diff options
Diffstat (limited to 'modules/asm/asm_amd64/mp_monty.S')
| -rw-r--r-- | modules/asm/asm_amd64/mp_monty.S | 397 |
1 files changed, 397 insertions, 0 deletions
diff --git a/modules/asm/asm_amd64/mp_monty.S b/modules/asm/asm_amd64/mp_monty.S new file mode 100644 index 000000000..3dd4040bc --- /dev/null +++ b/modules/asm/asm_amd64/mp_monty.S @@ -0,0 +1,397 @@ +/************************************************* +* Montgomery Reduction Source File * +* (C) 2008 Jack Lloyd * +*************************************************/ + +#include <botan/asm_macr.h> + +START_LISTING(mp_monty.S) + +START_FUNCTION(bigint_monty_redc) + pushq %r15 # + pushq %r14 # + pushq %r13 # + pushq %r12 # + pushq %rbp # + pushq %rbx # + + movq %rdi, %r14 # z + movq %rdx, %r12 # x + movl %esi, %ebp # z_size + + xorl %esi, %esi # j.76 + movq %r8, -16(%rsp) # u, u + movl %ecx, %ebx # x_size, x_size + movl %ecx, %r8d # x_size, blocks_of_8 + andl $-8, %r8d #, blocks_of_8 + testl %ecx, %ecx # x_size + je .L3 #, + mov %ecx, %eax # x_size, pretmp.71 + leal 1(%rbx), %r15d #, k.73 + salq $3, %rax #, + xorl %r13d, %r13d # j + movq %rax, -8(%rsp) #, pretmp.21 + .p2align 4,,10 + .p2align 3 +.L11: + mov %r13d, %eax # j, j + movq -16(%rsp), %rdi # u, y + leaq (%r14,%rax,8), %r11 #, z_j + xorl %r9d, %r9d # i + imulq (%r11), %rdi #* z_j, y + xorl %r10d, %r10d # carry + testl %r8d, %r8d # blocks_of_8 + je .L7 #, + .p2align 4,,10 + .p2align 3 +.LOOP_MUL_ADD: + mov %r9d, %ecx # i, i + addl $8, %r9d #, i + salq $3, %rcx #, D.2315 + leaq (%r11,%rcx), %rsi #, tmp130 + leaq (%r12,%rcx), %rcx #, tmp131 + + movq 8*0(%rcx), %rax + mulq %rdi # y + addq %r10, %rax # carry + adcq $0,%rdx + addq 8*0(%rsi), %rax + adcq $0,%rdx + movq %rdx,%r10 # carry + movq %rax, 8*0 (%rsi) + + movq 8*1(%rcx), %rax + mulq %rdi # y + addq %r10, %rax # carry + adcq $0,%rdx + addq 8*1(%rsi), %rax + adcq $0,%rdx + movq %rdx,%r10 # carry + movq %rax, 8*1 (%rsi) + + movq 8*2(%rcx), %rax + mulq %rdi # y + addq %r10, %rax # carry + adcq $0,%rdx + addq 8*2(%rsi), %rax + adcq $0,%rdx + movq %rdx,%r10 # carry + movq %rax, 8*2 (%rsi) + + movq 8*3(%rcx), %rax + mulq %rdi # y + addq %r10, %rax # carry + adcq $0,%rdx + addq 8*3(%rsi), %rax + adcq $0,%rdx + movq %rdx,%r10 # carry + movq %rax, 8*3 (%rsi) + + movq 8*4(%rcx), %rax + mulq %rdi # y + addq %r10, %rax # carry + adcq $0,%rdx + addq 8*4(%rsi), %rax + adcq $0,%rdx + movq %rdx,%r10 # carry + movq %rax, 8*4 (%rsi) + + movq 8*5(%rcx), %rax + mulq %rdi # y + addq %r10, %rax # carry + adcq $0,%rdx + addq 8*5(%rsi), %rax + adcq $0,%rdx + movq %rdx,%r10 # carry + movq %rax, 8*5 (%rsi) + + movq 8*6(%rcx), %rax + mulq %rdi # y + addq %r10, %rax # carry + adcq $0,%rdx + addq 8*6(%rsi), %rax + adcq $0,%rdx + movq %rdx,%r10 # carry + movq %rax, 8*6 (%rsi) + + movq 8*7(%rcx), %rax + mulq %rdi # y + addq %r10, %rax # carry + adcq $0,%rdx + addq 8*7(%rsi), %rax + adcq $0,%rdx + movq %rdx,%r10 # carry + movq %rax, 8*7 (%rsi) + + cmpl %r9d, %r8d # i, blocks_of_8 + jne .LOOP_MUL_ADD #, + cmpl %r8d, %ebx # blocks_of_8, x_size + je .L8 #, +.L7: + movl %r8d, %esi # blocks_of_8, i + .p2align 4,,10 + .p2align 3 +.L5: + mov %esi, %eax # i, i + movq %rdi, %rcx # y, b + leaq (%r11, %rax,8), %r9 #, D.2325 + incl %esi # i + movq (%r12, %rax,8), %rax #* x, tmp133 + + mulq %rcx # b + addq (%r9), %rax #* D.2325, a + adcq $0,%rdx # + addq %r10, %rax # carry, a + adcq $0,%rdx # + + cmpl %esi, %ebx # i, x_size + movq %rdx, %r10 #, carry + movq %rax, (%r9) # a,* D.2325 + jne .L5 #, +.L8: + movq -8(%rsp), %rdx # pretmp.21, + leaq (%r11,%rdx), %rax #, D.2332 + movq (%rax), %rcx #* D.2332, D.2333 + leaq (%r10,%rcx), %rdx #, z_sum + movq %rdx, (%rax) # z_sum,* D.2332 + cmpq %rdx, %rcx # z_sum, D.2333 + jbe .L9 #, + cmpl %ebp, %r15d # z_size, k.73 + je .L9 #, + movl %r15d, %ecx # k.73, k + jmp .L10 # + .p2align 4,,10 + .p2align 3 +.L31: + incl %ecx # k + cmpl %ecx, %ebp # k, z_size + .p2align 4,,4 + .p2align 3 + je .L9 #, +.L10: + mov %ecx, %edx # k, k + leaq (%r11,%rdx,8), %rdx #, D.2342 + movq (%rdx), %rax #* D.2342, tmp136 + incq %rax # D.2344 + movq %rax, (%rdx) # D.2344,* D.2342 + testq %rax, %rax # D.2344 + je .L31 #, +.L9: + incl %r13d # j + decl %ebp # z_size + cmpl %r13d, %ebx # j, x_size + jne .L11 #, + movl %ebx, %esi # x_size, j.76 +.L3: + leal (%rbx,%rbx), %eax #, tmp137 + mov %eax, %eax + leaq (%r14, %rax,8), %rdi #, D.2349 + cmpq $0, (%rdi) #,* D.2349 + jne .L12 #, + testl %ebx, %ebx # x_size + je .L12 #, + leal -1(%rbx), %ecx #, j + leal (%rsi,%rcx), %edx #, tmp141 + mov %ecx, %eax # j, j + movq (%r14,%rdx,8), %rbp #* z, + cmpq %rbp, (%r12, %rax,8) #,* x + jb .L12 #, + ja .L_EXIT #, + leal -2(%rsi,%rbx), %edx #, ivtmp.45 + jmp .L14 # + .p2align 4,,10 + .p2align 3 +.L15: + mov %edx, %eax # ivtmp.45, ivtmp.45 + decl %ecx # j + movq (%r14, %rax,8), %rsi #* z, D.2360 + mov %ecx, %eax # j, j + movq (%r12, %rax,8), %rax #* x, temp.68 + cmpq %rax, %rsi + ja .L12 #, + decl %edx # ivtmp.45 + cmpq %rax, %rsi + jb .L_EXIT #, +.L14: + testl %ecx, %ecx # j + jne .L15 #, +.L12: + xorl %ecx, %ecx # j + xorl %r10d, %r10d # carry + mov %ebx, %esi # x_size, pretmp.19 + testl %r8d, %r8d # blocks_of_8 + je .L17 #, + .p2align 4,,10 + .p2align 3 +.L22: + mov %ecx, %edx # j, D.2375 + addl $8, %ecx #, j + leaq (%rdx,%rsi), %rax #, tmp146 + leaq (%r12,%rdx,8), %rdx #, tmp150 + leaq (%r14, %rax,8), %rax #, tmp148 + + rorq %r10 # carry + + movq 8*0(%rdx), %r10 + sbbq %r10, 8*0(%rax) + + movq 8*1(%rdx), %r10 + sbbq %r10, 8*1(%rax) + + movq 8*2(%rdx), %r10 + sbbq %r10, 8*2(%rax) + + movq 8*3(%rdx), %r10 + sbbq %r10, 8*3(%rax) + + movq 8*4(%rdx), %r10 + sbbq %r10, 8*4(%rax) + + movq 8*5(%rdx), %r10 + sbbq %r10, 8*5(%rax) + + movq 8*6(%rdx), %r10 + sbbq %r10, 8*6(%rax) + + movq 8*7(%rdx), %r10 + sbbq %r10, 8*7(%rax) + + sbbq %r10,%r10 # carry + negq %r10 # carry + + cmpl %ecx, %r8d # j, blocks_of_8 + jne .L22 #, +.L17: + cmpl %r8d, %ebx # blocks_of_8, x_size + je .L19 #, + leal (%r8,%rbx), %r9d #, ivtmp.33 + movl %r8d, %esi # blocks_of_8, j + .p2align 4,,10 + .p2align 3 +.L20: + mov %r9d, %eax # ivtmp.33, ivtmp.33 + mov %esi, %ecx # j, j + leaq (%r14, %rax,8), %rax #, D.2387 + incl %esi # j + movq (%rax), %rdx #* D.2387, tmp153 + incl %r9d # ivtmp.33 + + rorq %r10 # carry + sbbq (%r12,%rcx,8),%rdx #* x, x + sbbq %r10,%r10 # carry + negq %r10 # carry + + cmpl %esi, %ebx # j, x_size + movq %rdx, (%rax) # x,* D.2387 + jne .L20 #, +.L19: + testq %r10, %r10 # carry + je .L_EXIT #, + decq (%rdi) #* D.2349 +.L_EXIT: + popq %rbx # + popq %rbp # + popq %r12 # + popq %r13 # + popq %r14 # + popq %r15 # +END_FUNCTION(bigint_monty_redc) + + +#if 0 + #define Z_ARR ARG_1 // rdi +#define Z_SIZE ARG_2_32 // esi +// X_ARR is ARG_3 == rdx, moved b/c needed for multiply +#define X_SIZE ARG_4_32 // ecx +#define U ARG_5 // r8 + +/* + We need all arguments for a while (we can reuse U eventually) + So only temp registers are + TEMP_1 %r10 + TEMP_2 %r11 + TEMP_3 = ARG_6 = %r9 + void return, so also + R0 %rax (aka TEMP_9) + is free (but needed for multiply) + + Can push: + %rbx (base pointer, callee saved) + %rpb (frame pointer, callee saved) + %r12-%r15 (callee saved) + + Can push base/frame pointers since this is a leaf function + and does not reference any data. +*/ + + push %r12 + push %r13 + push %r14 + push %r15 + +#define LOOP_CTR_I %r12 +#define LOOP_CTR_J %r13 + +#define CARRY TEMP_1 +#define Z_WORD TEMP_2 +#define X_ARR TEMP_3 +#define MUL_LO %rax +#define MUL_HI %rdx + + ASSIGN(X_ARR, ARG_3) + + /* + ZEROIZE(CARRY) + + ASSIGN(LOOP_CTR, X_SIZE) + + JUMP_IF_ZERO(LOOP_CTR, .L_MULADD_DONE) + JUMP_IF_LT(LOOP_CTR, 8, .LOOP_MULADD1) + +#define MULADD_OP(N) \ + ASSIGN(MUL_LO, ARRAY8(X_ARR, N)) ; \ + ASSIGN(Z_WORD, ARRAY8(Z_ARR, N)) ; \ + MUL(Y) ; \ + ADD(Z_WORD, CARRY) ; \ + ASSIGN(CARRY, MUL_HI) ; \ + ADD_LAST_CARRY(CARRY) ; \ + ADD(Z_WORD, MUL_LO) ; \ + ADD_LAST_CARRY(CARRY) ; \ + ASSIGN(ARRAY8(Z_ARR, N), Z_WORD) + +ALIGN +.LOOP_MULADD8: + MULADD_OP(0) + MULADD_OP(1) + MULADD_OP(2) + MULADD_OP(3) + MULADD_OP(4) + MULADD_OP(5) + MULADD_OP(6) + MULADD_OP(7) + + SUB_IMM(LOOP_CTR, 8) + ADD_IMM(Z_ARR, 64) + ADD_IMM(X_ARR, 64) + cmp IMM(8), LOOP_CTR + jge .LOOP_MULADD8 + + JUMP_IF_ZERO(LOOP_CTR, .L_MULADD_DONE) + +ALIGN +.LOOP_MULADD1: + MULADD_OP(0) + + SUB_IMM(LOOP_CTR, 1) + ADD_IMM(Z_ARR, 8) + ADD_IMM(X_ARR, 8) + + cmp IMM(0), LOOP_CTR + jne .LOOP_MULADD1 +*/ + + pop %r15 + pop %r14 + pop %r13 + pop %r12 +#endif |