Incoming new memcpy implementation

2024-11-10 08:38:20 +00:00 · 2018-05-18 01:59:00 +10:00 · 2018-05-18 01:59:00 +10:00 · ffec6c2014
commit ffec6c2014
parent d097531926
1 changed files with 129 additions and 0 deletions
--- a/common/memcpySSE2.h
+++ b/common/memcpySSE2.h
@ -0,0 +1,129 @@
+/*
+KVMGFX Client - A KVM Client for VGA Passthrough
+Copyright (C) 2017 Geoffrey McRae <geoff@hostfission.com>
+https://looking-glass.hostfission.com
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; either version 2 of the License, or (at your option) any later
+version.
+
+This program is distributed in the hope that it will be useful, but WITHOUT ANY
+WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A
+PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+*/
+
+#pragma once
+#include <stdbool.h>
+#include <stdint.h>
+#include <string.h>
+#include <tmmintrin.h>
+#include <immintrin.h>
+
+#include "debug.h"
+
+#define OP(...) #__VA_ARGS__ "\n\t"
+
+inline static void memcpySSE(void *dst, const void * src, size_t length)
+{
+#if defined(__x86_64__) || defined(__i386__)
+  void * end = dst + (length & ~0x7F);
+  size_t rem = (7 - ((length & 0x7F) >> 4)) * 10;
+
+  __asm__ __volatile__ (
+    // save the registers we intend to alter, failure to do so causes problems
+    // when gcc -O3 is used
+    OP(push %[dst])
+    OP(push %[src])
+    OP(push %[end])
+
+    // perform 128 byte SIMD block copy
+    OP(cmp %[dst],%[end])
+    OP(je ramain_%=)
+    OP(loop_%=:)
+    OP(vmovaps  0x00(%[src]),%%xmm0)
+    OP(vmovaps  0x10(%[src]),%%xmm1)
+    OP(vmovaps  0x20(%[src]),%%xmm2)
+    OP(vmovaps  0x30(%[src]),%%xmm3)
+    OP(vmovaps  0x40(%[src]),%%xmm4)
+    OP(vmovaps  0x50(%[src]),%%xmm5)
+    OP(vmovaps  0x60(%[src]),%%xmm6)
+    OP(vmovaps  0x70(%[src]),%%xmm7)
+    OP(vmovntdq %%xmm0,0x00(%[dst]))
+    OP(vmovntdq %%xmm1,0x10(%[dst]))
+    OP(vmovntdq %%xmm2,0x20(%[dst]))
+    OP(vmovntdq %%xmm3,0x30(%[dst]))
+    OP(vmovntdq %%xmm4,0x40(%[dst]))
+    OP(vmovntdq %%xmm5,0x50(%[dst]))
+    OP(vmovntdq %%xmm6,0x60(%[dst]))
+    OP(vmovntdq %%xmm7,0x70(%[dst]))
+    OP(add      $0x80,%[dst])
+    OP(add      $0x80,%[src])
+    OP(cmp      %[dst],%[end])
+    OP(jne      loop_%=)
+
+    // copy any remaining 16 byte blocks
+    OP(remain_%=:)
+#ifdef __x86_64__
+    OP(leaq (%%rip), %[end])
+    OP(add  $10,%[end])
+#else
+    OP(call .+5)
+    OP(pop  %[end])
+    OP(add  $8,%[end])
+#endif
+    OP(add  %[rem],%[end])
+    OP(jmp  *%[end])
+
+   // jump table
+    OP(vmovaps  0x60(%[src]),%%xmm0)
+    OP(vmovntdq %%xmm0,0x60(%[dst]))
+    OP(vmovaps  0x50(%[src]),%%xmm1)
+    OP(vmovntdq %%xmm1,0x50(%[dst]))
+    OP(vmovaps  0x40(%[src]),%%xmm2)
+    OP(vmovntdq %%xmm2,0x40(%[dst]))
+    OP(vmovaps  0x30(%[src]),%%xmm3)
+    OP(vmovntdq %%xmm3,0x30(%[dst]))
+    OP(vmovaps  0x20(%[src]),%%xmm4)
+    OP(vmovntdq %%xmm4,0x20(%[dst]))
+    OP(vmovaps  0x10(%[src]),%%xmm5)
+    OP(vmovntdq %%xmm5,0x10(%[dst]))
+    OP(vmovaps  0x00(%[src]),%%xmm6)
+    OP(vmovntdq %%xmm6,0x00(%[dst]))
+
+   // alignment as the previous two instructions are only 4 bytes
+    OP(nop)
+    OP(nop)
+
+   // restore the registers
+    OP(pop %[end])
+    OP(pop %[src])
+    OP(pop %[dst])
+   :
+   : [dst]"r" (dst),
+     [src]"r" (src),
+     [end]"c" (end),
+     [rem]"d" (rem)
+   : "xmm0",
+     "xmm1",
+     "xmm2",
+     "xmm3",
+     "xmm4",
+     "xmm5",
+     "xmm6",
+     "xmm7",
+     "memory"
+  );
+
+  //copy any remaining bytes
+  for(size_t i = (length & 0xF); i; --i)
+    ((uint8_t *)dst)[length - i] =
+      ((uint8_t *)src)[length - i];
+#else
+  memcpy(dst, src, length);
+#endif
+}