commit 81d15be76923ec23b86e72bbc2d3a2a6542a325e
author Shalini Salomi Bodapati <shalini.salomi.bodapati@intel.com> Thu May 30 11:00:42 2019 +0530
committer Vladimir Marko <vmarko@google.com> Wed Jun 05 16:06:49 2019 +0000
tree 57bf31c2951cbeebddd72c94a415711423c88f88
parent 5fd673488c25991ecc85d26c2dc54415773eb407

Add AVX support for packed add/sub instructions on x86

Test: ./test.py --host, test-art-host-gtest

Change-Id: I48d05e6f6befd54657d962119a543b27a8a51d71
Signed-off-by: Shalini Salomi Bodapati <shalini.salomi.bodapati@intel.com>

compiler/utils/x86/assembler_x86.h

diff --git a/compiler/utils/x86/assembler_x86.h b/compiler/utils/x86/assembler_x86.h
index e84294a..17039f0 100644
--- a/compiler/utils/x86/assembler_x86.h
+++ b/compiler/utils/x86/assembler_x86.h
@@ -417,6 +417,11 @@
   void mulps(XmmRegister dst, XmmRegister src);
   void divps(XmmRegister dst, XmmRegister src);
 
+  void vaddps(XmmRegister dst, XmmRegister add_left, XmmRegister add_right);
+  void vsubps(XmmRegister dst, XmmRegister add_left, XmmRegister add_right);
+  void vsubpd(XmmRegister dst, XmmRegister add_left, XmmRegister add_right);
+  void vaddpd(XmmRegister dst, XmmRegister add_left, XmmRegister add_right);
+
   void movapd(XmmRegister dst, XmmRegister src);     // move
   void movapd(XmmRegister dst, const Address& src);  // load aligned
   void movupd(XmmRegister dst, const Address& src);  // load unaligned
@@ -465,17 +470,29 @@
   void paddb(XmmRegister dst, XmmRegister src);  // no addr variant (for now)
   void psubb(XmmRegister dst, XmmRegister src);
 
+  void vpaddb(XmmRegister dst, XmmRegister add_left, XmmRegister add_right);
+  void vpaddw(XmmRegister dst, XmmRegister add_left, XmmRegister add_right);
+
   void paddw(XmmRegister dst, XmmRegister src);
   void psubw(XmmRegister dst, XmmRegister src);
   void pmullw(XmmRegister dst, XmmRegister src);
 
+  void vpsubb(XmmRegister dst, XmmRegister src1, XmmRegister src2);
+  void vpsubw(XmmRegister dst, XmmRegister src1, XmmRegister src2);
+  void vpsubd(XmmRegister dst, XmmRegister src1, XmmRegister src2);
+
   void paddd(XmmRegister dst, XmmRegister src);
   void psubd(XmmRegister dst, XmmRegister src);
   void pmulld(XmmRegister dst, XmmRegister src);
 
+  void vpaddd(XmmRegister dst, XmmRegister src1, XmmRegister src2);
+
   void paddq(XmmRegister dst, XmmRegister src);
   void psubq(XmmRegister dst, XmmRegister src);
 
+  void vpaddq(XmmRegister dst, XmmRegister add_left, XmmRegister add_right);
+  void vpsubq(XmmRegister dst, XmmRegister add_left, XmmRegister add_right);
+
   void paddusb(XmmRegister dst, XmmRegister src);
   void paddsb(XmmRegister dst, XmmRegister src);
   void paddusw(XmmRegister dst, XmmRegister src);