[pcsx_rearmed.git] / deps / libchdr / deps / lzma-22.01 / src / Asm / x86 / Sha256Opt.asm

; Sha256Opt.asm -- SHA-256 optimized code for SHA-256 x86 hardware instructions\r
; 2022-04-17 : Igor Pavlov : Public domain\r
\r
include 7zAsm.asm\r
\r
MY_ASM_START\r
\r
; .data\r
; public K\r
\r
; we can use external SHA256_K_ARRAY defined in Sha256.c\r
; but we must guarantee that SHA256_K_ARRAY is aligned for 16-bytes\r
\r
COMMENT @\r
ifdef x64\r
K_CONST equ SHA256_K_ARRAY\r
else\r
K_CONST equ _SHA256_K_ARRAY\r
endif\r
EXTRN   K_CONST:xmmword\r
@\r
\r
CONST   SEGMENT\r
\r
align 16\r
Reverse_Endian_Mask db 3,2,1,0, 7,6,5,4, 11,10,9,8, 15,14,13,12\r
\r
; COMMENT @\r
align 16\r
K_CONST \\r
DD 0428a2f98H, 071374491H, 0b5c0fbcfH, 0e9b5dba5H\r
DD 03956c25bH, 059f111f1H, 0923f82a4H, 0ab1c5ed5H\r
DD 0d807aa98H, 012835b01H, 0243185beH, 0550c7dc3H\r
DD 072be5d74H, 080deb1feH, 09bdc06a7H, 0c19bf174H\r
DD 0e49b69c1H, 0efbe4786H, 00fc19dc6H, 0240ca1ccH\r
DD 02de92c6fH, 04a7484aaH, 05cb0a9dcH, 076f988daH\r
DD 0983e5152H, 0a831c66dH, 0b00327c8H, 0bf597fc7H\r
DD 0c6e00bf3H, 0d5a79147H, 006ca6351H, 014292967H\r
DD 027b70a85H, 02e1b2138H, 04d2c6dfcH, 053380d13H\r
DD 0650a7354H, 0766a0abbH, 081c2c92eH, 092722c85H\r
DD 0a2bfe8a1H, 0a81a664bH, 0c24b8b70H, 0c76c51a3H\r
DD 0d192e819H, 0d6990624H, 0f40e3585H, 0106aa070H\r
DD 019a4c116H, 01e376c08H, 02748774cH, 034b0bcb5H\r
DD 0391c0cb3H, 04ed8aa4aH, 05b9cca4fH, 0682e6ff3H\r
DD 0748f82eeH, 078a5636fH, 084c87814H, 08cc70208H\r
DD 090befffaH, 0a4506cebH, 0bef9a3f7H, 0c67178f2H\r
; @\r
\r
CONST   ENDS\r
\r
; _TEXT$SHA256OPT SEGMENT 'CODE'\r
\r
ifndef x64\r
    .686\r
    .xmm\r
endif\r
        \r
; jwasm-based assemblers for linux and linker from new versions of binutils\r
; can generate incorrect code for load [ARRAY + offset] instructions.\r
; 22.00: we load K_CONST offset to (rTable) register to avoid jwasm+binutils problem \r
        rTable  equ r0\r
        ; rTable  equ K_CONST\r
        \r
ifdef x64\r
        rNum    equ REG_ABI_PARAM_2\r
    if (IS_LINUX eq 0)\r
        LOCAL_SIZE equ (16 * 2)\r
    endif\r
else\r
        rNum    equ r3\r
        LOCAL_SIZE equ (16 * 1)\r
endif\r
\r
rState equ REG_ABI_PARAM_0\r
rData  equ REG_ABI_PARAM_1\r
\r
\r
\r
\r
\r
\r
MY_SHA_INSTR macro cmd, a1, a2\r
        db 0fH, 038H, cmd, (0c0H + a1 * 8 + a2)\r
endm\r
\r
cmd_sha256rnds2 equ 0cbH\r
cmd_sha256msg1  equ 0ccH\r
cmd_sha256msg2  equ 0cdH\r
\r
MY_sha256rnds2 macro a1, a2\r
        MY_SHA_INSTR  cmd_sha256rnds2, a1, a2\r
endm\r
\r
MY_sha256msg1 macro a1, a2\r
        MY_SHA_INSTR  cmd_sha256msg1, a1, a2\r
endm\r
\r
MY_sha256msg2 macro a1, a2\r
        MY_SHA_INSTR  cmd_sha256msg2, a1, a2\r
endm\r
\r
MY_PROLOG macro\r
    ifdef x64\r
      if (IS_LINUX eq 0)\r
        movdqa  [r4 + 8], xmm6\r
        movdqa  [r4 + 8 + 16], xmm7\r
        sub     r4, LOCAL_SIZE + 8\r
        movdqa  [r4     ], xmm8\r
        movdqa  [r4 + 16], xmm9\r
      endif\r
    else ; x86\r
        push    r3\r
        push    r5\r
        mov     r5, r4\r
        NUM_PUSH_REGS   equ 2\r
        PARAM_OFFSET    equ (REG_SIZE * (1 + NUM_PUSH_REGS))\r
      if (IS_CDECL gt 0)\r
        mov     rState, [r4 + PARAM_OFFSET]\r
        mov     rData,  [r4 + PARAM_OFFSET + REG_SIZE * 1]\r
        mov     rNum,   [r4 + PARAM_OFFSET + REG_SIZE * 2]\r
      else ; fastcall\r
        mov     rNum,   [r4 + PARAM_OFFSET]\r
      endif\r
        and     r4, -16\r
        sub     r4, LOCAL_SIZE\r
    endif\r
endm\r
\r
MY_EPILOG macro\r
    ifdef x64\r
      if (IS_LINUX eq 0)\r
        movdqa  xmm8, [r4]\r
        movdqa  xmm9, [r4 + 16]\r
        add     r4, LOCAL_SIZE + 8\r
        movdqa  xmm6, [r4 + 8]\r
        movdqa  xmm7, [r4 + 8 + 16]\r
      endif\r
    else ; x86\r
        mov     r4, r5\r
        pop     r5\r
        pop     r3\r
    endif\r
    MY_ENDP\r
endm\r
\r
\r
msg        equ xmm0\r
tmp        equ xmm0\r
state0_N   equ 2\r
state1_N   equ 3\r
w_regs     equ 4\r
\r
\r
state1_save equ xmm1\r
state0  equ @CatStr(xmm, %state0_N)\r
state1  equ @CatStr(xmm, %state1_N)\r
\r
\r
ifdef x64\r
        state0_save  equ  xmm8\r
        mask2        equ  xmm9\r
else\r
        state0_save  equ  [r4]\r
        mask2        equ  xmm0\r
endif\r
\r
LOAD_MASK macro\r
        movdqa  mask2, XMMWORD PTR Reverse_Endian_Mask\r
endm\r
\r
LOAD_W macro k:req\r
        movdqu  @CatStr(xmm, %(w_regs + k)), [rData + (16 * (k))]\r
        pshufb  @CatStr(xmm, %(w_regs + k)), mask2\r
endm\r
\r
\r
; pre1 <= 4 && pre2 >= 1 && pre1 > pre2 && (pre1 - pre2) <= 1\r
pre1 equ 3\r
pre2 equ 2\r
   \r
\r
\r
RND4 macro k\r
        movdqa  msg, xmmword ptr [rTable + (k) * 16]\r
        paddd   msg, @CatStr(xmm, %(w_regs + ((k + 0) mod 4)))\r
        MY_sha256rnds2 state0_N, state1_N\r
        pshufd   msg, msg, 0eH\r
        \r
    if (k GE (4 - pre1)) AND (k LT (16 - pre1))\r
        ; w4[0] = msg1(w4[-4], w4[-3])\r
        MY_sha256msg1 (w_regs + ((k + pre1) mod 4)), (w_regs + ((k + pre1 - 3) mod 4))\r
    endif\r
        \r
        MY_sha256rnds2 state1_N, state0_N\r
\r
    if (k GE (4 - pre2)) AND (k LT (16 - pre2))\r
        movdqa  tmp, @CatStr(xmm, %(w_regs + ((k + pre2 - 1) mod 4)))\r
        palignr tmp, @CatStr(xmm, %(w_regs + ((k + pre2 - 2) mod 4))), 4\r
        paddd   @CatStr(xmm, %(w_regs + ((k + pre2) mod 4))), tmp\r
        ; w4[0] = msg2(w4[0], w4[-1])\r
        MY_sha256msg2 %(w_regs + ((k + pre2) mod 4)), %(w_regs + ((k + pre2 - 1) mod 4))\r
    endif\r
endm\r
\r
\r
\r
\r
\r
REVERSE_STATE macro\r
                               ; state0 ; dcba\r
                               ; state1 ; hgfe\r
        pshufd      tmp, state0, 01bH   ; abcd\r
        pshufd   state0, state1, 01bH   ; efgh\r
        movdqa   state1, state0         ; efgh\r
        punpcklqdq  state0, tmp         ; cdgh\r
        punpckhqdq  state1, tmp         ; abef\r
endm\r
\r
\r
MY_PROC Sha256_UpdateBlocks_HW, 3\r
    MY_PROLOG\r
\r
        lea     rTable, [K_CONST]\r
\r
        cmp     rNum, 0\r
        je      end_c\r
\r
        movdqu   state0, [rState]       ; dcba\r
        movdqu   state1, [rState + 16]  ; hgfe\r
\r
        REVERSE_STATE\r
       \r
        ifdef x64\r
        LOAD_MASK\r
        endif\r
\r
    align 16\r
    nextBlock:\r
        movdqa  state0_save, state0\r
        movdqa  state1_save, state1\r
        \r
        ifndef x64\r
        LOAD_MASK\r
        endif\r
        \r
        LOAD_W 0\r
        LOAD_W 1\r
        LOAD_W 2\r
        LOAD_W 3\r
\r
        \r
        k = 0\r
        rept 16\r
          RND4 k\r
          k = k + 1\r
        endm\r
\r
        paddd   state0, state0_save\r
        paddd   state1, state1_save\r
\r
        add     rData, 64\r
        sub     rNum, 1\r
        jnz     nextBlock\r
        \r
        REVERSE_STATE\r
\r
        movdqu  [rState], state0\r
        movdqu  [rState + 16], state1\r
       \r
  end_c:\r
MY_EPILOG\r
\r
; _TEXT$SHA256OPT ENDS\r
\r
end\r
Commit	Line	Data
9e052883	1	; Sha256Opt.asm -- SHA-256 optimized code for SHA-256 x86 hardware instructions\r
	2	; 2022-04-17 : Igor Pavlov : Public domain\r
	3	\r
	4	include 7zAsm.asm\r
	5	\r
	6	MY_ASM_START\r
	7	\r
	8	; .data\r
	9	; public K\r
	10	\r
	11	; we can use external SHA256_K_ARRAY defined in Sha256.c\r
	12	; but we must guarantee that SHA256_K_ARRAY is aligned for 16-bytes\r
	13	\r
	14	COMMENT @\r
	15	ifdef x64\r
	16	K_CONST equ SHA256_K_ARRAY\r
	17	else\r
	18	K_CONST equ _SHA256_K_ARRAY\r
	19	endif\r
	20	EXTRN K_CONST:xmmword\r
	21	@\r
	22	\r
	23	CONST SEGMENT\r
	24	\r
	25	align 16\r
	26	Reverse_Endian_Mask db 3,2,1,0, 7,6,5,4, 11,10,9,8, 15,14,13,12\r
	27	\r
	28	; COMMENT @\r
	29	align 16\r
	30	K_CONST \\r
	31	DD 0428a2f98H, 071374491H, 0b5c0fbcfH, 0e9b5dba5H\r
	32	DD 03956c25bH, 059f111f1H, 0923f82a4H, 0ab1c5ed5H\r
	33	DD 0d807aa98H, 012835b01H, 0243185beH, 0550c7dc3H\r
	34	DD 072be5d74H, 080deb1feH, 09bdc06a7H, 0c19bf174H\r
	35	DD 0e49b69c1H, 0efbe4786H, 00fc19dc6H, 0240ca1ccH\r
	36	DD 02de92c6fH, 04a7484aaH, 05cb0a9dcH, 076f988daH\r
	37	DD 0983e5152H, 0a831c66dH, 0b00327c8H, 0bf597fc7H\r
	38	DD 0c6e00bf3H, 0d5a79147H, 006ca6351H, 014292967H\r
	39	DD 027b70a85H, 02e1b2138H, 04d2c6dfcH, 053380d13H\r
	40	DD 0650a7354H, 0766a0abbH, 081c2c92eH, 092722c85H\r
	41	DD 0a2bfe8a1H, 0a81a664bH, 0c24b8b70H, 0c76c51a3H\r
	42	DD 0d192e819H, 0d6990624H, 0f40e3585H, 0106aa070H\r
	43	DD 019a4c116H, 01e376c08H, 02748774cH, 034b0bcb5H\r
	44	DD 0391c0cb3H, 04ed8aa4aH, 05b9cca4fH, 0682e6ff3H\r
	45	DD 0748f82eeH, 078a5636fH, 084c87814H, 08cc70208H\r
	46	DD 090befffaH, 0a4506cebH, 0bef9a3f7H, 0c67178f2H\r
	47	; @\r
	48	\r
	49	CONST ENDS\r
	50	\r
	51	; _TEXT$SHA256OPT SEGMENT 'CODE'\r
	52	\r
	53	ifndef x64\r
	54	.686\r
	55	.xmm\r
	56	endif\r
	57	\r
	58	; jwasm-based assemblers for linux and linker from new versions of binutils\r
	59	; can generate incorrect code for load [ARRAY + offset] instructions.\r
	60	; 22.00: we load K_CONST offset to (rTable) register to avoid jwasm+binutils problem \r
	61	rTable equ r0\r
	62	; rTable equ K_CONST\r
	63	\r
	64	ifdef x64\r
65	rNum equ REG_ABI_PARAM_2\r
66	if (IS_LINUX eq 0)\r
67	LOCAL_SIZE equ (16 * 2)\r
68	endif\r
69	else\r
70	rNum equ r3\r
71	LOCAL_SIZE equ (16 * 1)\r
72	endif\r
73	\r
74	rState equ REG_ABI_PARAM_0\r
75	rData equ REG_ABI_PARAM_1\r
76	\r
77	\r
78	\r
79	\r
80	\r
81	\r
82	MY_SHA_INSTR macro cmd, a1, a2\r
83	db 0fH, 038H, cmd, (0c0H + a1 * 8 + a2)\r
84	endm\r
85	\r
86	cmd_sha256rnds2 equ 0cbH\r
87	cmd_sha256msg1 equ 0ccH\r
88	cmd_sha256msg2 equ 0cdH\r
89	\r
90	MY_sha256rnds2 macro a1, a2\r
91	MY_SHA_INSTR cmd_sha256rnds2, a1, a2\r
92	endm\r
93	\r
94	MY_sha256msg1 macro a1, a2\r
95	MY_SHA_INSTR cmd_sha256msg1, a1, a2\r
96	endm\r
97	\r
98	MY_sha256msg2 macro a1, a2\r
99	MY_SHA_INSTR cmd_sha256msg2, a1, a2\r
100	endm\r
101	\r
102	MY_PROLOG macro\r
103	ifdef x64\r
104	if (IS_LINUX eq 0)\r
105	movdqa [r4 + 8], xmm6\r
106	movdqa [r4 + 8 + 16], xmm7\r
107	sub r4, LOCAL_SIZE + 8\r
108	movdqa [r4 ], xmm8\r
109	movdqa [r4 + 16], xmm9\r
110	endif\r
111	else ; x86\r
112	push r3\r
113	push r5\r
114	mov r5, r4\r
115	NUM_PUSH_REGS equ 2\r
116	PARAM_OFFSET equ (REG_SIZE * (1 + NUM_PUSH_REGS))\r
117	if (IS_CDECL gt 0)\r
118	mov rState, [r4 + PARAM_OFFSET]\r
119	mov rData, [r4 + PARAM_OFFSET + REG_SIZE * 1]\r
120	mov rNum, [r4 + PARAM_OFFSET + REG_SIZE * 2]\r
121	else ; fastcall\r
122	mov rNum, [r4 + PARAM_OFFSET]\r
123	endif\r
124	and r4, -16\r
125	sub r4, LOCAL_SIZE\r
126	endif\r
127	endm\r
128	\r
129	MY_EPILOG macro\r
130	ifdef x64\r
131	if (IS_LINUX eq 0)\r
132	movdqa xmm8, [r4]\r
133	movdqa xmm9, [r4 + 16]\r
134	add r4, LOCAL_SIZE + 8\r
135	movdqa xmm6, [r4 + 8]\r
136	movdqa xmm7, [r4 + 8 + 16]\r
137	endif\r
138	else ; x86\r
139	mov r4, r5\r
140	pop r5\r
141	pop r3\r
142	endif\r
143	MY_ENDP\r
144	endm\r
145	\r
146	\r
147	msg equ xmm0\r
148	tmp equ xmm0\r
149	state0_N equ 2\r
150	state1_N equ 3\r
151	w_regs equ 4\r
152	\r
153	\r
154	state1_save equ xmm1\r
155	state0 equ @CatStr(xmm, %state0_N)\r
156	state1 equ @CatStr(xmm, %state1_N)\r
157	\r
158	\r
159	ifdef x64\r
160	state0_save equ xmm8\r
161	mask2 equ xmm9\r
162	else\r
163	state0_save equ [r4]\r
164	mask2 equ xmm0\r
165	endif\r
166	\r
167	LOAD_MASK macro\r
168	movdqa mask2, XMMWORD PTR Reverse_Endian_Mask\r
169	endm\r
170	\r
171	LOAD_W macro k:req\r
172	movdqu @CatStr(xmm, %(w_regs + k)), [rData + (16 * (k))]\r
173	pshufb @CatStr(xmm, %(w_regs + k)), mask2\r
174	endm\r
175	\r
176	\r
177	; pre1 <= 4 && pre2 >= 1 && pre1 > pre2 && (pre1 - pre2) <= 1\r
178	pre1 equ 3\r
179	pre2 equ 2\r
180	\r
181	\r
182	\r
183	RND4 macro k\r
184	movdqa msg, xmmword ptr [rTable + (k) * 16]\r
185	paddd msg, @CatStr(xmm, %(w_regs + ((k + 0) mod 4)))\r
186	MY_sha256rnds2 state0_N, state1_N\r
187	pshufd msg, msg, 0eH\r
188	\r
189	if (k GE (4 - pre1)) AND (k LT (16 - pre1))\r
190	; w4[0] = msg1(w4[-4], w4[-3])\r
191	MY_sha256msg1 (w_regs + ((k + pre1) mod 4)), (w_regs + ((k + pre1 - 3) mod 4))\r
192	endif\r
193	\r
194	MY_sha256rnds2 state1_N, state0_N\r
195	\r
196	if (k GE (4 - pre2)) AND (k LT (16 - pre2))\r
197	movdqa tmp, @CatStr(xmm, %(w_regs + ((k + pre2 - 1) mod 4)))\r
198	palignr tmp, @CatStr(xmm, %(w_regs + ((k + pre2 - 2) mod 4))), 4\r
199	paddd @CatStr(xmm, %(w_regs + ((k + pre2) mod 4))), tmp\r
200	; w4[0] = msg2(w4[0], w4[-1])\r
201	MY_sha256msg2 %(w_regs + ((k + pre2) mod 4)), %(w_regs + ((k + pre2 - 1) mod 4))\r
202	endif\r
203	endm\r
204	\r
205	\r
206	\r
207	\r
208	\r
209	REVERSE_STATE macro\r
210	; state0 ; dcba\r
211	; state1 ; hgfe\r
212	pshufd tmp, state0, 01bH ; abcd\r
213	pshufd state0, state1, 01bH ; efgh\r
214	movdqa state1, state0 ; efgh\r
215	punpcklqdq state0, tmp ; cdgh\r
216	punpckhqdq state1, tmp ; abef\r
217	endm\r
218	\r
219	\r
220	MY_PROC Sha256_UpdateBlocks_HW, 3\r
221	MY_PROLOG\r
222	\r
223	lea rTable, [K_CONST]\r
224	\r
225	cmp rNum, 0\r
226	je end_c\r
227	\r
228	movdqu state0, [rState] ; dcba\r
229	movdqu state1, [rState + 16] ; hgfe\r
230	\r
231	REVERSE_STATE\r
232	\r
233	ifdef x64\r
234	LOAD_MASK\r
235	endif\r
236	\r
237	align 16\r
238	nextBlock:\r
239	movdqa state0_save, state0\r
240	movdqa state1_save, state1\r
241	\r
242	ifndef x64\r
243	LOAD_MASK\r
244	endif\r
245	\r
246	LOAD_W 0\r
247	LOAD_W 1\r
248	LOAD_W 2\r
249	LOAD_W 3\r
250	\r
251	\r
252	k = 0\r
253	rept 16\r
254	RND4 k\r
255	k = k + 1\r
256	endm\r
257	\r
258	paddd state0, state0_save\r
259	paddd state1, state1_save\r
260	\r
261	add rData, 64\r
262	sub rNum, 1\r
263	jnz nextBlock\r
264	\r
265	REVERSE_STATE\r
266	\r
267	movdqu [rState], state0\r
268	movdqu [rState + 16], state1\r
269	\r
270	end_c:\r
271	MY_EPILOG\r
272	\r
273	; _TEXT$SHA256OPT ENDS\r
274	\r
275	end\r