X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTarget%2FX86%2FX86InstrFMA.td;h=69cd5a568ba87abe8cefa11a3aa26bb58e130132;hb=22abf7e17f47f26691fdf4c590ebd88ebf560c73;hp=56638002d8eb2ef9788ee2733b99404054db44c2;hpb=dfb1e4babd2e825d951d42bcb45438b48c45b155;p=oota-llvm.git

diff --git a/lib/Target/X86/X86InstrFMA.td b/lib/Target/X86/X86InstrFMA.td
index 56638002d8e..69cd5a568ba 100644
--- a/lib/Target/X86/X86InstrFMA.td
+++ b/lib/Target/X86/X86InstrFMA.td
@@ -42,7 +42,7 @@ multiclass fma3p_rm<bits<8> opc, string OpcodeStr,
                    !strconcat(OpcodeStr,
                               "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
                    [(set VR256:$dst, (OpVT256 (Op VR256:$src2, VR256:$src1,
-                                               VR256:$src3)))]>;
+                                               VR256:$src3)))]>, VEX_L;
 
   let mayLoad = 1 in
   def mY    : FMA3<opc, MRMSrcMem, (outs VR256:$dst),
@@ -51,7 +51,7 @@ multiclass fma3p_rm<bits<8> opc, string OpcodeStr,
                               "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
                    [(set VR256:$dst,
                      (OpVT256 (Op VR256:$src2, VR256:$src1,
-                               (MemFrag256 addr:$src3))))]>;
+                               (MemFrag256 addr:$src3))))]>, VEX_L;
 }
 } // Constraints = "$src1 = $dst"
 
@@ -60,57 +60,57 @@ multiclass fma3p_forms<bits<8> opc132, bits<8> opc213, bits<8> opc231,
                        PatFrag MemFrag128, PatFrag MemFrag256,
                        SDNode Op, ValueType OpTy128, ValueType OpTy256> {
   defm r213 : fma3p_rm<opc213,
-                       !strconcat(OpcodeStr, !strconcat("213", PackTy)),
+                       !strconcat(OpcodeStr, "213", PackTy),
                        MemFrag128, MemFrag256, OpTy128, OpTy256, Op>;
 let neverHasSideEffects = 1 in {
   defm r132 : fma3p_rm<opc132,
-                       !strconcat(OpcodeStr, !strconcat("132", PackTy)),
+                       !strconcat(OpcodeStr, "132", PackTy),
                        MemFrag128, MemFrag256, OpTy128, OpTy256>;
   defm r231 : fma3p_rm<opc231,
-                       !strconcat(OpcodeStr, !strconcat("231", PackTy)),
+                       !strconcat(OpcodeStr, "231", PackTy),
                        MemFrag128, MemFrag256, OpTy128, OpTy256>;
 } // neverHasSideEffects = 1
 }
 
 // Fused Multiply-Add
 let ExeDomain = SSEPackedSingle in {
-  defm VFMADDPS    : fma3p_forms<0x98, 0xA8, 0xB8, "vfmadd", "ps", memopv4f32,
-                                 memopv8f32, X86Fmadd, v4f32, v8f32>;
-  defm VFMSUBPS    : fma3p_forms<0x9A, 0xAA, 0xBA, "vfmsub", "ps", memopv4f32,
-                                 memopv8f32, X86Fmsub, v4f32, v8f32>;
+  defm VFMADDPS    : fma3p_forms<0x98, 0xA8, 0xB8, "vfmadd", "ps", loadv4f32,
+                                 loadv8f32, X86Fmadd, v4f32, v8f32>;
+  defm VFMSUBPS    : fma3p_forms<0x9A, 0xAA, 0xBA, "vfmsub", "ps", loadv4f32,
+                                 loadv8f32, X86Fmsub, v4f32, v8f32>;
   defm VFMADDSUBPS : fma3p_forms<0x96, 0xA6, 0xB6, "vfmaddsub", "ps",
-                                 memopv4f32, memopv8f32, X86Fmaddsub,
+                                 loadv4f32, loadv8f32, X86Fmaddsub,
                                  v4f32, v8f32>;
   defm VFMSUBADDPS : fma3p_forms<0x97, 0xA7, 0xB7, "vfmsubadd", "ps",
-                                 memopv4f32, memopv8f32, X86Fmsubadd,
+                                 loadv4f32, loadv8f32, X86Fmsubadd,
                                  v4f32, v8f32>;
 }
 
 let ExeDomain = SSEPackedDouble in {
-  defm VFMADDPD    : fma3p_forms<0x98, 0xA8, 0xB8, "vfmadd", "pd", memopv2f64,
-                                 memopv4f64, X86Fmadd, v2f64, v4f64>, VEX_W;
-  defm VFMSUBPD    : fma3p_forms<0x9A, 0xAA, 0xBA, "vfmsub", "pd", memopv2f64,
-                                 memopv4f64, X86Fmsub, v2f64, v4f64>, VEX_W;
+  defm VFMADDPD    : fma3p_forms<0x98, 0xA8, 0xB8, "vfmadd", "pd", loadv2f64,
+                                 loadv4f64, X86Fmadd, v2f64, v4f64>, VEX_W;
+  defm VFMSUBPD    : fma3p_forms<0x9A, 0xAA, 0xBA, "vfmsub", "pd", loadv2f64,
+                                 loadv4f64, X86Fmsub, v2f64, v4f64>, VEX_W;
   defm VFMADDSUBPD : fma3p_forms<0x96, 0xA6, 0xB6, "vfmaddsub", "pd",
-                                 memopv2f64, memopv4f64, X86Fmaddsub,
+                                 loadv2f64, loadv4f64, X86Fmaddsub,
                                  v2f64, v4f64>, VEX_W;
   defm VFMSUBADDPD : fma3p_forms<0x97, 0xA7, 0xB7, "vfmsubadd", "pd",
-                                 memopv2f64, memopv4f64, X86Fmsubadd,
+                                 loadv2f64, loadv4f64, X86Fmsubadd,
                                  v2f64, v4f64>, VEX_W;
 }
 
 // Fused Negative Multiply-Add
 let ExeDomain = SSEPackedSingle in {
-  defm VFNMADDPS : fma3p_forms<0x9C, 0xAC, 0xBC, "vfnmadd", "ps",  memopv4f32,
-                               memopv8f32, X86Fnmadd, v4f32, v8f32>;
-  defm VFNMSUBPS : fma3p_forms<0x9E, 0xAE, 0xBE, "vfnmsub", "ps",  memopv4f32,
-                               memopv8f32, X86Fnmsub, v4f32, v8f32>;
+  defm VFNMADDPS : fma3p_forms<0x9C, 0xAC, 0xBC, "vfnmadd", "ps",  loadv4f32,
+                               loadv8f32, X86Fnmadd, v4f32, v8f32>;
+  defm VFNMSUBPS : fma3p_forms<0x9E, 0xAE, 0xBE, "vfnmsub", "ps",  loadv4f32,
+                               loadv8f32, X86Fnmsub, v4f32, v8f32>;
 }
 let ExeDomain = SSEPackedDouble in {
-  defm VFNMADDPD : fma3p_forms<0x9C, 0xAC, 0xBC, "vfnmadd", "pd", memopv2f64,
-                               memopv4f64, X86Fnmadd, v2f64, v4f64>, VEX_W;
+  defm VFNMADDPD : fma3p_forms<0x9C, 0xAC, 0xBC, "vfnmadd", "pd", loadv2f64,
+                               loadv4f64, X86Fnmadd, v2f64, v4f64>, VEX_W;
   defm VFNMSUBPD : fma3p_forms<0x9E, 0xAE, 0xBE, "vfnmsub", "pd",
-                               memopv2f64, memopv4f64, X86Fnmsub, v2f64,
+                               loadv2f64, loadv4f64, X86Fnmsub, v2f64,
                                v4f64>, VEX_W;
 }
 
@@ -160,15 +160,15 @@ multiclass fma3s_forms<bits<8> opc132, bits<8> opc213, bits<8> opc231,
                        X86MemOperand x86memop, Operand memop, PatFrag mem_frag,
                        ComplexPattern mem_cpat> {
 let neverHasSideEffects = 1 in {
-  defm r132 : fma3s_rm<opc132, !strconcat(OpStr, !strconcat("132", PackTy)),
+  defm r132 : fma3s_rm<opc132, !strconcat(OpStr, "132", PackTy),
                        x86memop, RC, OpVT, mem_frag>;
-  defm r231 : fma3s_rm<opc231, !strconcat(OpStr, !strconcat("231", PackTy)),
+  defm r231 : fma3s_rm<opc231, !strconcat(OpStr, "231", PackTy),
                        x86memop, RC, OpVT, mem_frag>;
 }
 
-defm r213 : fma3s_rm<opc213, !strconcat(OpStr, !strconcat("213", PackTy)),
+defm r213 : fma3s_rm<opc213, !strconcat(OpStr, "213", PackTy),
                      x86memop, RC, OpVT, mem_frag, OpNode>,
-            fma3s_rm_int<opc213, !strconcat(OpStr, !strconcat("213", PackTy)),
+            fma3s_rm_int<opc213, !strconcat(OpStr, "213", PackTy),
                          memop, mem_cpat, Int, RC>;
 }
 
@@ -206,25 +206,26 @@ multiclass fma4s<bits<8> opc, string OpcodeStr, RegisterClass RC,
            !strconcat(OpcodeStr,
            "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
            [(set RC:$dst,
-             (OpVT (OpNode RC:$src1, RC:$src2, RC:$src3)))]>, VEX_W, MemOp4;
+             (OpVT (OpNode RC:$src1, RC:$src2, RC:$src3)))]>, VEX_W, VEX_LIG, MemOp4;
   def rm : FMA4<opc, MRMSrcMem, (outs RC:$dst),
            (ins RC:$src1, RC:$src2, x86memop:$src3),
            !strconcat(OpcodeStr,
            "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
            [(set RC:$dst, (OpNode RC:$src1, RC:$src2,
-                           (mem_frag addr:$src3)))]>, VEX_W, MemOp4;
+                           (mem_frag addr:$src3)))]>, VEX_W, VEX_LIG, MemOp4;
   def mr : FMA4<opc, MRMSrcMem, (outs RC:$dst),
            (ins RC:$src1, x86memop:$src2, RC:$src3),
            !strconcat(OpcodeStr,
            "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
            [(set RC:$dst,
-             (OpNode RC:$src1, (mem_frag addr:$src2), RC:$src3))]>;
+             (OpNode RC:$src1, (mem_frag addr:$src2), RC:$src3))]>, VEX_LIG;
 // For disassembler
-let isCodeGenOnly = 1 in
+let isCodeGenOnly = 1, hasSideEffects = 0 in
   def rr_REV : FMA4<opc, MRMSrcReg, (outs RC:$dst),
                (ins RC:$src1, RC:$src2, RC:$src3),
                !strconcat(OpcodeStr,
-               "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), []>;
+               "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), []>,
+               VEX_LIG;
 }
 
 multiclass fma4s_int<bits<8> opc, string OpcodeStr, Operand memop,
@@ -235,19 +236,19 @@ multiclass fma4s_int<bits<8> opc, string OpcodeStr, Operand memop,
                !strconcat(OpcodeStr,
                "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
                [(set VR128:$dst,
-                 (Int VR128:$src1, VR128:$src2, VR128:$src3))]>, VEX_W, MemOp4;
+                 (Int VR128:$src1, VR128:$src2, VR128:$src3))]>, VEX_W, VEX_LIG, MemOp4;
   def rm_Int : FMA4<opc, MRMSrcMem, (outs VR128:$dst),
                (ins VR128:$src1, VR128:$src2, memop:$src3),
                !strconcat(OpcodeStr,
                "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
                [(set VR128:$dst, (Int VR128:$src1, VR128:$src2,
-                                  mem_cpat:$src3))]>, VEX_W, MemOp4;
+                                  mem_cpat:$src3))]>, VEX_W, VEX_LIG, MemOp4;
   def mr_Int : FMA4<opc, MRMSrcMem, (outs VR128:$dst),
                (ins VR128:$src1, memop:$src2, VR128:$src3),
                !strconcat(OpcodeStr,
                "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
                [(set VR128:$dst,
-                 (Int VR128:$src1, mem_cpat:$src2, VR128:$src3))]>;
+                 (Int VR128:$src1, mem_cpat:$src2, VR128:$src3))]>, VEX_LIG;
 }
 
 multiclass fma4p<bits<8> opc, string OpcodeStr, SDNode OpNode,
@@ -280,21 +281,21 @@ multiclass fma4p<bits<8> opc, string OpcodeStr, SDNode OpNode,
            "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
            [(set VR256:$dst,
              (OpVT256 (OpNode VR256:$src1, VR256:$src2, VR256:$src3)))]>,
-           VEX_W, MemOp4;
+           VEX_W, MemOp4, VEX_L;
   def rmY : FMA4<opc, MRMSrcMem, (outs VR256:$dst),
            (ins VR256:$src1, VR256:$src2, f256mem:$src3),
            !strconcat(OpcodeStr,
            "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
            [(set VR256:$dst, (OpNode VR256:$src1, VR256:$src2,
-                              (ld_frag256 addr:$src3)))]>, VEX_W, MemOp4;
+                              (ld_frag256 addr:$src3)))]>, VEX_W, MemOp4, VEX_L;
   def mrY : FMA4<opc, MRMSrcMem, (outs VR256:$dst),
            (ins VR256:$src1, f256mem:$src2, VR256:$src3),
            !strconcat(OpcodeStr,
            "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
-           [(set VR256:$dst,
-             (OpNode VR256:$src1, (ld_frag256 addr:$src2), VR256:$src3))]>;
+           [(set VR256:$dst, (OpNode VR256:$src1,
+                              (ld_frag256 addr:$src2), VR256:$src3))]>, VEX_L;
 // For disassembler
-let isCodeGenOnly = 1 in {
+let isCodeGenOnly = 1, hasSideEffects = 0 in {
   def rr_REV : FMA4<opc, MRMSrcReg, (outs VR128:$dst),
                (ins VR128:$src1, VR128:$src2, VR128:$src3),
                !strconcat(OpcodeStr,
@@ -302,12 +303,11 @@ let isCodeGenOnly = 1 in {
   def rrY_REV : FMA4<opc, MRMSrcReg, (outs VR256:$dst),
                 (ins VR256:$src1, VR256:$src2, VR256:$src3),
                 !strconcat(OpcodeStr,
-                "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), []>;
+                "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), []>,
+                VEX_L;
 } // isCodeGenOnly = 1
 }
 
-let Predicates = [HasFMA4] in {
-
 defm VFMADDSS4  : fma4s<0x6A, "vfmaddss", FR32, f32mem, f32, X86Fmadd, loadf32>,
                   fma4s_int<0x6A, "vfmaddss", ssmem, sse_load_f32,
                             int_x86_fma_vfmadd_ss>;
@@ -337,29 +337,33 @@ defm VFNMSUBSD4 : fma4s<0x7F, "vfnmsubsd", FR64, f64mem, f64,
                   fma4s_int<0x7F, "vfnmsubsd", sdmem, sse_load_f64,
                             int_x86_fma_vfnmsub_sd>;
 
-defm VFMADDPS4    : fma4p<0x68, "vfmaddps", X86Fmadd, v4f32, v8f32,
-                          memopv4f32, memopv8f32>;
-defm VFMADDPD4    : fma4p<0x69, "vfmaddpd", X86Fmadd, v2f64, v4f64,
-                          memopv2f64, memopv4f64>;
-defm VFMSUBPS4    : fma4p<0x6C, "vfmsubps", X86Fmsub, v4f32, v8f32,
-                          memopv4f32, memopv8f32>;
-defm VFMSUBPD4    : fma4p<0x6D, "vfmsubpd", X86Fmsub, v2f64, v4f64,
-                          memopv2f64, memopv4f64>;
-defm VFNMADDPS4   : fma4p<0x78, "vfnmaddps", X86Fnmadd, v4f32, v8f32,
-                          memopv4f32, memopv8f32>;
-defm VFNMADDPD4   : fma4p<0x79, "vfnmaddpd", X86Fnmadd, v2f64, v4f64,
-                          memopv2f64, memopv4f64>;
-defm VFNMSUBPS4   : fma4p<0x7C, "vfnmsubps", X86Fnmsub, v4f32, v8f32,
-                          memopv4f32, memopv8f32>;
-defm VFNMSUBPD4   : fma4p<0x7D, "vfnmsubpd", X86Fnmsub, v2f64, v4f64,
-                          memopv2f64, memopv4f64>;
-defm VFMADDSUBPS4 : fma4p<0x5C, "vfmaddsubps", X86Fmaddsub, v4f32, v8f32,
-                          memopv4f32, memopv8f32>;
-defm VFMADDSUBPD4 : fma4p<0x5D, "vfmaddsubpd", X86Fmaddsub, v2f64, v4f64,
-                          memopv2f64, memopv4f64>;
-defm VFMSUBADDPS4 : fma4p<0x5E, "vfmsubaddps", X86Fmsubadd, v4f32, v8f32,
-                          memopv4f32, memopv8f32>;
-defm VFMSUBADDPD4 : fma4p<0x5F, "vfmsubaddpd", X86Fmsubadd, v2f64, v4f64,
-                          memopv2f64, memopv4f64>;
-} // HasFMA4
+let ExeDomain = SSEPackedSingle in {
+  defm VFMADDPS4    : fma4p<0x68, "vfmaddps", X86Fmadd, v4f32, v8f32,
+                            loadv4f32, loadv8f32>;
+  defm VFMSUBPS4    : fma4p<0x6C, "vfmsubps", X86Fmsub, v4f32, v8f32,
+                            loadv4f32, loadv8f32>;
+  defm VFNMADDPS4   : fma4p<0x78, "vfnmaddps", X86Fnmadd, v4f32, v8f32,
+                            loadv4f32, loadv8f32>;
+  defm VFNMSUBPS4   : fma4p<0x7C, "vfnmsubps", X86Fnmsub, v4f32, v8f32,
+                            loadv4f32, loadv8f32>;
+  defm VFMADDSUBPS4 : fma4p<0x5C, "vfmaddsubps", X86Fmaddsub, v4f32, v8f32,
+                            loadv4f32, loadv8f32>;
+  defm VFMSUBADDPS4 : fma4p<0x5E, "vfmsubaddps", X86Fmsubadd, v4f32, v8f32,
+                            loadv4f32, loadv8f32>;
+}
+
+let ExeDomain = SSEPackedDouble in {
+  defm VFMADDPD4    : fma4p<0x69, "vfmaddpd", X86Fmadd, v2f64, v4f64,
+                            loadv2f64, loadv4f64>;
+  defm VFMSUBPD4    : fma4p<0x6D, "vfmsubpd", X86Fmsub, v2f64, v4f64,
+                            loadv2f64, loadv4f64>;
+  defm VFNMADDPD4   : fma4p<0x79, "vfnmaddpd", X86Fnmadd, v2f64, v4f64,
+                            loadv2f64, loadv4f64>;
+  defm VFNMSUBPD4   : fma4p<0x7D, "vfnmsubpd", X86Fnmsub, v2f64, v4f64,
+                            loadv2f64, loadv4f64>;
+  defm VFMADDSUBPD4 : fma4p<0x5D, "vfmaddsubpd", X86Fmaddsub, v2f64, v4f64,
+                            loadv2f64, loadv4f64>;
+  defm VFMSUBADDPD4 : fma4p<0x5F, "vfmsubaddpd", X86Fmsubadd, v2f64, v4f64,
+                            loadv2f64, loadv4f64>;
+}