39 #define PF0(x)		"	prefetchnta "PF_OFFS(x)"(%[p1])		;\n"
40 #define LD(x, y)	"	movaps "OFFS(x)"(%[p1]), %%xmm"#y"	;\n"
41 #define ST(x, y)	"	movaps %%xmm"#y", "OFFS(x)"(%[p1])	;\n"
55 			op(i + 1, 1)			\
60 xor_sse_2(unsigned long bytes, unsigned long * __restrict p1,  in xor_sse_2()  argument
71 			LD(i + 1, 1)			\  in xor_sse_2()
79 			XO1(i + 1, 1)			\  in xor_sse_2()
83 			ST(i + 1, 1)			\  in xor_sse_2()
92 	" 1:                            ;\n"  in xor_sse_2()
99 	"       add %[inc], %[p1]       ;\n"  in xor_sse_2()
102 	"       jnz 1b                  ;\n"  in xor_sse_2()
104 	  [p1] "+r" (p1), [p2] "+r" (p2)  in xor_sse_2()
112 xor_sse_2_pf64(unsigned long bytes, unsigned long * __restrict p1,  in xor_sse_2_pf64()  argument
127 	" 1:                            ;\n"  in xor_sse_2_pf64()
134 	"       add %[inc], %[p1]       ;\n"  in xor_sse_2_pf64()
137 	"       jnz 1b                  ;\n"  in xor_sse_2_pf64()
139 	  [p1] "+r" (p1), [p2] "+r" (p2)  in xor_sse_2_pf64()
147 xor_sse_3(unsigned long bytes, unsigned long * __restrict p1,  in xor_sse_3()  argument
161 			LD(i + 1, 1)			\  in xor_sse_3()
169 			XO1(i + 1, 1)			\  in xor_sse_3()
173 			XO2(i + 1, 1)			\  in xor_sse_3()
177 			ST(i + 1, 1)			\  in xor_sse_3()
186 	" 1:                            ;\n"  in xor_sse_3()
193 	"       add %[inc], %[p1]       ;\n"  in xor_sse_3()
197 	"       jnz 1b                  ;\n"  in xor_sse_3()
199 	  [p1] "+r" (p1), [p2] "+r" (p2), [p3] "+r" (p3)  in xor_sse_3()
207 xor_sse_3_pf64(unsigned long bytes, unsigned long * __restrict p1,  in xor_sse_3_pf64()  argument
224 	" 1:                            ;\n"  in xor_sse_3_pf64()
231 	"       add %[inc], %[p1]       ;\n"  in xor_sse_3_pf64()
235 	"       jnz 1b                  ;\n"  in xor_sse_3_pf64()
237 	  [p1] "+r" (p1), [p2] "+r" (p2), [p3] "+r" (p3)  in xor_sse_3_pf64()
245 xor_sse_4(unsigned long bytes, unsigned long * __restrict p1,  in xor_sse_4()  argument
260 			LD(i + 1, 1)			\  in xor_sse_4()
266 			XO1(i + 1, 1)			\  in xor_sse_4()
274 			XO2(i + 1, 1)			\  in xor_sse_4()
278 			XO3(i + 1, 1)			\  in xor_sse_4()
282 			ST(i + 1, 1)			\  in xor_sse_4()
291 	" 1:                            ;\n"  in xor_sse_4()
298 	"       add %[inc], %[p1]       ;\n"  in xor_sse_4()
303 	"       jnz 1b                  ;\n"  in xor_sse_4()
304 	: [cnt] "+r" (lines), [p1] "+r" (p1),  in xor_sse_4()
313 xor_sse_4_pf64(unsigned long bytes, unsigned long * __restrict p1,  in xor_sse_4_pf64()  argument
332 	" 1:                            ;\n"  in xor_sse_4_pf64()
339 	"       add %[inc], %[p1]       ;\n"  in xor_sse_4_pf64()
344 	"       jnz 1b                  ;\n"  in xor_sse_4_pf64()
345 	: [cnt] "+r" (lines), [p1] "+r" (p1),  in xor_sse_4_pf64()
354 xor_sse_5(unsigned long bytes, unsigned long * __restrict p1,  in xor_sse_5()  argument
370 			LD(i + 1, 1)			\  in xor_sse_5()
376 			XO1(i + 1, 1)			\  in xor_sse_5()
382 			XO2(i + 1, 1)			\  in xor_sse_5()
390 			XO3(i + 1, 1)			\  in xor_sse_5()
394 			XO4(i + 1, 1)			\  in xor_sse_5()
398 			ST(i + 1, 1)			\  in xor_sse_5()
407 	" 1:                            ;\n"  in xor_sse_5()
414 	"       add %[inc], %[p1]       ;\n"  in xor_sse_5()
420 	"       jnz 1b                  ;\n"  in xor_sse_5()
421 	: [cnt] "+r" (lines), [p1] "+r" (p1), [p2] "+r" (p2),  in xor_sse_5()
430 xor_sse_5_pf64(unsigned long bytes, unsigned long * __restrict p1,  in xor_sse_5_pf64()  argument
451 	" 1:                            ;\n"  in xor_sse_5_pf64()
458 	"       add %[inc], %[p1]       ;\n"  in xor_sse_5_pf64()
464 	"       jnz 1b                  ;\n"  in xor_sse_5_pf64()
465 	: [cnt] "+r" (lines), [p1] "+r" (p1), [p2] "+r" (p2),  in xor_sse_5_pf64()