From 4958f35a2ebc307049ff2104ffb944f5f457feb3 Mon Sep 17 00:00:00 2001 From: Diego Biurrun Date: Fri, 6 Dec 2013 12:22:40 +0000 Subject: [PATCH] dsputil: Move apply_window_int16 to ac3dsp The (optimized) functions are used nowhere else. --- libavcodec/ac3dsp.c | 15 +++++++++++++++ libavcodec/ac3dsp.h | 14 ++++++++++++++ libavcodec/ac3enc_fixed.c | 11 ----------- libavcodec/ac3enc_float.c | 12 ------------ libavcodec/ac3enc_template.c | 12 ++++-------- libavcodec/arm/ac3dsp_init_arm.c | 3 +++ libavcodec/arm/ac3dsp_neon.S | 23 +++++++++++++++++++++++ libavcodec/arm/dsputil_init_neon.c | 5 ----- libavcodec/arm/dsputil_neon.S | 23 ----------------------- libavcodec/dsputil.c | 14 -------------- libavcodec/dsputil.h | 14 -------------- libavcodec/x86/ac3dsp_init.c | 28 +++++++++++++++++++++++++++- libavcodec/x86/dsputil_init.c | 28 ---------------------------- 13 files changed, 86 insertions(+), 116 deletions(-) diff --git a/libavcodec/ac3dsp.c b/libavcodec/ac3dsp.c index ef19f6a449..e792bcfce5 100644 --- a/libavcodec/ac3dsp.c +++ b/libavcodec/ac3dsp.c @@ -23,6 +23,7 @@ #include "avcodec.h" #include "ac3.h" #include "ac3dsp.h" +#include "mathops.h" static void ac3_exponent_min_c(uint8_t *exp, int num_reuse_blocks, int nb_coefs) { @@ -196,6 +197,19 @@ static void ac3_downmix_c(float **samples, float (*matrix)[2], } } +static void apply_window_int16_c(int16_t *output, const int16_t *input, + const int16_t *window, unsigned int len) +{ + int i; + int len2 = len >> 1; + + for (i = 0; i < len2; i++) { + int16_t w = window[i]; + output[i] = (MUL16(input[i], w) + (1 << 14)) >> 15; + output[len-i-1] = (MUL16(input[len-i-1], w) + (1 << 14)) >> 15; + } +} + av_cold void ff_ac3dsp_init(AC3DSPContext *c, int bit_exact) { c->ac3_exponent_min = ac3_exponent_min_c; @@ -208,6 +222,7 @@ av_cold void ff_ac3dsp_init(AC3DSPContext *c, int bit_exact) c->compute_mantissa_size = ac3_compute_mantissa_size_c; c->extract_exponents = ac3_extract_exponents_c; c->downmix = ac3_downmix_c; + c->apply_window_int16 = apply_window_int16_c; if (ARCH_ARM) ff_ac3dsp_init_arm(c, bit_exact); diff --git a/libavcodec/ac3dsp.h b/libavcodec/ac3dsp.h index 882eb7681a..e350c8824d 100644 --- a/libavcodec/ac3dsp.h +++ b/libavcodec/ac3dsp.h @@ -128,6 +128,20 @@ typedef struct AC3DSPContext { void (*downmix)(float **samples, float (*matrix)[2], int out_ch, int in_ch, int len); + + /** + * Apply symmetric window in 16-bit fixed-point. + * @param output destination array + * constraints: 16-byte aligned + * @param input source array + * constraints: 16-byte aligned + * @param window window array + * constraints: 16-byte aligned, at least len/2 elements + * @param len full window length + * constraints: multiple of ? greater than zero + */ + void (*apply_window_int16)(int16_t *output, const int16_t *input, + const int16_t *window, unsigned int len); } AC3DSPContext; void ff_ac3dsp_init (AC3DSPContext *c, int bit_exact); diff --git a/libavcodec/ac3enc_fixed.c b/libavcodec/ac3enc_fixed.c index 6e251899d8..7d0ddab11b 100644 --- a/libavcodec/ac3enc_fixed.c +++ b/libavcodec/ac3enc_fixed.c @@ -65,17 +65,6 @@ av_cold int AC3_NAME(mdct_init)(AC3EncodeContext *s) } -/* - * Apply KBD window to input samples prior to MDCT. - */ -static void apply_window(void *dsp, int16_t *output, const int16_t *input, - const int16_t *window, unsigned int len) -{ - DSPContext *dsp0 = dsp; - dsp0->apply_window_int16(output, input, window, len); -} - - /* * Normalize the input samples to use the maximum available precision. * This assumes signed 16-bit input samples. diff --git a/libavcodec/ac3enc_float.c b/libavcodec/ac3enc_float.c index 86a5cff95f..71db68cbef 100644 --- a/libavcodec/ac3enc_float.c +++ b/libavcodec/ac3enc_float.c @@ -83,18 +83,6 @@ av_cold int ff_ac3_float_mdct_init(AC3EncodeContext *s) } -/* - * Apply KBD window to input samples prior to MDCT. - */ -static void apply_window(void *dsp, float *output, - const float *input, const float *window, - unsigned int len) -{ - AVFloatDSPContext *fdsp = dsp; - fdsp->vector_fmul(output, input, window, len); -} - - /* * Normalize the input samples. * Not needed for the floating-point encoder. diff --git a/libavcodec/ac3enc_template.c b/libavcodec/ac3enc_template.c index ae9100c3ac..1b887268ed 100644 --- a/libavcodec/ac3enc_template.c +++ b/libavcodec/ac3enc_template.c @@ -34,10 +34,6 @@ static void scale_coefficients(AC3EncodeContext *s); -static void apply_window(void *dsp, SampleType *output, - const SampleType *input, const SampleType *window, - unsigned int len); - static int normalize_samples(AC3EncodeContext *s); static void clip_coefficients(DSPContext *dsp, CoefType *coef, unsigned int len); @@ -102,11 +98,11 @@ static void apply_mdct(AC3EncodeContext *s) const SampleType *input_samples = &s->planar_samples[ch][blk * AC3_BLOCK_SIZE]; #if CONFIG_AC3ENC_FLOAT - apply_window(&s->fdsp, s->windowed_samples, input_samples, - s->mdct_window, AC3_WINDOW_SIZE); + s->fdsp.vector_fmul(s->windowed_samples, input_samples, + s->mdct_window, AC3_WINDOW_SIZE); #else - apply_window(&s->dsp, s->windowed_samples, input_samples, - s->mdct_window, AC3_WINDOW_SIZE); + s->ac3dsp.apply_window_int16(s->windowed_samples, input_samples, + s->mdct_window, AC3_WINDOW_SIZE); #endif if (s->fixed_point) diff --git a/libavcodec/arm/ac3dsp_init_arm.c b/libavcodec/arm/ac3dsp_init_arm.c index d7cb95b669..a48353a099 100644 --- a/libavcodec/arm/ac3dsp_init_arm.c +++ b/libavcodec/arm/ac3dsp_init_arm.c @@ -31,6 +31,8 @@ void ff_ac3_lshift_int16_neon(int16_t *src, unsigned len, unsigned shift); void ff_ac3_rshift_int32_neon(int32_t *src, unsigned len, unsigned shift); void ff_float_to_fixed24_neon(int32_t *dst, const float *src, unsigned int len); void ff_ac3_extract_exponents_neon(uint8_t *exp, int32_t *coef, int nb_coefs); +void ff_apply_window_int16_neon(int16_t *dst, const int16_t *src, + const int16_t *window, unsigned n); void ff_ac3_bit_alloc_calc_bap_armv6(int16_t *mask, int16_t *psd, int start, int end, @@ -56,5 +58,6 @@ av_cold void ff_ac3dsp_init_arm(AC3DSPContext *c, int bit_exact) c->ac3_rshift_int32 = ff_ac3_rshift_int32_neon; c->float_to_fixed24 = ff_float_to_fixed24_neon; c->extract_exponents = ff_ac3_extract_exponents_neon; + c->apply_window_int16 = ff_apply_window_int16_neon; } } diff --git a/libavcodec/arm/ac3dsp_neon.S b/libavcodec/arm/ac3dsp_neon.S index 82ff8af9fb..f97b1907df 100644 --- a/libavcodec/arm/ac3dsp_neon.S +++ b/libavcodec/arm/ac3dsp_neon.S @@ -108,3 +108,26 @@ function ff_ac3_extract_exponents_neon, export=1 bgt 1b bx lr endfunc + +function ff_apply_window_int16_neon, export=1 + push {r4,lr} + add r4, r1, r3, lsl #1 + add lr, r0, r3, lsl #1 + sub r4, r4, #16 + sub lr, lr, #16 + mov r12, #-16 +1: + vld1.16 {q0}, [r1,:128]! + vld1.16 {q2}, [r2,:128]! + vld1.16 {q1}, [r4,:128], r12 + vrev64.16 q3, q2 + vqrdmulh.s16 q0, q0, q2 + vqrdmulh.s16 d2, d2, d7 + vqrdmulh.s16 d3, d3, d6 + vst1.16 {q0}, [r0,:128]! + vst1.16 {q1}, [lr,:128], r12 + subs r3, r3, #16 + bgt 1b + + pop {r4,pc} +endfunc diff --git a/libavcodec/arm/dsputil_init_neon.c b/libavcodec/arm/dsputil_init_neon.c index 793fab1cf8..0926c84428 100644 --- a/libavcodec/arm/dsputil_init_neon.c +++ b/libavcodec/arm/dsputil_init_neon.c @@ -45,9 +45,6 @@ int32_t ff_scalarproduct_int16_neon(const int16_t *v1, const int16_t *v2, int le int32_t ff_scalarproduct_and_madd_int16_neon(int16_t *v1, const int16_t *v2, const int16_t *v3, int len, int mul); -void ff_apply_window_int16_neon(int16_t *dst, const int16_t *src, - const int16_t *window, unsigned n); - av_cold void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx) { 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) c->scalarproduct_int16 = ff_scalarproduct_int16_neon; c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_neon; - - c->apply_window_int16 = ff_apply_window_int16_neon; } diff --git a/libavcodec/arm/dsputil_neon.S b/libavcodec/arm/dsputil_neon.S index 81662c04ad..e30bd10b17 100644 --- a/libavcodec/arm/dsputil_neon.S +++ b/libavcodec/arm/dsputil_neon.S @@ -169,29 +169,6 @@ NOVFP ldr r2, [sp] bx lr endfunc -function ff_apply_window_int16_neon, export=1 - push {r4,lr} - add r4, r1, r3, lsl #1 - add lr, r0, r3, lsl #1 - sub r4, r4, #16 - sub lr, lr, #16 - mov r12, #-16 -1: - vld1.16 {q0}, [r1,:128]! - vld1.16 {q2}, [r2,:128]! - vld1.16 {q1}, [r4,:128], r12 - vrev64.16 q3, q2 - vqrdmulh.s16 q0, q0, q2 - vqrdmulh.s16 d2, d2, d7 - vqrdmulh.s16 d3, d3, d6 - vst1.16 {q0}, [r0,:128]! - vst1.16 {q1}, [lr,:128], r12 - subs r3, r3, #16 - bgt 1b - - pop {r4,pc} -endfunc - function ff_vector_clip_int32_neon, export=1 vdup.32 q0, r2 vdup.32 q1, r3 diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c index 2aac64f09b..56207e805b 100644 --- a/libavcodec/dsputil.c +++ b/libavcodec/dsputil.c @@ -2380,19 +2380,6 @@ static int32_t scalarproduct_and_madd_int16_c(int16_t *v1, const int16_t *v2, co return res; } -static void apply_window_int16_c(int16_t *output, const int16_t *input, - const int16_t *window, unsigned int len) -{ - int i; - int len2 = len >> 1; - - for (i = 0; i < len2; i++) { - int16_t w = window[i]; - output[i] = (MUL16(input[i], w) + (1 << 14)) >> 15; - output[len-i-1] = (MUL16(input[len-i-1], w) + (1 << 14)) >> 15; - } -} - static void vector_clip_int32_c(int32_t *dst, const int32_t *src, int32_t min, int32_t max, unsigned int len) { @@ -2628,7 +2615,6 @@ av_cold void ff_dsputil_init(DSPContext* c, AVCodecContext *avctx) c->vector_clipf = vector_clipf_c; c->scalarproduct_int16 = scalarproduct_int16_c; c->scalarproduct_and_madd_int16 = scalarproduct_and_madd_int16_c; - c->apply_window_int16 = apply_window_int16_c; c->vector_clip_int32 = vector_clip_int32_c; c->shrink[0]= av_image_copy_plane; diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h index f571a9927d..7bd92e53d3 100644 --- a/libavcodec/dsputil.h +++ b/libavcodec/dsputil.h @@ -271,20 +271,6 @@ typedef struct DSPContext { */ int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul); - /** - * Apply symmetric window in 16-bit fixed-point. - * @param output destination array - * constraints: 16-byte aligned - * @param input source array - * constraints: 16-byte aligned - * @param window window array - * constraints: 16-byte aligned, at least len/2 elements - * @param len full window length - * constraints: multiple of ? greater than zero - */ - void (*apply_window_int16)(int16_t *output, const int16_t *input, - const int16_t *window, unsigned int len); - /** * Clip each element in an array of int32_t to a given minimum and maximum value. * @param dst destination array diff --git a/libavcodec/x86/ac3dsp_init.c b/libavcodec/x86/ac3dsp_init.c index ca10864323..9fbed3c0b4 100644 --- a/libavcodec/x86/ac3dsp_init.c +++ b/libavcodec/x86/ac3dsp_init.c @@ -51,6 +51,19 @@ void ff_ac3_extract_exponents_3dnow(uint8_t *exp, int32_t *coef, int nb_coefs); void ff_ac3_extract_exponents_sse2 (uint8_t *exp, int32_t *coef, int nb_coefs); void ff_ac3_extract_exponents_ssse3(uint8_t *exp, int32_t *coef, int nb_coefs); +void ff_apply_window_int16_round_mmxext(int16_t *output, const int16_t *input, + const int16_t *window, unsigned int len); +void ff_apply_window_int16_round_sse2(int16_t *output, const int16_t *input, + const int16_t *window, unsigned int len); +void ff_apply_window_int16_mmxext(int16_t *output, const int16_t *input, + const int16_t *window, unsigned int len); +void ff_apply_window_int16_sse2(int16_t *output, const int16_t *input, + const int16_t *window, unsigned int len); +void ff_apply_window_int16_ssse3(int16_t *output, const int16_t *input, + const int16_t *window, unsigned int len); +void ff_apply_window_int16_ssse3_atom(int16_t *output, const int16_t *input, + const int16_t *window, unsigned int len); + #if HAVE_SSE_INLINE && HAVE_7REGS #define IF1(x) x @@ -196,6 +209,11 @@ av_cold void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact) if (EXTERNAL_MMXEXT(cpu_flags)) { c->ac3_exponent_min = ff_ac3_exponent_min_mmxext; c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_mmxext; + if (bit_exact) { + c->apply_window_int16 = ff_apply_window_int16_mmxext; + } else { + c->apply_window_int16 = ff_apply_window_int16_round_mmxext; + } } if (EXTERNAL_SSE(cpu_flags)) { c->float_to_fixed24 = ff_float_to_fixed24_sse; @@ -210,11 +228,19 @@ av_cold void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact) c->ac3_lshift_int16 = ff_ac3_lshift_int16_sse2; c->ac3_rshift_int32 = ff_ac3_rshift_int32_sse2; } + if (bit_exact) { + c->apply_window_int16 = ff_apply_window_int16_sse2; + } else if (!(cpu_flags & AV_CPU_FLAG_SSE2SLOW)) { + c->apply_window_int16 = ff_apply_window_int16_round_sse2; + } } if (EXTERNAL_SSSE3(cpu_flags)) { c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_ssse3; - if (!(cpu_flags & AV_CPU_FLAG_ATOM)) { + if (cpu_flags & AV_CPU_FLAG_ATOM) { + c->apply_window_int16 = ff_apply_window_int16_ssse3_atom; + } else { c->extract_exponents = ff_ac3_extract_exponents_ssse3; + c->apply_window_int16 = ff_apply_window_int16_ssse3; } } diff --git a/libavcodec/x86/dsputil_init.c b/libavcodec/x86/dsputil_init.c index 2ac93516ca..c339e8fdd0 100644 --- a/libavcodec/x86/dsputil_init.c +++ b/libavcodec/x86/dsputil_init.c @@ -83,19 +83,6 @@ int32_t ff_scalarproduct_and_madd_int16_ssse3(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul); -void ff_apply_window_int16_round_mmxext(int16_t *output, const int16_t *input, - const int16_t *window, unsigned int len); -void ff_apply_window_int16_round_sse2(int16_t *output, const int16_t *input, - const int16_t *window, unsigned int len); -void ff_apply_window_int16_mmxext(int16_t *output, const int16_t *input, - const int16_t *window, unsigned int len); -void ff_apply_window_int16_sse2(int16_t *output, const int16_t *input, - const int16_t *window, unsigned int len); -void ff_apply_window_int16_ssse3(int16_t *output, const int16_t *input, - const int16_t *window, unsigned int len); -void ff_apply_window_int16_ssse3_atom(int16_t *output, const int16_t *input, - const int16_t *window, unsigned int len); - void ff_bswap32_buf_ssse3(uint32_t *dst, const uint32_t *src, int w); void ff_bswap32_buf_sse2(uint32_t *dst, const uint32_t *src, int w); @@ -596,12 +583,6 @@ static av_cold void dsputil_init_mmxext(DSPContext *c, AVCodecContext *avctx, c->scalarproduct_int16 = ff_scalarproduct_int16_mmxext; c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_mmxext; - - if (avctx->flags & CODEC_FLAG_BITEXACT) { - c->apply_window_int16 = ff_apply_window_int16_mmxext; - } else { - c->apply_window_int16 = ff_apply_window_int16_round_mmxext; - } #endif /* HAVE_MMXEXT_EXTERNAL */ } @@ -651,11 +632,6 @@ static av_cold void dsputil_init_sse2(DSPContext *c, AVCodecContext *avctx, } else { c->vector_clip_int32 = ff_vector_clip_int32_sse2; } - if (avctx->flags & CODEC_FLAG_BITEXACT) { - c->apply_window_int16 = ff_apply_window_int16_sse2; - } else if (!(cpu_flags & AV_CPU_FLAG_SSE2SLOW)) { - c->apply_window_int16 = ff_apply_window_int16_round_sse2; - } c->bswap_buf = ff_bswap32_buf_sse2; #endif /* HAVE_SSE2_EXTERNAL */ } @@ -668,10 +644,6 @@ static av_cold void dsputil_init_ssse3(DSPContext *c, AVCodecContext *avctx, if (cpu_flags & AV_CPU_FLAG_SSE4) // not really SSE4, just slow on Conroe c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_sse4; - if (cpu_flags & AV_CPU_FLAG_ATOM) - c->apply_window_int16 = ff_apply_window_int16_ssse3_atom; - else - c->apply_window_int16 = ff_apply_window_int16_ssse3; if (!(cpu_flags & (AV_CPU_FLAG_SSE42 | AV_CPU_FLAG_3DNOW))) // cachesplit c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_ssse3; c->bswap_buf = ff_bswap32_buf_ssse3;