]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - arch/x86/include/asm/xor_64.h
x86/xor: Add alternative SSE implementation only prefetching once per 64-byte line
[~shefty/rdma-dev.git] / arch / x86 / include / asm / xor_64.h
1 #ifndef _ASM_X86_XOR_64_H
2 #define _ASM_X86_XOR_64_H
3
4 static struct xor_block_template xor_block_sse = {
5         .name = "generic_sse",
6         .do_2 = xor_sse_2,
7         .do_3 = xor_sse_3,
8         .do_4 = xor_sse_4,
9         .do_5 = xor_sse_5,
10 };
11
12
13 /* Also try the AVX routines */
14 #include <asm/xor_avx.h>
15
16 /* We force the use of the SSE xor block because it can write around L2.
17    We may also be able to load into the L1 only depending on how the cpu
18    deals with a load to a line that is being prefetched.  */
19 #undef XOR_TRY_TEMPLATES
20 #define XOR_TRY_TEMPLATES                       \
21 do {                                            \
22         AVX_XOR_SPEED;                          \
23         xor_speed(&xor_block_sse_pf64);         \
24         xor_speed(&xor_block_sse);              \
25 } while (0)
26
27 #endif /* _ASM_X86_XOR_64_H */