port0, Port1]>] >,
+
   // SSE binary operations
   // arithmetic fp scalar
   InstrItinData<IIC_SSE_ALU_F32S_RR, [InstrStage<5, [Port1]>] >,