[PATCH] D16629: [X86][AVX512] add intrinsics of Scalar FP to integer conversion with rounding mode

[Igor Breger via llvm-commits]

igorb added inline comments.

================
Comment at: E:/Work/LLVM/llvmOrg/lib/Target/X86/X86InstrAVX512.td:4836
@@ -4834,3 +4835,3 @@
     let mayLoad = 1 in
-    def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins memop:$src),
-                !strconcat(asm,"\t{$src, $dst|$dst, $src}"), []>, EVEX, VEX_LIG;
+    def rm : SI<opc, MRMSrcMem, (outs DstVT.RC:$dst), (ins SrcVT.MemOp:$src),
+                !strconcat(asm,"\t{$src, $dst|$dst, $src}"),
----------------
scalar memory - SrcVT.ScalarMemOp

================
Comment at: E:/Work/LLVM/llvmOrg/lib/Target/X86/X86InstrAVX512.td:4872
@@ +4871,3 @@
+let Predicates = [HasAVX512] in {
+  def : Pat<(i32 (int_x86_sse_cvtss2si (v4f32 VR128X:$src))),
+            (VCVTSS2SIZrr (COPY_TO_REGCLASS VR128X:$src, FR32X))>;
----------------
could you please add comment  ( why sse intrinsic are here ).

================
Comment at: E:/Work/LLVM/llvmOrg/lib/Target/X86/X86IntrinsicsInfo.h:2035
@@ -2034,1 +2034,3 @@
   X86_INTRINSIC_DATA(avx512_vcomi_ss, COMI_RM, X86ISD::COMI, X86ISD::UCOMI),
+  X86_INTRINSIC_DATA(avx512_vcvtsd2si32, INTR_TYPE_2OP, X86ISD::SCALAR_FP_TO_SINT_RND, 0),
+  X86_INTRINSIC_DATA(avx512_vcvtsd2si64, INTR_TYPE_2OP, X86ISD::SCALAR_FP_TO_SINT_RND, 0),
----------------
please wrap


Repository:
  rL LLVM

http://reviews.llvm.org/D16629