The x86 runs short on registers because numerous elements are not static. In addition, splitting them allows more optimized code, at least for x86.
Unified · Split
| @@ -47,16 +47,43 @@ void ff_synth_filter_float_neon(FFTContext *imdct, · 46 unmodified lines | |||
| 47 | float out[32], const float in[32], | 47 | float out[32], const float in[32], |
| 48 | float scale); | 48 | float scale); |
| 49 | 49 | ||
| 50 | static void lfe_fir0_vfp(float *out, const float *in, const float *coefs, | ||
| 51 | float scale) | ||
| 52 | { | ||
| 53 | ff_dca_lfe_fir_vfp(out, in, coefs, 32, scale); | ||
| 54 | } | ||
| 55 | |||
| 56 | static void lfe_fir1_vfp(float *out, const float *in, const float *coefs, | ||
| 57 | float scale) | ||
| 58 | { | ||
| 59 | ff_dca_lfe_fir_vfp(out, in, coefs, 64, scale); | ||
| 60 | } | ||
| 61 | |||
| 62 | static void lfe_fir0_neon(float *out, const float *in, const float *coefs, | ||
| 63 | float scale) | ||
| 64 | { | ||
| 65 | ff_dca_lfe_fir_neon(out, in, coefs, 32, scale); | ||
| 66 | } | ||
| 67 | |||
| 68 | static void lfe_fir1_neon(float *out, const float *in, const float *coefs, | ||
| 69 | float scale) | ||
| 70 | { | ||
| 71 | ff_dca_lfe_fir_neon(out, in, coefs, 64, scale); | ||
| 72 | } | ||
| 73 | |||
| 50 | av_cold void ff_dcadsp_init_arm(DCADSPContext *s) | 74 | av_cold void ff_dcadsp_init_arm(DCADSPContext *s) |
| 51 | { | 75 | { |
| 52 | int cpu_flags = av_get_cpu_flags(); | 76 | int cpu_flags = av_get_cpu_flags(); |
| 53 | 77 | ||
| 54 | if (have_vfp(cpu_flags) && !have_vfpv3(cpu_flags)) { | 78 | if (have_vfp(cpu_flags) && !have_vfpv3(cpu_flags)) { |
| 55 | s->lfe_fir = ff_dca_lfe_fir_vfp; | 79 | s->lfe_fir[0] = lfe_fir0_vfp; |
| 80 | s->lfe_fir[1] = lfe_fir1_vfp; | ||
| 56 | s->qmf_32_subbands = ff_dca_qmf_32_subbands_vfp; | 81 | s->qmf_32_subbands = ff_dca_qmf_32_subbands_vfp; |
| 57 | } | 82 | } |
| 58 | if (have_neon(cpu_flags)) | 83 | if (have_neon(cpu_flags)) { |
| 59 | s->lfe_fir = ff_dca_lfe_fir_neon; | 84 | s->lfe_fir[0] = lfe_fir0_neon; |
| 85 | s->lfe_fir[1] = lfe_fir1_neon; | ||
| 86 | } | ||
| 60 | } | 87 | } |
| 61 | 88 | ||
| 62 | av_cold void ff_synth_filter_init_arm(SynthFilterContext *s) | 89 | av_cold void ff_synth_filter_init_arm(SynthFilterContext *s) |
| @@ -1118,23 +1118,23 @@ static void lfe_interpolation_fir(DCAContext *s, int decimation_select, · 1117 unmodified lines | |||
| 1118 | * samples_out: An array holding interpolated samples | 1118 | * samples_out: An array holding interpolated samples |
| 1119 | */ | 1119 | */ |
| 1120 | 1120 | ||
| 1121 | int decifactor; | 1121 | int idx; |
| 1122 | const float *prCoeff; | 1122 | const float *prCoeff; |
| 1123 | int deciindex; | 1123 | int deciindex; |
| 1124 | 1124 | ||
| 1125 | /* Select decimation filter */ | 1125 | /* Select decimation filter */ |
| 1126 | if (decimation_select == 1) { | 1126 | if (decimation_select == 1) { |
| 1127 | decifactor = 64; | 1127 | idx = 1; |
| 1128 | prCoeff = lfe_fir_128; | 1128 | prCoeff = lfe_fir_128; |
| 1129 | } else { | 1129 | } else { |
| 1130 | decifactor = 32; | 1130 | idx = 0; |
| 1131 | prCoeff = lfe_fir_64; | 1131 | prCoeff = lfe_fir_64; |
| 1132 | } | 1132 | } |
| 1133 | /* Interpolation */ | 1133 | /* Interpolation */ |
| 1134 | for (deciindex = 0; deciindex < num_deci_sample; deciindex++) { | 1134 | for (deciindex = 0; deciindex < num_deci_sample; deciindex++) { |
| 1135 | s->dcadsp.lfe_fir(samples_out, samples_in, prCoeff, decifactor, scale); | 1135 | s->dcadsp.lfe_fir[idx](samples_out, samples_in, prCoeff, scale); |
| 1136 | samples_in++; | 1136 | samples_in++; |
| 1137 | samples_out += 2 * decifactor; | 1137 | samples_out += 2 * 32 * (1 + idx); |
| 1138 | } | 1138 | } |
| 1139 | } | 1139 | } |
| 1140 | 1140 | ||
| @@ -32,8 +32,9 @@ static void int8x8_fmul_int32_c(float *dst, const int8_t *src, int scale) · 31 unmodified lines | |||
| 32 | dst[i] = src[i] * fscale; | 32 | dst[i] = src[i] * fscale; |
| 33 | } | 33 | } |
| 34 | 34 | ||
| 35 | static void dca_lfe_fir_c(float *out, const float *in, const float *coefs, | 35 | static inline void |
| 36 | int decifactor, float scale) | 36 | dca_lfe_fir(float *out, const float *in, const float *coefs, |
| 37 | int decifactor, float scale) | ||
| 37 | { | 38 | { |
| 38 | float *out2 = out + decifactor; | 39 | float *out2 = out + decifactor; |
| 39 | const float *cf0 = coefs; | 40 | const float *cf0 = coefs; |
| @@ -82,9 +83,22 @@ static void dca_qmf_32_subbands(float samples_in[32][8], int sb_act, · 42 unmodified lines | |||
| 82 | } | 83 | } |
| 83 | } | 84 | } |
| 84 | 85 | ||
| 86 | static void dca_lfe_fir0_c(float *out, const float *in, const float *coefs, | ||
| 87 | float scale) | ||
| 88 | { | ||
| 89 | dca_lfe_fir(out, in, coefs, 32, scale); | ||
| 90 | } | ||
| 91 | |||
| 92 | static void dca_lfe_fir1_c(float *out, const float *in, const float *coefs, | ||
| 93 | float scale) | ||
| 94 | { | ||
| 95 | dca_lfe_fir(out, in, coefs, 64, scale); | ||
| 96 | } | ||
| 97 | |||
| 85 | av_cold void ff_dcadsp_init(DCADSPContext *s) | 98 | av_cold void ff_dcadsp_init(DCADSPContext *s) |
| 86 | { | 99 | { |
| 87 | s->lfe_fir = dca_lfe_fir_c; | 100 | s->lfe_fir[0] = dca_lfe_fir0_c; |
| 101 | s->lfe_fir[1] = dca_lfe_fir1_c; | ||
| 88 | s->qmf_32_subbands = dca_qmf_32_subbands; | 102 | s->qmf_32_subbands = dca_qmf_32_subbands; |
| 89 | s->int8x8_fmul_int32 = int8x8_fmul_int32_c; | 103 | s->int8x8_fmul_int32 = int8x8_fmul_int32_c; |
| 90 | if (ARCH_ARM) ff_dcadsp_init_arm(s); | 104 | if (ARCH_ARM) ff_dcadsp_init_arm(s); |
| @@ -23,8 +23,8 @@ · 22 unmodified lines | |||
| 23 | #include "synth_filter.h" | 23 | #include "synth_filter.h" |
| 24 | 24 | ||
| 25 | typedef struct DCADSPContext { | 25 | typedef struct DCADSPContext { |
| 26 | void (*lfe_fir)(float *out, const float *in, const float *coefs, | 26 | void (*lfe_fir[2])(float *out, const float *in, const float *coefs, |
| 27 | int decifactor, float scale); | 27 | float scale); |
| 28 | void (*qmf_32_subbands)(float samples_in[32][8], int sb_act, | 28 | void (*qmf_32_subbands)(float samples_in[32][8], int sb_act, |
| 29 | SynthFilterContext *synth, FFTContext *imdct, | 29 | SynthFilterContext *synth, FFTContext *imdct, |
| 30 | float synth_buf_ptr[512], | 30 | float synth_buf_ptr[512], |