* commit '4958f35a2ebc307049ff2104ffb944f5f457feb3':
dsputil: Move apply_window_int16 to ac3dsp
Conflicts:
libavcodec/arm/ac3dsp_init_arm.c
libavcodec/arm/ac3dsp_neon.S
libavcodec/x86/ac3dsp_init.c
Merged-by: Michael Niedermayer <michaelni@gmx.at>
... | ... |
@@ -239,6 +239,19 @@ static void ac3_downmix_c(float **samples, float (*matrix)[2], |
239 | 239 |
} |
240 | 240 |
} |
241 | 241 |
|
242 |
+static void apply_window_int16_c(int16_t *output, const int16_t *input, |
|
243 |
+ const int16_t *window, unsigned int len) |
|
244 |
+{ |
|
245 |
+ int i; |
|
246 |
+ int len2 = len >> 1; |
|
247 |
+ |
|
248 |
+ for (i = 0; i < len2; i++) { |
|
249 |
+ int16_t w = window[i]; |
|
250 |
+ output[i] = (MUL16(input[i], w) + (1 << 14)) >> 15; |
|
251 |
+ output[len-i-1] = (MUL16(input[len-i-1], w) + (1 << 14)) >> 15; |
|
252 |
+ } |
|
253 |
+} |
|
254 |
+ |
|
242 | 255 |
av_cold void ff_ac3dsp_init(AC3DSPContext *c, int bit_exact) |
243 | 256 |
{ |
244 | 257 |
c->ac3_exponent_min = ac3_exponent_min_c; |
... | ... |
@@ -253,6 +266,7 @@ av_cold void ff_ac3dsp_init(AC3DSPContext *c, int bit_exact) |
253 | 253 |
c->sum_square_butterfly_int32 = ac3_sum_square_butterfly_int32_c; |
254 | 254 |
c->sum_square_butterfly_float = ac3_sum_square_butterfly_float_c; |
255 | 255 |
c->downmix = ac3_downmix_c; |
256 |
+ c->apply_window_int16 = apply_window_int16_c; |
|
256 | 257 |
|
257 | 258 |
if (ARCH_ARM) |
258 | 259 |
ff_ac3dsp_init_arm(c, bit_exact); |
... | ... |
@@ -134,6 +134,20 @@ typedef struct AC3DSPContext { |
134 | 134 |
|
135 | 135 |
void (*downmix)(float **samples, float (*matrix)[2], int out_ch, |
136 | 136 |
int in_ch, int len); |
137 |
+ |
|
138 |
+ /** |
|
139 |
+ * Apply symmetric window in 16-bit fixed-point. |
|
140 |
+ * @param output destination array |
|
141 |
+ * constraints: 16-byte aligned |
|
142 |
+ * @param input source array |
|
143 |
+ * constraints: 16-byte aligned |
|
144 |
+ * @param window window array |
|
145 |
+ * constraints: 16-byte aligned, at least len/2 elements |
|
146 |
+ * @param len full window length |
|
147 |
+ * constraints: multiple of ? greater than zero |
|
148 |
+ */ |
|
149 |
+ void (*apply_window_int16)(int16_t *output, const int16_t *input, |
|
150 |
+ const int16_t *window, unsigned int len); |
|
137 | 151 |
} AC3DSPContext; |
138 | 152 |
|
139 | 153 |
void ff_ac3dsp_init (AC3DSPContext *c, int bit_exact); |
... | ... |
@@ -71,17 +71,6 @@ av_cold int AC3_NAME(mdct_init)(AC3EncodeContext *s) |
71 | 71 |
|
72 | 72 |
|
73 | 73 |
/* |
74 |
- * Apply KBD window to input samples prior to MDCT. |
|
75 |
- */ |
|
76 |
-static void apply_window(void *dsp, int16_t *output, const int16_t *input, |
|
77 |
- const int16_t *window, unsigned int len) |
|
78 |
-{ |
|
79 |
- DSPContext *dsp0 = dsp; |
|
80 |
- dsp0->apply_window_int16(output, input, window, len); |
|
81 |
-} |
|
82 |
- |
|
83 |
- |
|
84 |
-/* |
|
85 | 74 |
* Normalize the input samples to use the maximum available precision. |
86 | 75 |
* This assumes signed 16-bit input samples. |
87 | 76 |
*/ |
... | ... |
@@ -88,18 +88,6 @@ av_cold int ff_ac3_float_mdct_init(AC3EncodeContext *s) |
88 | 88 |
|
89 | 89 |
|
90 | 90 |
/* |
91 |
- * Apply KBD window to input samples prior to MDCT. |
|
92 |
- */ |
|
93 |
-static void apply_window(void *dsp, float *output, |
|
94 |
- const float *input, const float *window, |
|
95 |
- unsigned int len) |
|
96 |
-{ |
|
97 |
- AVFloatDSPContext *fdsp = dsp; |
|
98 |
- fdsp->vector_fmul(output, input, window, len); |
|
99 |
-} |
|
100 |
- |
|
101 |
- |
|
102 |
-/* |
|
103 | 91 |
* Normalize the input samples. |
104 | 92 |
* Not needed for the floating-point encoder. |
105 | 93 |
*/ |
... | ... |
@@ -34,10 +34,6 @@ |
34 | 34 |
|
35 | 35 |
static void scale_coefficients(AC3EncodeContext *s); |
36 | 36 |
|
37 |
-static void apply_window(void *dsp, SampleType *output, |
|
38 |
- const SampleType *input, const SampleType *window, |
|
39 |
- unsigned int len); |
|
40 |
- |
|
41 | 37 |
static int normalize_samples(AC3EncodeContext *s); |
42 | 38 |
|
43 | 39 |
static void clip_coefficients(DSPContext *dsp, CoefType *coef, unsigned int len); |
... | ... |
@@ -105,11 +101,11 @@ static void apply_mdct(AC3EncodeContext *s) |
105 | 105 |
const SampleType *input_samples = &s->planar_samples[ch][blk * AC3_BLOCK_SIZE]; |
106 | 106 |
|
107 | 107 |
#if CONFIG_AC3ENC_FLOAT |
108 |
- apply_window(&s->fdsp, s->windowed_samples, input_samples, |
|
109 |
- s->mdct_window, AC3_WINDOW_SIZE); |
|
108 |
+ s->fdsp.vector_fmul(s->windowed_samples, input_samples, |
|
109 |
+ s->mdct_window, AC3_WINDOW_SIZE); |
|
110 | 110 |
#else |
111 |
- apply_window(&s->dsp, s->windowed_samples, input_samples, |
|
112 |
- s->mdct_window, AC3_WINDOW_SIZE); |
|
111 |
+ s->ac3dsp.apply_window_int16(s->windowed_samples, input_samples, |
|
112 |
+ s->mdct_window, AC3_WINDOW_SIZE); |
|
113 | 113 |
#endif |
114 | 114 |
|
115 | 115 |
if (s->fixed_point) |
... | ... |
@@ -31,6 +31,8 @@ void ff_ac3_lshift_int16_neon(int16_t *src, unsigned len, unsigned shift); |
31 | 31 |
void ff_ac3_rshift_int32_neon(int32_t *src, unsigned len, unsigned shift); |
32 | 32 |
void ff_float_to_fixed24_neon(int32_t *dst, const float *src, unsigned int len); |
33 | 33 |
void ff_ac3_extract_exponents_neon(uint8_t *exp, int32_t *coef, int nb_coefs); |
34 |
+void ff_apply_window_int16_neon(int16_t *dst, const int16_t *src, |
|
35 |
+ const int16_t *window, unsigned n); |
|
34 | 36 |
void ff_ac3_sum_square_butterfly_int32_neon(int64_t sum[4], |
35 | 37 |
const int32_t *coef0, |
36 | 38 |
const int32_t *coef1, |
... | ... |
@@ -64,6 +66,7 @@ av_cold void ff_ac3dsp_init_arm(AC3DSPContext *c, int bit_exact) |
64 | 64 |
c->ac3_rshift_int32 = ff_ac3_rshift_int32_neon; |
65 | 65 |
c->float_to_fixed24 = ff_float_to_fixed24_neon; |
66 | 66 |
c->extract_exponents = ff_ac3_extract_exponents_neon; |
67 |
+ c->apply_window_int16 = ff_apply_window_int16_neon; |
|
67 | 68 |
c->sum_square_butterfly_int32 = ff_ac3_sum_square_butterfly_int32_neon; |
68 | 69 |
c->sum_square_butterfly_float = ff_ac3_sum_square_butterfly_float_neon; |
69 | 70 |
} |
... | ... |
@@ -109,6 +109,29 @@ function ff_ac3_extract_exponents_neon, export=1 |
109 | 109 |
bx lr |
110 | 110 |
endfunc |
111 | 111 |
|
112 |
+function ff_apply_window_int16_neon, export=1 |
|
113 |
+ push {r4,lr} |
|
114 |
+ add r4, r1, r3, lsl #1 |
|
115 |
+ add lr, r0, r3, lsl #1 |
|
116 |
+ sub r4, r4, #16 |
|
117 |
+ sub lr, lr, #16 |
|
118 |
+ mov r12, #-16 |
|
119 |
+1: |
|
120 |
+ vld1.16 {q0}, [r1,:128]! |
|
121 |
+ vld1.16 {q2}, [r2,:128]! |
|
122 |
+ vld1.16 {q1}, [r4,:128], r12 |
|
123 |
+ vrev64.16 q3, q2 |
|
124 |
+ vqrdmulh.s16 q0, q0, q2 |
|
125 |
+ vqrdmulh.s16 d2, d2, d7 |
|
126 |
+ vqrdmulh.s16 d3, d3, d6 |
|
127 |
+ vst1.16 {q0}, [r0,:128]! |
|
128 |
+ vst1.16 {q1}, [lr,:128], r12 |
|
129 |
+ subs r3, r3, #16 |
|
130 |
+ bgt 1b |
|
131 |
+ |
|
132 |
+ pop {r4,pc} |
|
133 |
+endfunc |
|
134 |
+ |
|
112 | 135 |
function ff_ac3_sum_square_butterfly_int32_neon, export=1 |
113 | 136 |
vmov.i64 q0, #0 |
114 | 137 |
vmov.i64 q1, #0 |
... | ... |
@@ -45,9 +45,6 @@ int32_t ff_scalarproduct_int16_neon(const int16_t *v1, const int16_t *v2, int le |
45 | 45 |
int32_t ff_scalarproduct_and_madd_int16_neon(int16_t *v1, const int16_t *v2, |
46 | 46 |
const int16_t *v3, int len, int mul); |
47 | 47 |
|
48 |
-void ff_apply_window_int16_neon(int16_t *dst, const int16_t *src, |
|
49 |
- const int16_t *window, unsigned n); |
|
50 |
- |
|
51 | 48 |
av_cold void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx) |
52 | 49 |
{ |
53 | 50 |
const int high_bit_depth = avctx->bits_per_raw_sample > 8; |
... | ... |
@@ -76,6 +73,4 @@ av_cold void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx) |
76 | 76 |
|
77 | 77 |
c->scalarproduct_int16 = ff_scalarproduct_int16_neon; |
78 | 78 |
c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_neon; |
79 |
- |
|
80 |
- c->apply_window_int16 = ff_apply_window_int16_neon; |
|
81 | 79 |
} |
... | ... |
@@ -169,29 +169,6 @@ NOVFP ldr r2, [sp] |
169 | 169 |
bx lr |
170 | 170 |
endfunc |
171 | 171 |
|
172 |
-function ff_apply_window_int16_neon, export=1 |
|
173 |
- push {r4,lr} |
|
174 |
- add r4, r1, r3, lsl #1 |
|
175 |
- add lr, r0, r3, lsl #1 |
|
176 |
- sub r4, r4, #16 |
|
177 |
- sub lr, lr, #16 |
|
178 |
- mov r12, #-16 |
|
179 |
-1: |
|
180 |
- vld1.16 {q0}, [r1,:128]! |
|
181 |
- vld1.16 {q2}, [r2,:128]! |
|
182 |
- vld1.16 {q1}, [r4,:128], r12 |
|
183 |
- vrev64.16 q3, q2 |
|
184 |
- vqrdmulh.s16 q0, q0, q2 |
|
185 |
- vqrdmulh.s16 d2, d2, d7 |
|
186 |
- vqrdmulh.s16 d3, d3, d6 |
|
187 |
- vst1.16 {q0}, [r0,:128]! |
|
188 |
- vst1.16 {q1}, [lr,:128], r12 |
|
189 |
- subs r3, r3, #16 |
|
190 |
- bgt 1b |
|
191 |
- |
|
192 |
- pop {r4,pc} |
|
193 |
-endfunc |
|
194 |
- |
|
195 | 172 |
function ff_vector_clip_int32_neon, export=1 |
196 | 173 |
vdup.32 q0, r2 |
197 | 174 |
vdup.32 q1, r3 |
... | ... |
@@ -2495,19 +2495,6 @@ static int32_t scalarproduct_and_madd_int16_c(int16_t *v1, const int16_t *v2, co |
2495 | 2495 |
return res; |
2496 | 2496 |
} |
2497 | 2497 |
|
2498 |
-static void apply_window_int16_c(int16_t *output, const int16_t *input, |
|
2499 |
- const int16_t *window, unsigned int len) |
|
2500 |
-{ |
|
2501 |
- int i; |
|
2502 |
- int len2 = len >> 1; |
|
2503 |
- |
|
2504 |
- for (i = 0; i < len2; i++) { |
|
2505 |
- int16_t w = window[i]; |
|
2506 |
- output[i] = (MUL16(input[i], w) + (1 << 14)) >> 15; |
|
2507 |
- output[len-i-1] = (MUL16(input[len-i-1], w) + (1 << 14)) >> 15; |
|
2508 |
- } |
|
2509 |
-} |
|
2510 |
- |
|
2511 | 2498 |
static void vector_clip_int32_c(int32_t *dst, const int32_t *src, int32_t min, |
2512 | 2499 |
int32_t max, unsigned int len) |
2513 | 2500 |
{ |
... | ... |
@@ -2799,7 +2786,6 @@ av_cold void ff_dsputil_init(DSPContext* c, AVCodecContext *avctx) |
2799 | 2799 |
c->vector_clipf = vector_clipf_c; |
2800 | 2800 |
c->scalarproduct_int16 = scalarproduct_int16_c; |
2801 | 2801 |
c->scalarproduct_and_madd_int16 = scalarproduct_and_madd_int16_c; |
2802 |
- c->apply_window_int16 = apply_window_int16_c; |
|
2803 | 2802 |
c->vector_clip_int32 = vector_clip_int32_c; |
2804 | 2803 |
|
2805 | 2804 |
c->shrink[0]= av_image_copy_plane; |
... | ... |
@@ -275,20 +275,6 @@ typedef struct DSPContext { |
275 | 275 |
int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul); |
276 | 276 |
|
277 | 277 |
/** |
278 |
- * Apply symmetric window in 16-bit fixed-point. |
|
279 |
- * @param output destination array |
|
280 |
- * constraints: 16-byte aligned |
|
281 |
- * @param input source array |
|
282 |
- * constraints: 16-byte aligned |
|
283 |
- * @param window window array |
|
284 |
- * constraints: 16-byte aligned, at least len/2 elements |
|
285 |
- * @param len full window length |
|
286 |
- * constraints: multiple of ? greater than zero |
|
287 |
- */ |
|
288 |
- void (*apply_window_int16)(int16_t *output, const int16_t *input, |
|
289 |
- const int16_t *window, unsigned int len); |
|
290 |
- |
|
291 |
- /** |
|
292 | 278 |
* Clip each element in an array of int32_t to a given minimum and maximum value. |
293 | 279 |
* @param dst destination array |
294 | 280 |
* constraints: 16-byte aligned |
... | ... |
@@ -51,6 +51,19 @@ void ff_ac3_extract_exponents_3dnow(uint8_t *exp, int32_t *coef, int nb_coefs); |
51 | 51 |
void ff_ac3_extract_exponents_sse2 (uint8_t *exp, int32_t *coef, int nb_coefs); |
52 | 52 |
void ff_ac3_extract_exponents_ssse3(uint8_t *exp, int32_t *coef, int nb_coefs); |
53 | 53 |
|
54 |
+void ff_apply_window_int16_round_mmxext(int16_t *output, const int16_t *input, |
|
55 |
+ const int16_t *window, unsigned int len); |
|
56 |
+void ff_apply_window_int16_round_sse2(int16_t *output, const int16_t *input, |
|
57 |
+ const int16_t *window, unsigned int len); |
|
58 |
+void ff_apply_window_int16_mmxext(int16_t *output, const int16_t *input, |
|
59 |
+ const int16_t *window, unsigned int len); |
|
60 |
+void ff_apply_window_int16_sse2(int16_t *output, const int16_t *input, |
|
61 |
+ const int16_t *window, unsigned int len); |
|
62 |
+void ff_apply_window_int16_ssse3(int16_t *output, const int16_t *input, |
|
63 |
+ const int16_t *window, unsigned int len); |
|
64 |
+void ff_apply_window_int16_ssse3_atom(int16_t *output, const int16_t *input, |
|
65 |
+ const int16_t *window, unsigned int len); |
|
66 |
+ |
|
54 | 67 |
#if ARCH_X86_32 && defined(__INTEL_COMPILER) |
55 | 68 |
# undef HAVE_7REGS |
56 | 69 |
# define HAVE_7REGS 0 |
... | ... |
@@ -201,6 +214,11 @@ av_cold void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact) |
201 | 201 |
if (EXTERNAL_MMXEXT(cpu_flags)) { |
202 | 202 |
c->ac3_exponent_min = ff_ac3_exponent_min_mmxext; |
203 | 203 |
c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_mmxext; |
204 |
+ if (bit_exact) { |
|
205 |
+ c->apply_window_int16 = ff_apply_window_int16_mmxext; |
|
206 |
+ } else { |
|
207 |
+ c->apply_window_int16 = ff_apply_window_int16_round_mmxext; |
|
208 |
+ } |
|
204 | 209 |
} |
205 | 210 |
if (EXTERNAL_SSE(cpu_flags)) { |
206 | 211 |
c->float_to_fixed24 = ff_float_to_fixed24_sse; |
... | ... |
@@ -215,11 +233,19 @@ av_cold void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact) |
215 | 215 |
c->ac3_lshift_int16 = ff_ac3_lshift_int16_sse2; |
216 | 216 |
c->ac3_rshift_int32 = ff_ac3_rshift_int32_sse2; |
217 | 217 |
} |
218 |
+ if (bit_exact) { |
|
219 |
+ c->apply_window_int16 = ff_apply_window_int16_sse2; |
|
220 |
+ } else if (!(cpu_flags & AV_CPU_FLAG_SSE2SLOW)) { |
|
221 |
+ c->apply_window_int16 = ff_apply_window_int16_round_sse2; |
|
222 |
+ } |
|
218 | 223 |
} |
219 | 224 |
if (EXTERNAL_SSSE3(cpu_flags)) { |
220 | 225 |
c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_ssse3; |
221 |
- if (!(cpu_flags & AV_CPU_FLAG_ATOM)) { |
|
226 |
+ if (cpu_flags & AV_CPU_FLAG_ATOM) { |
|
227 |
+ c->apply_window_int16 = ff_apply_window_int16_ssse3_atom; |
|
228 |
+ } else { |
|
222 | 229 |
c->extract_exponents = ff_ac3_extract_exponents_ssse3; |
230 |
+ c->apply_window_int16 = ff_apply_window_int16_ssse3; |
|
223 | 231 |
} |
224 | 232 |
} |
225 | 233 |
|
... | ... |
@@ -86,19 +86,6 @@ int32_t ff_scalarproduct_and_madd_int16_ssse3(int16_t *v1, const int16_t *v2, |
86 | 86 |
const int16_t *v3, |
87 | 87 |
int order, int mul); |
88 | 88 |
|
89 |
-void ff_apply_window_int16_round_mmxext(int16_t *output, const int16_t *input, |
|
90 |
- const int16_t *window, unsigned int len); |
|
91 |
-void ff_apply_window_int16_round_sse2(int16_t *output, const int16_t *input, |
|
92 |
- const int16_t *window, unsigned int len); |
|
93 |
-void ff_apply_window_int16_mmxext(int16_t *output, const int16_t *input, |
|
94 |
- const int16_t *window, unsigned int len); |
|
95 |
-void ff_apply_window_int16_sse2(int16_t *output, const int16_t *input, |
|
96 |
- const int16_t *window, unsigned int len); |
|
97 |
-void ff_apply_window_int16_ssse3(int16_t *output, const int16_t *input, |
|
98 |
- const int16_t *window, unsigned int len); |
|
99 |
-void ff_apply_window_int16_ssse3_atom(int16_t *output, const int16_t *input, |
|
100 |
- const int16_t *window, unsigned int len); |
|
101 |
- |
|
102 | 89 |
void ff_bswap32_buf_ssse3(uint32_t *dst, const uint32_t *src, int w); |
103 | 90 |
void ff_bswap32_buf_sse2(uint32_t *dst, const uint32_t *src, int w); |
104 | 91 |
|
... | ... |
@@ -586,12 +573,6 @@ static av_cold void dsputil_init_mmxext(DSPContext *c, AVCodecContext *avctx, |
586 | 586 |
|
587 | 587 |
c->scalarproduct_int16 = ff_scalarproduct_int16_mmxext; |
588 | 588 |
c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_mmxext; |
589 |
- |
|
590 |
- if (avctx->flags & CODEC_FLAG_BITEXACT) { |
|
591 |
- c->apply_window_int16 = ff_apply_window_int16_mmxext; |
|
592 |
- } else { |
|
593 |
- c->apply_window_int16 = ff_apply_window_int16_round_mmxext; |
|
594 |
- } |
|
595 | 589 |
#endif /* HAVE_MMXEXT_EXTERNAL */ |
596 | 590 |
} |
597 | 591 |
|
... | ... |
@@ -647,11 +628,6 @@ static av_cold void dsputil_init_sse2(DSPContext *c, AVCodecContext *avctx, |
647 | 647 |
} else { |
648 | 648 |
c->vector_clip_int32 = ff_vector_clip_int32_sse2; |
649 | 649 |
} |
650 |
- if (avctx->flags & CODEC_FLAG_BITEXACT) { |
|
651 |
- c->apply_window_int16 = ff_apply_window_int16_sse2; |
|
652 |
- } else if (!(cpu_flags & AV_CPU_FLAG_SSE2SLOW)) { |
|
653 |
- c->apply_window_int16 = ff_apply_window_int16_round_sse2; |
|
654 |
- } |
|
655 | 650 |
c->bswap_buf = ff_bswap32_buf_sse2; |
656 | 651 |
#endif /* HAVE_SSE2_EXTERNAL */ |
657 | 652 |
} |
... | ... |
@@ -664,10 +640,6 @@ static av_cold void dsputil_init_ssse3(DSPContext *c, AVCodecContext *avctx, |
664 | 664 |
if (cpu_flags & AV_CPU_FLAG_SSE4) // not really SSE4, just slow on Conroe |
665 | 665 |
c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_sse4; |
666 | 666 |
|
667 |
- if (cpu_flags & AV_CPU_FLAG_ATOM) |
|
668 |
- c->apply_window_int16 = ff_apply_window_int16_ssse3_atom; |
|
669 |
- else |
|
670 |
- c->apply_window_int16 = ff_apply_window_int16_ssse3; |
|
671 | 667 |
if (!(cpu_flags & (AV_CPU_FLAG_SSE42 | AV_CPU_FLAG_3DNOW))) // cachesplit |
672 | 668 |
c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_ssse3; |
673 | 669 |
c->bswap_buf = ff_bswap32_buf_ssse3; |