diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-07 18:45:59 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-07 18:45:59 +0000 |
commit | 19fcec84d8d7d21e796c7624e521b60d28ee21ed (patch) | |
tree | 42d26aa27d1e3f7c0b8bd3fd14e7d7082f5008dc /src/crypto/isa-l/isa-l_crypto/md5_mb/md5_mb_mgr_flush_avx512.asm | |
parent | Initial commit. (diff) | |
download | ceph-upstream.tar.xz ceph-upstream.zip |
Adding upstream version 16.2.11+ds.upstream/16.2.11+dsupstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'src/crypto/isa-l/isa-l_crypto/md5_mb/md5_mb_mgr_flush_avx512.asm')
-rw-r--r-- | src/crypto/isa-l/isa-l_crypto/md5_mb/md5_mb_mgr_flush_avx512.asm | 313 |
1 files changed, 313 insertions, 0 deletions
diff --git a/src/crypto/isa-l/isa-l_crypto/md5_mb/md5_mb_mgr_flush_avx512.asm b/src/crypto/isa-l/isa-l_crypto/md5_mb/md5_mb_mgr_flush_avx512.asm new file mode 100644 index 000000000..e8d4ca03c --- /dev/null +++ b/src/crypto/isa-l/isa-l_crypto/md5_mb/md5_mb_mgr_flush_avx512.asm @@ -0,0 +1,313 @@ +;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; +; Copyright(c) 2011-2016 Intel Corporation All rights reserved. +; +; Redistribution and use in source and binary forms, with or without +; modification, are permitted provided that the following conditions +; are met: +; * Redistributions of source code must retain the above copyright +; notice, this list of conditions and the following disclaimer. +; * Redistributions in binary form must reproduce the above copyright +; notice, this list of conditions and the following disclaimer in +; the documentation and/or other materials provided with the +; distribution. +; * Neither the name of Intel Corporation nor the names of its +; contributors may be used to endorse or promote products derived +; from this software without specific prior written permission. +; +; THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +; "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +; LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +; A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +; OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +; SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +; LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +; DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +; THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +; (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +; OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; + +%include "md5_job.asm" +%include "md5_mb_mgr_datastruct.asm" +%include "reg_sizes.asm" + +%ifdef HAVE_AS_KNOWS_AVX512 +extern md5_mb_x16x2_avx512 +default rel + +%if 1 +%ifidn __OUTPUT_FORMAT__, elf64 +;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; +; UN*X register definitions +%define arg1 rdi ; rcx +%define arg2 rsi ; rdx + +;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; + +%else + +;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; +; WINDOWS register definitions +%define arg1 rcx +%define arg2 rdx + +;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; +%endif + +; Common register definitions + +%define state arg1 +%define len2 arg2 + +; idx must be a register not clobberred by md5_mb_x16_avx512 +%define idx rbp + +%define unused_lanes ymm7 +%define lane r9 + +%define lane_data r10 + +%define job_rax rax +%define tmp rax + +%define num_lanes_inuse r8 + +%endif ;; if 1 + +; STACK_SPACE needs to be an odd multiple of 8 +_XMM_SAVE_SIZE equ 10*16 +_GPR_SAVE_SIZE equ 8*8 +_ALIGN_SIZE equ 8 + +_XMM_SAVE equ 0 +_GPR_SAVE equ _XMM_SAVE + _XMM_SAVE_SIZE +STACK_SPACE equ _GPR_SAVE + _GPR_SAVE_SIZE + _ALIGN_SIZE + +%define APPEND(a,b) a %+ b + +;; Byte shift in MEM addr, read a extra byte [addr+16] +%macro MEM_VPSRLDDQ 2 +%define %%addr %1 +%define %%TMP_YMM %2 + vmovdqu %%TMP_YMM, [%%addr + 1] + vmovdqu [%%addr], %%TMP_YMM + mov [%%addr + 31], byte 0 +%endmacro + +;; Byte shift in MEM addr, read a extra byte [addr-1] +%macro MEM_VPSLLDDQ 2 +%define %%addr %1 +%define %%TMP_YMM %2 + vmovdqu %%TMP_YMM, [%%addr-1] + vmovdqu [%%addr], %%TMP_YMM + mov [%%addr], byte 0 +%endmacro + +align 64 +default rel +section .text + +; JOB* md5_mb_mgr_flush_avx512(MB_MGR_HMAC_OOO *state) +; arg 1 : rcx : state +global md5_mb_mgr_flush_avx512:function +md5_mb_mgr_flush_avx512: + sub rsp, STACK_SPACE + mov [rsp + _GPR_SAVE + 8*0], rbx + mov [rsp + _GPR_SAVE + 8*3], rbp + mov [rsp + _GPR_SAVE + 8*4], r12 + mov [rsp + _GPR_SAVE + 8*5], r13 + mov [rsp + _GPR_SAVE + 8*6], r14 + mov [rsp + _GPR_SAVE + 8*7], r15 +%ifidn __OUTPUT_FORMAT__, win64 + mov [rsp + _GPR_SAVE + 8*1], rsi + mov [rsp + _GPR_SAVE + 8*2], rdi + vmovdqa [rsp + _XMM_SAVE + 16*0], xmm6 + vmovdqa [rsp + _XMM_SAVE + 16*1], xmm7 + vmovdqa [rsp + _XMM_SAVE + 16*2], xmm8 + vmovdqa [rsp + _XMM_SAVE + 16*3], xmm9 + vmovdqa [rsp + _XMM_SAVE + 16*4], xmm10 + vmovdqa [rsp + _XMM_SAVE + 16*5], xmm11 + vmovdqa [rsp + _XMM_SAVE + 16*6], xmm12 + vmovdqa [rsp + _XMM_SAVE + 16*7], xmm13 + vmovdqa [rsp + _XMM_SAVE + 16*8], xmm14 + vmovdqa [rsp + _XMM_SAVE + 16*9], xmm15 +%endif + + mov DWORD(num_lanes_inuse), [state + _num_lanes_inuse] + cmp num_lanes_inuse, 0 + jz return_null + + ; find a lane with a non-null job + xor idx, idx +%assign I 1 +%rep 31 + cmp qword [state + _ldata + I * _LANE_DATA_size + _job_in_lane], 0 + cmovne idx, [APPEND(lane_,I)] +%assign I (I+1) +%endrep + + ; copy idx to empty lanes +copy_lane_data: + mov tmp, [state + _args + _data_ptr + 8*idx] + +%assign I 0 +%rep 32 + cmp qword [state + _ldata + I * _LANE_DATA_size + _job_in_lane], 0 + jne APPEND(skip_,I) + mov [state + _args + _data_ptr + 8*I], tmp + mov dword [state + _lens + 4*I], 0xFFFFFFFF +APPEND(skip_,I): +%assign I (I+1) +%endrep + + ; Find min length + vmovdqu ymm0, [state + _lens + 0*32] + vmovdqu ymm1, [state + _lens + 1*32] + + vpminud ymm2, ymm0, ymm1 ; ymm2 has {D,C,B,A} + vpalignr ymm3, ymm3, ymm2, 8 ; ymm3 has {x,x,D,C} + vpminud ymm2, ymm2, ymm3 ; ymm2 has {x,x,E,F} + vpalignr ymm3, ymm3, ymm2, 4 ; ymm3 has {x,x,x,E} + vpminud ymm2, ymm2, ymm3 ; ymm2 has min value in low dword + vperm2i128 ymm3, ymm2, ymm2, 1 ; ymm3 has halves of ymm2 reversed + vpminud ymm2, ymm2, ymm3 ; ymm2 has min value in low dword + ; Find min length + vmovdqu ymm5, [state + _lens + 2*32] + vmovdqu ymm6, [state + _lens + 3*32] + + vpminud ymm4, ymm5, ymm6 ; ymm4 has {D,C,B,A} + vpalignr ymm3, ymm3, ymm4, 8 ; ymm3 has {x,x,D,C} + vpminud ymm4, ymm4, ymm3 ; ymm4 has {x,x,E,F} + vpalignr ymm3, ymm3, ymm4, 4 ; ymm3 has {x,x,x,E} + vpminud ymm4, ymm4, ymm3 ; ymm4 has min value in low dword + vperm2i128 ymm3, ymm4, ymm4, 1 ; ymm3 has halves of ymm4 reversed + vpminud ymm4, ymm4, ymm3 ; ymm4 has min value in low dword + + vpminud ymm2, ymm2, ymm4 ; ymm2 has min value in low dword + vmovd DWORD(idx), xmm2 + mov len2, idx + and idx, 0x3F + shr len2, 6 + jz len_is_0 + + vpand ymm2, ymm2, [rel clear_low_6bits] + vpshufd ymm2, ymm2, 0 + + vpsubd ymm0, ymm0, ymm2 + vpsubd ymm1, ymm1, ymm2 + vpsubd ymm5, ymm5, ymm2 + vpsubd ymm6, ymm6, ymm2 + + vmovdqu [state + _lens + 0*32], ymm0 + vmovdqu [state + _lens + 1*32], ymm1 + vmovdqu [state + _lens + 2*32], ymm5 + vmovdqu [state + _lens + 3*32], ymm6 + + ; "state" and "args" are the same address, arg1 + ; len is arg2 + call md5_mb_x16x2_avx512 + ; state and idx are intact + +len_is_0: + ; process completed job "idx" + imul lane_data, idx, _LANE_DATA_size + lea lane_data, [state + _ldata + lane_data] + + mov job_rax, [lane_data + _job_in_lane] + mov lane, [state + _unused_lanes] + mov qword [lane_data + _job_in_lane], 0 + mov dword [job_rax + _status], STS_COMPLETED + + shl lane, 8 + or lane, idx + MEM_VPSLLDDQ (state + _unused_lanes), unused_lanes + mov [state + _unused_lanes], lane + + mov DWORD(num_lanes_inuse), [state + _num_lanes_inuse] + sub num_lanes_inuse, 1 + mov [state + _num_lanes_inuse], DWORD(num_lanes_inuse) + + mov dword [state + _lens + 4*idx], 0xFFFFFFFF + + vmovd xmm0, [state + _args_digest + 4*idx + 0*4*16*2] + vpinsrd xmm0, [state + _args_digest + 4*idx + 1*4*16*2], 1 + vpinsrd xmm0, [state + _args_digest + 4*idx + 2*4*16*2], 2 + vpinsrd xmm0, [state + _args_digest + 4*idx + 3*4*16*2], 3 + + vmovdqa [job_rax + _result_digest + 0*16], xmm0 + +return: + +%ifidn __OUTPUT_FORMAT__, win64 + vmovdqa xmm6, [rsp + _XMM_SAVE + 16*0] + vmovdqa xmm7, [rsp + _XMM_SAVE + 16*1] + vmovdqa xmm8, [rsp + _XMM_SAVE + 16*2] + vmovdqa xmm9, [rsp + _XMM_SAVE + 16*3] + vmovdqa xmm10, [rsp + _XMM_SAVE + 16*4] + vmovdqa xmm11, [rsp + _XMM_SAVE + 16*5] + vmovdqa xmm12, [rsp + _XMM_SAVE + 16*6] + vmovdqa xmm13, [rsp + _XMM_SAVE + 16*7] + vmovdqa xmm14, [rsp + _XMM_SAVE + 16*8] + vmovdqa xmm15, [rsp + _XMM_SAVE + 16*9] + mov rsi, [rsp + _GPR_SAVE + 8*1] + mov rdi, [rsp + _GPR_SAVE + 8*2] +%endif + mov rbx, [rsp + _GPR_SAVE + 8*0] + mov rbp, [rsp + _GPR_SAVE + 8*3] + mov r12, [rsp + _GPR_SAVE + 8*4] + mov r13, [rsp + _GPR_SAVE + 8*5] + mov r14, [rsp + _GPR_SAVE + 8*6] + mov r15, [rsp + _GPR_SAVE + 8*7] + add rsp, STACK_SPACE + + ret + +return_null: + xor job_rax, job_rax + jmp return + + +section .data align=16 + +align 16 +clear_low_6bits: + dq 0x00000000FFFFFFC0, 0x0000000000000000 + dq 0x00000000FFFFFFC0, 0x0000000000000000 +lane_1: dq 1 +lane_2: dq 2 +lane_3: dq 3 +lane_4: dq 4 +lane_5: dq 5 +lane_6: dq 6 +lane_7: dq 7 +lane_8: dq 8 +lane_9: dq 9 +lane_10: dq 10 +lane_11: dq 11 +lane_12: dq 12 +lane_13: dq 13 +lane_14: dq 14 +lane_15: dq 15 +lane_16: dq 16 +lane_17: dq 17 +lane_18: dq 18 +lane_19: dq 19 +lane_20: dq 20 +lane_21: dq 21 +lane_22: dq 22 +lane_23: dq 23 +lane_24: dq 24 +lane_25: dq 25 +lane_26: dq 26 +lane_27: dq 27 +lane_28: dq 28 +lane_29: dq 29 +lane_30: dq 30 +lane_31: dq 31 + +%else +%ifidn __OUTPUT_FORMAT__, win64 +global no_md5_mb_mgr_flush_avx512 +no_md5_mb_mgr_flush_avx512: +%endif +%endif ; HAVE_AS_KNOWS_AVX512 |