diff options
Diffstat (limited to 'libbb/hash_md5_sha_x86-64.S')
-rw-r--r-- | libbb/hash_md5_sha_x86-64.S | 127 |
1 files changed, 65 insertions, 62 deletions
diff --git a/libbb/hash_md5_sha_x86-64.S b/libbb/hash_md5_sha_x86-64.S index 51fde082a..f0daa30f6 100644 --- a/libbb/hash_md5_sha_x86-64.S +++ b/libbb/hash_md5_sha_x86-64.S | |||
@@ -27,60 +27,68 @@ sha1_process_block64: | |||
27 | # xmm7: all round constants | 27 | # xmm7: all round constants |
28 | # -64(%rsp): area for passing RCONST + W[] from vector to integer units | 28 | # -64(%rsp): area for passing RCONST + W[] from vector to integer units |
29 | 29 | ||
30 | movl 80(%rdi), %eax # a = ctx->hash[0] | ||
31 | movl 84(%rdi), %ebx # b = ctx->hash[1] | ||
32 | movl 88(%rdi), %ecx # c = ctx->hash[2] | ||
33 | movl 92(%rdi), %edx # d = ctx->hash[3] | ||
34 | movl 96(%rdi), %ebp # e = ctx->hash[4] | ||
35 | |||
30 | movaps sha1const(%rip), %xmm7 | 36 | movaps sha1const(%rip), %xmm7 |
31 | movaps bswap32_mask(%rip), %xmm4 | ||
32 | pshufd $0x00, %xmm7, %xmm6 | 37 | pshufd $0x00, %xmm7, %xmm6 |
33 | 38 | ||
34 | # Load W[] to xmm0..3, byteswapping on the fly. | 39 | # Load W[] to xmm0..3, byteswapping on the fly. |
35 | # | 40 | # |
36 | # For iterations 0..15, we pass RCONST+W[] in rsi,r8..r14 | 41 | # For iterations 0..15, we pass W[] in rsi,r8..r14 |
37 | # for use in RD1As instead of spilling them to stack. | 42 | # for use in RD1As instead of spilling them to stack. |
43 | # We lose parallelized addition of RCONST, but LEA | ||
44 | # can do two additions at once, so it is probably a wash. | ||
38 | # (We use rsi instead of rN because this makes two | 45 | # (We use rsi instead of rN because this makes two |
39 | # ADDs in two first RD1As shorter by one byte). | 46 | # LEAs in two first RD1As shorter by one byte). |
40 | movups 16*0(%rdi), %xmm0 | 47 | movq 4*0(%rdi), %rsi |
41 | pshufb %xmm4, %xmm0 | 48 | movq 4*2(%rdi), %r8 |
42 | movaps %xmm0, %xmm5 | 49 | bswapq %rsi |
43 | paddd %xmm6, %xmm5 | 50 | bswapq %r8 |
44 | movq %xmm5, %rsi | 51 | rolq $32, %rsi # rsi = W[1]:W[0] |
45 | # pextrq $1, %xmm5, %r8 #SSE4.1 insn | 52 | rolq $32, %r8 # r8 = W[3]:W[2] |
46 | # movhpd %xmm5, %r8 #can only move to mem, not to reg | 53 | movq %rsi, %xmm0 |
47 | shufps $0x0e, %xmm5, %xmm5 | 54 | movq %r8, %xmm4 |
48 | movq %xmm5, %r8 | 55 | punpcklqdq %xmm4, %xmm0 # xmm0 = r8:rsi = (W[0],W[1],W[2],W[3]) |
49 | 56 | # movaps %xmm0, %xmm4 # add RCONST, spill to stack | |
50 | movups 16*1(%rdi), %xmm1 | 57 | # paddd %xmm6, %xmm4 |
51 | pshufb %xmm4, %xmm1 | 58 | # movups %xmm4, -64+16*0(%rsp) |
52 | movaps %xmm1, %xmm5 | ||
53 | paddd %xmm6, %xmm5 | ||
54 | movq %xmm5, %r9 | ||
55 | shufps $0x0e, %xmm5, %xmm5 | ||
56 | movq %xmm5, %r10 | ||
57 | 59 | ||
58 | movups 16*2(%rdi), %xmm2 | 60 | movq 4*4(%rdi), %r9 |
59 | pshufb %xmm4, %xmm2 | 61 | movq 4*6(%rdi), %r10 |
60 | movaps %xmm2, %xmm5 | 62 | bswapq %r9 |
61 | paddd %xmm6, %xmm5 | 63 | bswapq %r10 |
62 | movq %xmm5, %r11 | 64 | rolq $32, %r9 # r9 = W[5]:W[4] |
63 | shufps $0x0e, %xmm5, %xmm5 | 65 | rolq $32, %r10 # r10 = W[7]:W[6] |
64 | movq %xmm5, %r12 | 66 | movq %r9, %xmm1 |
67 | movq %r10, %xmm4 | ||
68 | punpcklqdq %xmm4, %xmm1 # xmm1 = r10:r9 = (W[4],W[5],W[6],W[7]) | ||
65 | 69 | ||
66 | movups 16*3(%rdi), %xmm3 | 70 | movq 4*8(%rdi), %r11 |
67 | pshufb %xmm4, %xmm3 | 71 | movq 4*10(%rdi), %r12 |
68 | movaps %xmm3, %xmm5 | 72 | bswapq %r11 |
69 | paddd %xmm6, %xmm5 | 73 | bswapq %r12 |
70 | movq %xmm5, %r13 | 74 | rolq $32, %r11 # r11 = W[9]:W[8] |
71 | shufps $0x0e, %xmm5, %xmm5 | 75 | rolq $32, %r12 # r12 = W[11]:W[10] |
72 | movq %xmm5, %r14 | 76 | movq %r11, %xmm2 |
77 | movq %r12, %xmm4 | ||
78 | punpcklqdq %xmm4, %xmm2 # xmm2 = r12:r11 = (W[8],W[9],W[10],W[11]) | ||
73 | 79 | ||
74 | # MOVQs to GPRs (above) have somewhat high latency. | 80 | movq 4*12(%rdi), %r13 |
75 | # Load hash[] while they are completing: | 81 | movq 4*14(%rdi), %r14 |
76 | movl 80(%rdi), %eax # a = ctx->hash[0] | 82 | bswapq %r13 |
77 | movl 84(%rdi), %ebx # b = ctx->hash[1] | 83 | bswapq %r14 |
78 | movl 88(%rdi), %ecx # c = ctx->hash[2] | 84 | rolq $32, %r13 # r13 = W[13]:W[12] |
79 | movl 92(%rdi), %edx # d = ctx->hash[3] | 85 | rolq $32, %r14 # r14 = W[15]:W[14] |
80 | movl 96(%rdi), %ebp # e = ctx->hash[4] | 86 | movq %r13, %xmm3 |
87 | movq %r14, %xmm4 | ||
88 | punpcklqdq %xmm4, %xmm3 # xmm3 = r14:r13 = (W[12],W[13],W[14],W[15]) | ||
81 | 89 | ||
82 | # 0 | 90 | # 0 |
83 | addl %esi, %ebp # e += RCONST + W[n] | 91 | leal 0x5A827999(%rbp,%rsi), %ebp # e += RCONST + W[n] |
84 | shrq $32, %rsi | 92 | shrq $32, %rsi |
85 | movl %ecx, %edi # c | 93 | movl %ecx, %edi # c |
86 | xorl %edx, %edi # ^d | 94 | xorl %edx, %edi # ^d |
@@ -92,7 +100,7 @@ sha1_process_block64: | |||
92 | addl %edi, %ebp # e += rotl32(a,5) | 100 | addl %edi, %ebp # e += rotl32(a,5) |
93 | rorl $2, %ebx # b = rotl32(b,30) | 101 | rorl $2, %ebx # b = rotl32(b,30) |
94 | # 1 | 102 | # 1 |
95 | addl %esi, %edx # e += RCONST + W[n] | 103 | leal 0x5A827999(%rdx,%rsi), %edx # e += RCONST + W[n] |
96 | movl %ebx, %edi # c | 104 | movl %ebx, %edi # c |
97 | xorl %ecx, %edi # ^d | 105 | xorl %ecx, %edi # ^d |
98 | andl %eax, %edi # &b | 106 | andl %eax, %edi # &b |
@@ -103,7 +111,7 @@ sha1_process_block64: | |||
103 | addl %edi, %edx # e += rotl32(a,5) | 111 | addl %edi, %edx # e += rotl32(a,5) |
104 | rorl $2, %eax # b = rotl32(b,30) | 112 | rorl $2, %eax # b = rotl32(b,30) |
105 | # 2 | 113 | # 2 |
106 | addl %r8d, %ecx # e += RCONST + W[n] | 114 | leal 0x5A827999(%rcx,%r8), %ecx # e += RCONST + W[n] |
107 | shrq $32, %r8 | 115 | shrq $32, %r8 |
108 | movl %eax, %edi # c | 116 | movl %eax, %edi # c |
109 | xorl %ebx, %edi # ^d | 117 | xorl %ebx, %edi # ^d |
@@ -115,7 +123,7 @@ sha1_process_block64: | |||
115 | addl %edi, %ecx # e += rotl32(a,5) | 123 | addl %edi, %ecx # e += rotl32(a,5) |
116 | rorl $2, %ebp # b = rotl32(b,30) | 124 | rorl $2, %ebp # b = rotl32(b,30) |
117 | # 3 | 125 | # 3 |
118 | addl %r8d, %ebx # e += RCONST + W[n] | 126 | leal 0x5A827999(%rbx,%r8), %ebx # e += RCONST + W[n] |
119 | movl %ebp, %edi # c | 127 | movl %ebp, %edi # c |
120 | xorl %eax, %edi # ^d | 128 | xorl %eax, %edi # ^d |
121 | andl %edx, %edi # &b | 129 | andl %edx, %edi # &b |
@@ -126,7 +134,7 @@ sha1_process_block64: | |||
126 | addl %edi, %ebx # e += rotl32(a,5) | 134 | addl %edi, %ebx # e += rotl32(a,5) |
127 | rorl $2, %edx # b = rotl32(b,30) | 135 | rorl $2, %edx # b = rotl32(b,30) |
128 | # 4 | 136 | # 4 |
129 | addl %r9d, %eax # e += RCONST + W[n] | 137 | leal 0x5A827999(%rax,%r9), %eax # e += RCONST + W[n] |
130 | shrq $32, %r9 | 138 | shrq $32, %r9 |
131 | movl %edx, %edi # c | 139 | movl %edx, %edi # c |
132 | xorl %ebp, %edi # ^d | 140 | xorl %ebp, %edi # ^d |
@@ -138,7 +146,7 @@ sha1_process_block64: | |||
138 | addl %edi, %eax # e += rotl32(a,5) | 146 | addl %edi, %eax # e += rotl32(a,5) |
139 | rorl $2, %ecx # b = rotl32(b,30) | 147 | rorl $2, %ecx # b = rotl32(b,30) |
140 | # 5 | 148 | # 5 |
141 | addl %r9d, %ebp # e += RCONST + W[n] | 149 | leal 0x5A827999(%rbp,%r9), %ebp # e += RCONST + W[n] |
142 | movl %ecx, %edi # c | 150 | movl %ecx, %edi # c |
143 | xorl %edx, %edi # ^d | 151 | xorl %edx, %edi # ^d |
144 | andl %ebx, %edi # &b | 152 | andl %ebx, %edi # &b |
@@ -149,7 +157,7 @@ sha1_process_block64: | |||
149 | addl %edi, %ebp # e += rotl32(a,5) | 157 | addl %edi, %ebp # e += rotl32(a,5) |
150 | rorl $2, %ebx # b = rotl32(b,30) | 158 | rorl $2, %ebx # b = rotl32(b,30) |
151 | # 6 | 159 | # 6 |
152 | addl %r10d, %edx # e += RCONST + W[n] | 160 | leal 0x5A827999(%rdx,%r10), %edx # e += RCONST + W[n] |
153 | shrq $32, %r10 | 161 | shrq $32, %r10 |
154 | movl %ebx, %edi # c | 162 | movl %ebx, %edi # c |
155 | xorl %ecx, %edi # ^d | 163 | xorl %ecx, %edi # ^d |
@@ -161,7 +169,7 @@ sha1_process_block64: | |||
161 | addl %edi, %edx # e += rotl32(a,5) | 169 | addl %edi, %edx # e += rotl32(a,5) |
162 | rorl $2, %eax # b = rotl32(b,30) | 170 | rorl $2, %eax # b = rotl32(b,30) |
163 | # 7 | 171 | # 7 |
164 | addl %r10d, %ecx # e += RCONST + W[n] | 172 | leal 0x5A827999(%rcx,%r10), %ecx # e += RCONST + W[n] |
165 | movl %eax, %edi # c | 173 | movl %eax, %edi # c |
166 | xorl %ebx, %edi # ^d | 174 | xorl %ebx, %edi # ^d |
167 | andl %ebp, %edi # &b | 175 | andl %ebp, %edi # &b |
@@ -202,7 +210,7 @@ sha1_process_block64: | |||
202 | paddd %xmm6, %xmm5 | 210 | paddd %xmm6, %xmm5 |
203 | movups %xmm5, -64+16*0(%rsp) | 211 | movups %xmm5, -64+16*0(%rsp) |
204 | # 8 | 212 | # 8 |
205 | addl %r11d, %ebx # e += RCONST + W[n] | 213 | leal 0x5A827999(%rbx,%r11), %ebx # e += RCONST + W[n] |
206 | shrq $32, %r11 | 214 | shrq $32, %r11 |
207 | movl %ebp, %edi # c | 215 | movl %ebp, %edi # c |
208 | xorl %eax, %edi # ^d | 216 | xorl %eax, %edi # ^d |
@@ -214,7 +222,7 @@ sha1_process_block64: | |||
214 | addl %edi, %ebx # e += rotl32(a,5) | 222 | addl %edi, %ebx # e += rotl32(a,5) |
215 | rorl $2, %edx # b = rotl32(b,30) | 223 | rorl $2, %edx # b = rotl32(b,30) |
216 | # 9 | 224 | # 9 |
217 | addl %r11d, %eax # e += RCONST + W[n] | 225 | leal 0x5A827999(%rax,%r11), %eax # e += RCONST + W[n] |
218 | movl %edx, %edi # c | 226 | movl %edx, %edi # c |
219 | xorl %ebp, %edi # ^d | 227 | xorl %ebp, %edi # ^d |
220 | andl %ecx, %edi # &b | 228 | andl %ecx, %edi # &b |
@@ -225,7 +233,7 @@ sha1_process_block64: | |||
225 | addl %edi, %eax # e += rotl32(a,5) | 233 | addl %edi, %eax # e += rotl32(a,5) |
226 | rorl $2, %ecx # b = rotl32(b,30) | 234 | rorl $2, %ecx # b = rotl32(b,30) |
227 | # 10 | 235 | # 10 |
228 | addl %r12d, %ebp # e += RCONST + W[n] | 236 | leal 0x5A827999(%rbp,%r12), %ebp # e += RCONST + W[n] |
229 | shrq $32, %r12 | 237 | shrq $32, %r12 |
230 | movl %ecx, %edi # c | 238 | movl %ecx, %edi # c |
231 | xorl %edx, %edi # ^d | 239 | xorl %edx, %edi # ^d |
@@ -237,7 +245,7 @@ sha1_process_block64: | |||
237 | addl %edi, %ebp # e += rotl32(a,5) | 245 | addl %edi, %ebp # e += rotl32(a,5) |
238 | rorl $2, %ebx # b = rotl32(b,30) | 246 | rorl $2, %ebx # b = rotl32(b,30) |
239 | # 11 | 247 | # 11 |
240 | addl %r12d, %edx # e += RCONST + W[n] | 248 | leal 0x5A827999(%rdx,%r12), %edx # e += RCONST + W[n] |
241 | movl %ebx, %edi # c | 249 | movl %ebx, %edi # c |
242 | xorl %ecx, %edi # ^d | 250 | xorl %ecx, %edi # ^d |
243 | andl %eax, %edi # &b | 251 | andl %eax, %edi # &b |
@@ -279,7 +287,7 @@ sha1_process_block64: | |||
279 | paddd %xmm6, %xmm5 | 287 | paddd %xmm6, %xmm5 |
280 | movups %xmm5, -64+16*1(%rsp) | 288 | movups %xmm5, -64+16*1(%rsp) |
281 | # 12 | 289 | # 12 |
282 | addl %r13d, %ecx # e += RCONST + W[n] | 290 | leal 0x5A827999(%rcx,%r13), %ecx # e += RCONST + W[n] |
283 | shrq $32, %r13 | 291 | shrq $32, %r13 |
284 | movl %eax, %edi # c | 292 | movl %eax, %edi # c |
285 | xorl %ebx, %edi # ^d | 293 | xorl %ebx, %edi # ^d |
@@ -291,7 +299,7 @@ sha1_process_block64: | |||
291 | addl %edi, %ecx # e += rotl32(a,5) | 299 | addl %edi, %ecx # e += rotl32(a,5) |
292 | rorl $2, %ebp # b = rotl32(b,30) | 300 | rorl $2, %ebp # b = rotl32(b,30) |
293 | # 13 | 301 | # 13 |
294 | addl %r13d, %ebx # e += RCONST + W[n] | 302 | leal 0x5A827999(%rbx,%r13), %ebx # e += RCONST + W[n] |
295 | movl %ebp, %edi # c | 303 | movl %ebp, %edi # c |
296 | xorl %eax, %edi # ^d | 304 | xorl %eax, %edi # ^d |
297 | andl %edx, %edi # &b | 305 | andl %edx, %edi # &b |
@@ -302,7 +310,7 @@ sha1_process_block64: | |||
302 | addl %edi, %ebx # e += rotl32(a,5) | 310 | addl %edi, %ebx # e += rotl32(a,5) |
303 | rorl $2, %edx # b = rotl32(b,30) | 311 | rorl $2, %edx # b = rotl32(b,30) |
304 | # 14 | 312 | # 14 |
305 | addl %r14d, %eax # e += RCONST + W[n] | 313 | leal 0x5A827999(%rax,%r14), %eax # e += RCONST + W[n] |
306 | shrq $32, %r14 | 314 | shrq $32, %r14 |
307 | movl %edx, %edi # c | 315 | movl %edx, %edi # c |
308 | xorl %ebp, %edi # ^d | 316 | xorl %ebp, %edi # ^d |
@@ -314,7 +322,7 @@ sha1_process_block64: | |||
314 | addl %edi, %eax # e += rotl32(a,5) | 322 | addl %edi, %eax # e += rotl32(a,5) |
315 | rorl $2, %ecx # b = rotl32(b,30) | 323 | rorl $2, %ecx # b = rotl32(b,30) |
316 | # 15 | 324 | # 15 |
317 | addl %r14d, %ebp # e += RCONST + W[n] | 325 | leal 0x5A827999(%rbp,%r14), %ebp # e += RCONST + W[n] |
318 | movl %ecx, %edi # c | 326 | movl %ecx, %edi # c |
319 | xorl %edx, %edi # ^d | 327 | xorl %edx, %edi # ^d |
320 | andl %ebx, %edi # &b | 328 | andl %ebx, %edi # &b |
@@ -1467,11 +1475,6 @@ sha1_process_block64: | |||
1467 | ret | 1475 | ret |
1468 | .size sha1_process_block64, .-sha1_process_block64 | 1476 | .size sha1_process_block64, .-sha1_process_block64 |
1469 | 1477 | ||
1470 | .section .rodata.cst16.bswap32_mask, "aM", @progbits, 16 | ||
1471 | .balign 16 | ||
1472 | bswap32_mask: | ||
1473 | .octa 0x0c0d0e0f08090a0b0405060700010203 | ||
1474 | |||
1475 | .section .rodata.cst16.sha1const, "aM", @progbits, 16 | 1478 | .section .rodata.cst16.sha1const, "aM", @progbits, 16 |
1476 | .balign 16 | 1479 | .balign 16 |
1477 | sha1const: | 1480 | sha1const: |