AMDGPU/SI: Emit global variable sizes when targeting HSA

[oota-llvm.git] / lib / Target / X86 / X86InstrFMA.td

diff --git a/lib/Target/X86/X86InstrFMA.td b/lib/Target/X86/X86InstrFMA.td
index 0467a64d7e51283e472c7c3c837b78f1d55ac3fe..fd800cf077f7bb08acee5475fb73fec3701fd3f5 100644 (file)
--- a/lib/Target/X86/X86InstrFMA.td
+++ b/lib/Target/X86/X86InstrFMA.td
@@ -223,9 +223,12 @@ multiclass fma3s_int_forms<bits<8> opc132, bits<8> opc213, bits<8> opc231,
 multiclass fma3s<bits<8> opc132, bits<8> opc213, bits<8> opc231,
                  string OpStr, Intrinsic IntF32, Intrinsic IntF64,
                  SDNode OpNode> {
+  let ExeDomain = SSEPackedSingle in
   defm SS : fma3s_forms<opc132, opc213, opc231, OpStr, "ss", OpNode,
                         FR32, f32mem>,
             fma3s_int_forms<opc132, opc213, opc231, OpStr, "ss", VR128, ssmem>;
+
+  let ExeDomain = SSEPackedDouble in
   defm SD : fma3s_forms<opc132, opc213, opc231, OpStr, "sd", OpNode,
                         FR64, f64mem>,
             fma3s_int_forms<opc132, opc213, opc231, OpStr, "sd", VR128, sdmem>,
@@ -374,36 +377,23 @@ let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in {
 } // isCodeGenOnly = 1
 }
 
-defm VFMADDSS4  : fma4s<0x6A, "vfmaddss", FR32, f32mem, f32, X86Fmadd, loadf32>,
-                  fma4s_int<0x6A, "vfmaddss", ssmem, sse_load_f32,
-                            int_x86_fma_vfmadd_ss>;
-defm VFMADDSD4  : fma4s<0x6B, "vfmaddsd", FR64, f64mem, f64, X86Fmadd, loadf64>,
-                  fma4s_int<0x6B, "vfmaddsd", sdmem, sse_load_f64,
-                            int_x86_fma_vfmadd_sd>;
-defm VFMSUBSS4  : fma4s<0x6E, "vfmsubss", FR32, f32mem, f32, X86Fmsub, loadf32>,
-                  fma4s_int<0x6E, "vfmsubss", ssmem, sse_load_f32,
-                            int_x86_fma_vfmsub_ss>;
-defm VFMSUBSD4  : fma4s<0x6F, "vfmsubsd", FR64, f64mem, f64, X86Fmsub, loadf64>,
-                  fma4s_int<0x6F, "vfmsubsd", sdmem, sse_load_f64,
-                            int_x86_fma_vfmsub_sd>;
-defm VFNMADDSS4 : fma4s<0x7A, "vfnmaddss", FR32, f32mem, f32,
-                        X86Fnmadd, loadf32>,
-                  fma4s_int<0x7A, "vfnmaddss", ssmem, sse_load_f32,
-                            int_x86_fma_vfnmadd_ss>;
-defm VFNMADDSD4 : fma4s<0x7B, "vfnmaddsd", FR64, f64mem, f64,
-                        X86Fnmadd, loadf64>,
-                  fma4s_int<0x7B, "vfnmaddsd", sdmem, sse_load_f64,
-                            int_x86_fma_vfnmadd_sd>;
-defm VFNMSUBSS4 : fma4s<0x7E, "vfnmsubss", FR32, f32mem, f32,
-                        X86Fnmsub, loadf32>,
-                  fma4s_int<0x7E, "vfnmsubss", ssmem, sse_load_f32,
-                            int_x86_fma_vfnmsub_ss>;
-defm VFNMSUBSD4 : fma4s<0x7F, "vfnmsubsd", FR64, f64mem, f64,
-                        X86Fnmsub, loadf64>,
-                  fma4s_int<0x7F, "vfnmsubsd", sdmem, sse_load_f64,
-                            int_x86_fma_vfnmsub_sd>;
-
 let ExeDomain = SSEPackedSingle in {
+  // Scalar Instructions
+  defm VFMADDSS4  : fma4s<0x6A, "vfmaddss", FR32, f32mem, f32, X86Fmadd, loadf32>,
+                    fma4s_int<0x6A, "vfmaddss", ssmem, sse_load_f32,
+                              int_x86_fma_vfmadd_ss>;
+  defm VFMSUBSS4  : fma4s<0x6E, "vfmsubss", FR32, f32mem, f32, X86Fmsub, loadf32>,
+                    fma4s_int<0x6E, "vfmsubss", ssmem, sse_load_f32,
+                              int_x86_fma_vfmsub_ss>;
+  defm VFNMADDSS4 : fma4s<0x7A, "vfnmaddss", FR32, f32mem, f32,
+                          X86Fnmadd, loadf32>,
+                    fma4s_int<0x7A, "vfnmaddss", ssmem, sse_load_f32,
+                              int_x86_fma_vfnmadd_ss>;
+  defm VFNMSUBSS4 : fma4s<0x7E, "vfnmsubss", FR32, f32mem, f32,
+                          X86Fnmsub, loadf32>,
+                    fma4s_int<0x7E, "vfnmsubss", ssmem, sse_load_f32,
+                              int_x86_fma_vfnmsub_ss>;
+  // Packed Instructions
   defm VFMADDPS4    : fma4p<0x68, "vfmaddps", X86Fmadd, v4f32, v8f32,
                             loadv4f32, loadv8f32>;
   defm VFMSUBPS4    : fma4p<0x6C, "vfmsubps", X86Fmsub, v4f32, v8f32,
@@ -419,6 +409,22 @@ let ExeDomain = SSEPackedSingle in {
 }
 
 let ExeDomain = SSEPackedDouble in {
+  // Scalar Instructions
+  defm VFMADDSD4  : fma4s<0x6B, "vfmaddsd", FR64, f64mem, f64, X86Fmadd, loadf64>,
+                    fma4s_int<0x6B, "vfmaddsd", sdmem, sse_load_f64,
+                              int_x86_fma_vfmadd_sd>;
+  defm VFMSUBSD4  : fma4s<0x6F, "vfmsubsd", FR64, f64mem, f64, X86Fmsub, loadf64>,
+                    fma4s_int<0x6F, "vfmsubsd", sdmem, sse_load_f64,
+                              int_x86_fma_vfmsub_sd>;
+  defm VFNMADDSD4 : fma4s<0x7B, "vfnmaddsd", FR64, f64mem, f64,
+                          X86Fnmadd, loadf64>,
+                    fma4s_int<0x7B, "vfnmaddsd", sdmem, sse_load_f64,
+                              int_x86_fma_vfnmadd_sd>;
+  defm VFNMSUBSD4 : fma4s<0x7F, "vfnmsubsd", FR64, f64mem, f64,
+                          X86Fnmsub, loadf64>,
+                    fma4s_int<0x7F, "vfnmsubsd", sdmem, sse_load_f64,
+                              int_x86_fma_vfnmsub_sd>;
+  // Packed Instructions
   defm VFMADDPD4    : fma4p<0x69, "vfmaddpd", X86Fmadd, v2f64, v4f64,
                             loadv2f64, loadv4f64>;
   defm VFMSUBPD4    : fma4p<0x6D, "vfmsubpd", X86Fmsub, v2f64, v4f64,