1;  vim:filetype=nasm ts=8
2
3;  libFLAC - Free Lossless Audio Codec library
4;  Copyright (C) 2001-2009  Josh Coalson
5;  Copyright (C) 2011-2013  Xiph.Org Foundation
6;
7;  Redistribution and use in source and binary forms, with or without
8;  modification, are permitted provided that the following conditions
9;  are met:
10;
11;  - Redistributions of source code must retain the above copyright
12;  notice, this list of conditions and the following disclaimer.
13;
14;  - Redistributions in binary form must reproduce the above copyright
15;  notice, this list of conditions and the following disclaimer in the
16;  documentation and/or other materials provided with the distribution.
17;
18;  - Neither the name of the Xiph.org Foundation nor the names of its
19;  contributors may be used to endorse or promote products derived from
20;  this software without specific prior written permission.
21;
22;  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23;  ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24;  LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25;  A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR
26;  CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
27;  EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
28;  PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
29;  PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
30;  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
31;  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
32;  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33
34%include "nasm.h"
35
36	data_section
37
38cglobal FLAC__fixed_compute_best_predictor_asm_ia32_mmx_cmov
39
40	code_section
41
42; **********************************************************************
43;
44; unsigned FLAC__fixed_compute_best_predictor(const FLAC__int32 *data, unsigned data_len, FLAC__float residual_bits_per_sample[FLAC__MAX_FIXED_ORDER+1])
45; {
46; 	FLAC__int32 last_error_0 = data[-1];
47; 	FLAC__int32 last_error_1 = data[-1] - data[-2];
48; 	FLAC__int32 last_error_2 = last_error_1 - (data[-2] - data[-3]);
49; 	FLAC__int32 last_error_3 = last_error_2 - (data[-2] - 2*data[-3] + data[-4]);
50; 	FLAC__int32 error, save;
51; 	FLAC__uint32 total_error_0 = 0, total_error_1 = 0, total_error_2 = 0, total_error_3 = 0, total_error_4 = 0;
52; 	unsigned i, order;
53;
54; 	for(i = 0; i < data_len; i++) {
55; 		error  = data[i]     ; total_error_0 += local_abs(error);                      save = error;
56; 		error -= last_error_0; total_error_1 += local_abs(error); last_error_0 = save; save = error;
57; 		error -= last_error_1; total_error_2 += local_abs(error); last_error_1 = save; save = error;
58; 		error -= last_error_2; total_error_3 += local_abs(error); last_error_2 = save; save = error;
59; 		error -= last_error_3; total_error_4 += local_abs(error); last_error_3 = save;
60; 	}
61;
62; 	if(total_error_0 < min(min(min(total_error_1, total_error_2), total_error_3), total_error_4))
63; 		order = 0;
64; 	else if(total_error_1 < min(min(total_error_2, total_error_3), total_error_4))
65; 		order = 1;
66; 	else if(total_error_2 < min(total_error_3, total_error_4))
67; 		order = 2;
68; 	else if(total_error_3 < total_error_4)
69; 		order = 3;
70; 	else
71; 		order = 4;
72;
73; 	residual_bits_per_sample[0] = (FLAC__float)((data_len > 0 && total_error_0 > 0) ? log(M_LN2 * (FLAC__double)total_error_0 / (FLAC__double)data_len) / M_LN2 : 0.0);
74; 	residual_bits_per_sample[1] = (FLAC__float)((data_len > 0 && total_error_1 > 0) ? log(M_LN2 * (FLAC__double)total_error_1 / (FLAC__double)data_len) / M_LN2 : 0.0);
75; 	residual_bits_per_sample[2] = (FLAC__float)((data_len > 0 && total_error_2 > 0) ? log(M_LN2 * (FLAC__double)total_error_2 / (FLAC__double)data_len) / M_LN2 : 0.0);
76; 	residual_bits_per_sample[3] = (FLAC__float)((data_len > 0 && total_error_3 > 0) ? log(M_LN2 * (FLAC__double)total_error_3 / (FLAC__double)data_len) / M_LN2 : 0.0);
77; 	residual_bits_per_sample[4] = (FLAC__float)((data_len > 0 && total_error_4 > 0) ? log(M_LN2 * (FLAC__double)total_error_4 / (FLAC__double)data_len) / M_LN2 : 0.0);
78;
79; 	return order;
80; }
81	ALIGN 16
82cident FLAC__fixed_compute_best_predictor_asm_ia32_mmx_cmov
83
84	; esp + 36 == data[]
85	; esp + 40 == data_len
86	; esp + 44 == residual_bits_per_sample[]
87
88	push	ebp
89	push	ebx
90	push	esi
91	push	edi
92	sub	esp, byte 16
93	; qword [esp] == temp space for loading FLAC__uint64s to FPU regs
94
95	; ebx == &data[i]
96	; ecx == loop counter (i)
97	; ebp == order
98	; mm0 == total_error_1:total_error_0
99	; mm1 == total_error_2:total_error_3
100	; mm2 == :total_error_4
101	; mm3 == last_error_1:last_error_0
102	; mm4 == last_error_2:last_error_3
103
104	mov	ecx, [esp + 40]			; ecx = data_len
105	test	ecx, ecx
106	jz	near .data_len_is_0
107
108	mov	ebx, [esp + 36]			; ebx = data[]
109	movd	mm3, [ebx - 4]			; mm3 = 0:last_error_0
110	movd	mm2, [ebx - 8]			; mm2 = 0:data[-2]
111	movd	mm1, [ebx - 12]			; mm1 = 0:data[-3]
112	movd	mm0, [ebx - 16]			; mm0 = 0:data[-4]
113	movq	mm5, mm3			; mm5 = 0:last_error_0
114	psubd	mm5, mm2			; mm5 = 0:last_error_1
115	punpckldq	mm3, mm5		; mm3 = last_error_1:last_error_0
116	psubd	mm2, mm1			; mm2 = 0:data[-2] - data[-3]
117	psubd	mm5, mm2			; mm5 = 0:last_error_2
118	movq	mm4, mm5			; mm4 = 0:last_error_2
119	psubd	mm4, mm2			; mm4 = 0:last_error_2 - (data[-2] - data[-3])
120	paddd	mm4, mm1			; mm4 = 0:last_error_2 - (data[-2] - 2 * data[-3])
121	psubd	mm4, mm0			; mm4 = 0:last_error_3
122	punpckldq	mm4, mm5		; mm4 = last_error_2:last_error_3
123	pxor	mm0, mm0			; mm0 = total_error_1:total_error_0
124	pxor	mm1, mm1			; mm1 = total_error_2:total_error_3
125	pxor	mm2, mm2			; mm2 = 0:total_error_4
126
127	ALIGN 16
128.loop:
129	movd	mm7, [ebx]			; mm7 = 0:error_0
130	add	ebx, byte 4
131	movq	mm6, mm7			; mm6 = 0:error_0
132	psubd	mm7, mm3			; mm7 = :error_1
133	punpckldq	mm6, mm7		; mm6 = error_1:error_0
134	movq	mm5, mm6			; mm5 = error_1:error_0
135	movq	mm7, mm6			; mm7 = error_1:error_0
136	psubd	mm5, mm3			; mm5 = error_2:
137	movq	mm3, mm6			; mm3 = error_1:error_0
138	psrad	mm6, 31
139	pxor	mm7, mm6
140	psubd	mm7, mm6			; mm7 = abs(error_1):abs(error_0)
141	paddd	mm0, mm7			; mm0 = total_error_1:total_error_0
142	movq	mm6, mm5			; mm6 = error_2:
143	psubd	mm5, mm4			; mm5 = error_3:
144	punpckhdq	mm5, mm6		; mm5 = error_2:error_3
145	movq	mm7, mm5			; mm7 = error_2:error_3
146	movq	mm6, mm5			; mm6 = error_2:error_3
147	psubd	mm5, mm4			; mm5 = :error_4
148	movq	mm4, mm6			; mm4 = error_2:error_3
149	psrad	mm6, 31
150	pxor	mm7, mm6
151	psubd	mm7, mm6			; mm7 = abs(error_2):abs(error_3)
152	paddd	mm1, mm7			; mm1 = total_error_2:total_error_3
153	movq	mm6, mm5			; mm6 = :error_4
154	psrad	mm5, 31
155	pxor	mm6, mm5
156	psubd	mm6, mm5			; mm6 = :abs(error_4)
157	paddd	mm2, mm6			; mm2 = :total_error_4
158
159	dec	ecx
160	jnz	short .loop
161
162; 	if(total_error_0 < min(min(min(total_error_1, total_error_2), total_error_3), total_error_4))
163; 		order = 0;
164; 	else if(total_error_1 < min(min(total_error_2, total_error_3), total_error_4))
165; 		order = 1;
166; 	else if(total_error_2 < min(total_error_3, total_error_4))
167; 		order = 2;
168; 	else if(total_error_3 < total_error_4)
169; 		order = 3;
170; 	else
171; 		order = 4;
172	movq	mm3, mm0			; mm3 = total_error_1:total_error_0
173	movd	edi, mm2			; edi = total_error_4
174	movd	esi, mm1			; esi = total_error_3
175	movd	eax, mm0			; eax = total_error_0
176	punpckhdq	mm1, mm1		; mm1 = total_error_2:total_error_2
177	punpckhdq	mm3, mm3		; mm3 = total_error_1:total_error_1
178	movd	edx, mm1			; edx = total_error_2
179	movd	ecx, mm3			; ecx = total_error_1
180
181	xor	ebx, ebx
182	xor	ebp, ebp
183	inc	ebx
184	cmp	ecx, eax
185	cmovb	eax, ecx			; eax = min(total_error_0, total_error_1)
186	cmovbe	ebp, ebx
187	inc	ebx
188	cmp	edx, eax
189	cmovb	eax, edx			; eax = min(total_error_0, total_error_1, total_error_2)
190	cmovbe	ebp, ebx
191	inc	ebx
192	cmp	esi, eax
193	cmovb	eax, esi			; eax = min(total_error_0, total_error_1, total_error_2, total_error_3)
194	cmovbe	ebp, ebx
195	inc	ebx
196	cmp	edi, eax
197	cmovb	eax, edi			; eax = min(total_error_0, total_error_1, total_error_2, total_error_3, total_error_4)
198	cmovbe	ebp, ebx
199	movd	ebx, mm0			; ebx = total_error_0
200	emms
201
202	; 	residual_bits_per_sample[0] = (FLAC__float)((data_len > 0 && total_error_0 > 0) ? log(M_LN2 * (FLAC__double)total_error_0 / (FLAC__double)data_len) / M_LN2 : 0.0);
203	; 	residual_bits_per_sample[1] = (FLAC__float)((data_len > 0 && total_error_1 > 0) ? log(M_LN2 * (FLAC__double)total_error_1 / (FLAC__double)data_len) / M_LN2 : 0.0);
204	; 	residual_bits_per_sample[2] = (FLAC__float)((data_len > 0 && total_error_2 > 0) ? log(M_LN2 * (FLAC__double)total_error_2 / (FLAC__double)data_len) / M_LN2 : 0.0);
205	; 	residual_bits_per_sample[3] = (FLAC__float)((data_len > 0 && total_error_3 > 0) ? log(M_LN2 * (FLAC__double)total_error_3 / (FLAC__double)data_len) / M_LN2 : 0.0);
206	; 	residual_bits_per_sample[4] = (FLAC__float)((data_len > 0 && total_error_4 > 0) ? log(M_LN2 * (FLAC__double)total_error_4 / (FLAC__double)data_len) / M_LN2 : 0.0);
207	xor	eax, eax
208	fild	dword [esp + 40]		; ST = data_len (NOTE: assumes data_len is <2gigs)
209.rbps_0:
210	test	ebx, ebx
211	jz	.total_error_0_is_0
212	fld1					; ST = 1.0 data_len
213	mov	[esp], ebx
214	mov	[esp + 4], eax			; [esp] = (FLAC__uint64)total_error_0
215	mov	ebx, [esp + 44]
216	fild	qword [esp]			; ST = total_error_0 1.0 data_len
217	fdiv	st2				; ST = total_error_0/data_len 1.0 data_len
218	fldln2					; ST = ln2 total_error_0/data_len 1.0 data_len
219	fmulp	st1				; ST = ln2*total_error_0/data_len 1.0 data_len
220	fyl2x					; ST = log2(ln2*total_error_0/data_len) data_len
221	fstp	dword [ebx]			; residual_bits_per_sample[0] = log2(ln2*total_error_0/data_len)   ST = data_len
222	jmp	short .rbps_1
223.total_error_0_is_0:
224	mov	ebx, [esp + 44]
225	mov	[ebx], eax			; residual_bits_per_sample[0] = 0.0
226.rbps_1:
227	test	ecx, ecx
228	jz	.total_error_1_is_0
229	fld1					; ST = 1.0 data_len
230	mov	[esp], ecx
231	mov	[esp + 4], eax			; [esp] = (FLAC__uint64)total_error_1
232	fild	qword [esp]			; ST = total_error_1 1.0 data_len
233	fdiv	st2				; ST = total_error_1/data_len 1.0 data_len
234	fldln2					; ST = ln2 total_error_1/data_len 1.0 data_len
235	fmulp	st1				; ST = ln2*total_error_1/data_len 1.0 data_len
236	fyl2x					; ST = log2(ln2*total_error_1/data_len) data_len
237	fstp	dword [ebx + 4]			; residual_bits_per_sample[1] = log2(ln2*total_error_1/data_len)   ST = data_len
238	jmp	short .rbps_2
239.total_error_1_is_0:
240	mov	[ebx + 4], eax			; residual_bits_per_sample[1] = 0.0
241.rbps_2:
242	test	edx, edx
243	jz	.total_error_2_is_0
244	fld1					; ST = 1.0 data_len
245	mov	[esp], edx
246	mov	[esp + 4], eax			; [esp] = (FLAC__uint64)total_error_2
247	fild	qword [esp]			; ST = total_error_2 1.0 data_len
248	fdiv	st2				; ST = total_error_2/data_len 1.0 data_len
249	fldln2					; ST = ln2 total_error_2/data_len 1.0 data_len
250	fmulp	st1				; ST = ln2*total_error_2/data_len 1.0 data_len
251	fyl2x					; ST = log2(ln2*total_error_2/data_len) data_len
252	fstp	dword [ebx + 8]			; residual_bits_per_sample[2] = log2(ln2*total_error_2/data_len)   ST = data_len
253	jmp	short .rbps_3
254.total_error_2_is_0:
255	mov	[ebx + 8], eax			; residual_bits_per_sample[2] = 0.0
256.rbps_3:
257	test	esi, esi
258	jz	.total_error_3_is_0
259	fld1					; ST = 1.0 data_len
260	mov	[esp], esi
261	mov	[esp + 4], eax			; [esp] = (FLAC__uint64)total_error_3
262	fild	qword [esp]			; ST = total_error_3 1.0 data_len
263	fdiv	st2				; ST = total_error_3/data_len 1.0 data_len
264	fldln2					; ST = ln2 total_error_3/data_len 1.0 data_len
265	fmulp	st1				; ST = ln2*total_error_3/data_len 1.0 data_len
266	fyl2x					; ST = log2(ln2*total_error_3/data_len) data_len
267	fstp	dword [ebx + 12]		; residual_bits_per_sample[3] = log2(ln2*total_error_3/data_len)   ST = data_len
268	jmp	short .rbps_4
269.total_error_3_is_0:
270	mov	[ebx + 12], eax			; residual_bits_per_sample[3] = 0.0
271.rbps_4:
272	test	edi, edi
273	jz	.total_error_4_is_0
274	fld1					; ST = 1.0 data_len
275	mov	[esp], edi
276	mov	[esp + 4], eax			; [esp] = (FLAC__uint64)total_error_4
277	fild	qword [esp]			; ST = total_error_4 1.0 data_len
278	fdiv	st2				; ST = total_error_4/data_len 1.0 data_len
279	fldln2					; ST = ln2 total_error_4/data_len 1.0 data_len
280	fmulp	st1				; ST = ln2*total_error_4/data_len 1.0 data_len
281	fyl2x					; ST = log2(ln2*total_error_4/data_len) data_len
282	fstp	dword [ebx + 16]		; residual_bits_per_sample[4] = log2(ln2*total_error_4/data_len)   ST = data_len
283	jmp	short .rbps_end
284.total_error_4_is_0:
285	mov	[ebx + 16], eax			; residual_bits_per_sample[4] = 0.0
286.rbps_end:
287	fstp	st0				; ST = [empty]
288	jmp	short .end
289.data_len_is_0:
290	; data_len == 0, so residual_bits_per_sample[*] = 0.0
291	xor	ebp, ebp
292	mov	edi, [esp + 44]
293	mov	[edi], ebp
294	mov	[edi + 4], ebp
295	mov	[edi + 8], ebp
296	mov	[edi + 12], ebp
297	mov	[edi + 16], ebp
298	add	ebp, byte 4			; order = 4
299
300.end:
301	mov	eax, ebp			; return order
302	add	esp, byte 16
303	pop	edi
304	pop	esi
305	pop	ebx
306	pop	ebp
307	ret
308
309; end
310