From ac7eb4cb20ea84cfc911794722695d501b354ee9 Mon Sep 17 00:00:00 2001 From: Justin Ruggles Date: Mon, 24 Sep 2012 15:00:53 -0400 Subject: [PATCH 1/3] float_dsp: add vector_dmul_scalar() to multiply a vector of doubles Include x86-optimized versions for SSE2 and AVX. --- libavutil/float_dsp.c | 9 +++++++ libavutil/float_dsp.h | 15 ++++++++++++ libavutil/x86/float_dsp.asm | 45 ++++++++++++++++++++++++++++++++++ libavutil/x86/float_dsp_init.c | 9 +++++++ libavutil/x86/x86util.asm | 11 +++++++++ 5 files changed, 89 insertions(+) diff --git a/libavutil/float_dsp.c b/libavutil/float_dsp.c index b6b11818b5..22139defe4 100644 --- a/libavutil/float_dsp.c +++ b/libavutil/float_dsp.c @@ -44,11 +44,20 @@ static void vector_fmul_scalar_c(float *dst, const float *src, float mul, dst[i] = src[i] * mul; } +static void vector_dmul_scalar_c(double *dst, const double *src, double mul, + int len) +{ + int i; + for (i = 0; i < len; i++) + dst[i] = src[i] * mul; +} + void avpriv_float_dsp_init(AVFloatDSPContext *fdsp, int bit_exact) { fdsp->vector_fmul = vector_fmul_c; fdsp->vector_fmac_scalar = vector_fmac_scalar_c; fdsp->vector_fmul_scalar = vector_fmul_scalar_c; + fdsp->vector_dmul_scalar = vector_dmul_scalar_c; #if ARCH_ARM ff_float_dsp_init_arm(fdsp); diff --git a/libavutil/float_dsp.h b/libavutil/float_dsp.h index cb4b28f0e2..41b73c5b26 100644 --- a/libavutil/float_dsp.h +++ b/libavutil/float_dsp.h @@ -66,6 +66,21 @@ typedef struct AVFloatDSPContext { */ void (*vector_fmul_scalar)(float *dst, const float *src, float mul, int len); + + /** + * Multiply a vector of double by a scalar double. Source and + * destination vectors must overlap exactly or not at all. + * + * @param dst result vector + * constraints: 32-byte aligned + * @param src input vector + * constraints: 32-byte aligned + * @param mul scalar value + * @param len length of vector + * constraints: multiple of 8 + */ + void (*vector_dmul_scalar)(double *dst, const double *src, double mul, + int len); } AVFloatDSPContext; /** diff --git a/libavutil/x86/float_dsp.asm b/libavutil/x86/float_dsp.asm index 317df9c3c1..d8fd93a625 100644 --- a/libavutil/x86/float_dsp.asm +++ b/libavutil/x86/float_dsp.asm @@ -114,3 +114,48 @@ cglobal vector_fmul_scalar, 4,4,3, dst, src, mul, len INIT_XMM sse VECTOR_FMUL_SCALAR + +;------------------------------------------------------------------------------ +; void ff_vector_dmul_scalar(double *dst, const double *src, double mul, +; int len) +;------------------------------------------------------------------------------ + +%macro VECTOR_DMUL_SCALAR 0 +%if UNIX64 +cglobal vector_dmul_scalar, 3,3,3, dst, src, len +%else +cglobal vector_dmul_scalar, 4,4,3, dst, src, mul, len +%endif +%if ARCH_X86_32 + VBROADCASTSD xmm0, mulm +%else +%if WIN64 + movlhps xmm2, xmm2 +%if cpuflag(avx) + vinsertf128 ymm2, ymm2, xmm2, 1 +%endif + SWAP 0, 2 +%else + movlhps xmm0, xmm0 +%if cpuflag(avx) + vinsertf128 ymm0, ymm0, xmm0, 1 +%endif +%endif +%endif + lea lenq, [lend*8-2*mmsize] +.loop: + mulpd m1, m0, [srcq+lenq ] + mulpd m2, m0, [srcq+lenq+mmsize] + mova [dstq+lenq ], m1 + mova [dstq+lenq+mmsize], m2 + sub lenq, 2*mmsize + jge .loop + REP_RET +%endmacro + +INIT_XMM sse2 +VECTOR_DMUL_SCALAR +%if HAVE_AVX_EXTERNAL +INIT_YMM avx +VECTOR_DMUL_SCALAR +%endif diff --git a/libavutil/x86/float_dsp_init.c b/libavutil/x86/float_dsp_init.c index d14ec6a377..b3b7ff4c3a 100644 --- a/libavutil/x86/float_dsp_init.c +++ b/libavutil/x86/float_dsp_init.c @@ -35,6 +35,11 @@ extern void ff_vector_fmac_scalar_avx(float *dst, const float *src, float mul, extern void ff_vector_fmul_scalar_sse(float *dst, const float *src, float mul, int len); +extern void ff_vector_dmul_scalar_sse2(double *dst, const double *src, + double mul, int len); +extern void ff_vector_dmul_scalar_avx(double *dst, const double *src, + double mul, int len); + void ff_float_dsp_init_x86(AVFloatDSPContext *fdsp) { int mm_flags = av_get_cpu_flags(); @@ -44,8 +49,12 @@ void ff_float_dsp_init_x86(AVFloatDSPContext *fdsp) fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_sse; fdsp->vector_fmul_scalar = ff_vector_fmul_scalar_sse; } + if (EXTERNAL_SSE2(mm_flags)) { + fdsp->vector_dmul_scalar = ff_vector_dmul_scalar_sse2; + } if (EXTERNAL_AVX(mm_flags)) { fdsp->vector_fmul = ff_vector_fmul_avx; fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_avx; + fdsp->vector_dmul_scalar = ff_vector_dmul_scalar_avx; } } diff --git a/libavutil/x86/x86util.asm b/libavutil/x86/x86util.asm index d3c0d86056..16ee6cfe94 100644 --- a/libavutil/x86/x86util.asm +++ b/libavutil/x86/x86util.asm @@ -631,6 +631,17 @@ %endif %endmacro +%macro VBROADCASTSD 2 ; dst xmm/ymm, src m64 +%if cpuflag(avx) && mmsize == 32 + vbroadcastsd %1, %2 +%elif cpuflag(sse3) + movddup %1, %2 +%else ; sse2 + movsd %1, %2 + movlhps %1, %1 +%endif +%endmacro + %macro SHUFFLE_MASK_W 8 %rep 8 %if %1>=0x80 From 5312268b34dc948e87934ae665a79a57c2fe2648 Mon Sep 17 00:00:00 2001 From: Justin Ruggles Date: Fri, 28 Sep 2012 16:36:58 -0400 Subject: [PATCH 2/3] eval: treat dB as decibels instead of decibytes --- libavutil/eval.c | 27 ++++++++++++++++++++++++--- tests/ref/fate/eval | 6 ++++++ 2 files changed, 30 insertions(+), 3 deletions(-) diff --git a/libavutil/eval.c b/libavutil/eval.c index 7887cf6b0e..a0e968dc34 100644 --- a/libavutil/eval.c +++ b/libavutil/eval.c @@ -81,7 +81,11 @@ double av_strtod(const char *numstr, char **tail) d = strtod(numstr, &next); /* if parsing succeeded, check for and interpret postfixes */ if (next!=numstr) { - if (*next >= 'E' && *next <= 'z') { + if (next[0] == 'd' && next[1] == 'B') { + /* treat dB as decibels instead of decibytes */ + d = pow(10, d / 20); + next += 2; + } else if (*next >= 'E' && *next <= 'z') { int e= si_prefixes[*next - 'E']; if (e) { if (next[1] == 'i') { @@ -339,16 +343,31 @@ static int parse_pow(AVExpr **e, Parser *p, int *sign) return parse_primary(e, p); } +static int parse_dB(AVExpr **e, Parser *p, int *sign) +{ + /* do not filter out the negative sign when parsing a dB value. + for example, -3dB is not the same as -(3dB) */ + if (*p->s == '-') { + char *next; + strtod(p->s, &next); + if (next != p->s && next[0] == 'd' && next[1] == 'B') { + *sign = 0; + return parse_primary(e, p); + } + } + return parse_pow(e, p, sign); +} + static int parse_factor(AVExpr **e, Parser *p) { int sign, sign2, ret; AVExpr *e0, *e1, *e2; - if ((ret = parse_pow(&e0, p, &sign)) < 0) + if ((ret = parse_dB(&e0, p, &sign)) < 0) return ret; while(p->s[0]=='^'){ e1 = e0; p->s++; - if ((ret = parse_pow(&e2, p, &sign2)) < 0) { + if ((ret = parse_dB(&e2, p, &sign2)) < 0) { av_expr_free(e1); return ret; } @@ -629,6 +648,8 @@ int main(int argc, char **argv) "not(1)", "not(NAN)", "not(0)", + "6.0206dB", + "-3.0103dB", NULL }; diff --git a/tests/ref/fate/eval b/tests/ref/fate/eval index dfc5ae463d..b01e5a5718 100644 --- a/tests/ref/fate/eval +++ b/tests/ref/fate/eval @@ -184,5 +184,11 @@ Evaluating 'not(NAN)' Evaluating 'not(0)' 'not(0)' -> 1.000000 +Evaluating '6.0206dB' +'6.0206dB' -> 2.000000 + +Evaluating '-3.0103dB' +'-3.0103dB' -> 0.707107 + 12.700000 == 12.7 0.931323 == 0.931322575 From 9d5c62ba5b586c80af508b5914934b1c439f6652 Mon Sep 17 00:00:00 2001 From: Justin Ruggles Date: Sat, 29 Sep 2012 00:34:28 -0400 Subject: [PATCH 3/3] lavu/opt: do not filter out the initial sign character except for flags This allows parsing of special-case negative numbers like decibels. --- libavutil/opt.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-) diff --git a/libavutil/opt.c b/libavutil/opt.c index d2cb9ef95a..e199170075 100644 --- a/libavutil/opt.c +++ b/libavutil/opt.c @@ -154,10 +154,15 @@ static int set_string_number(void *obj, const AVOption *o, const char *val, void double d, num = 1; int64_t intnum = 1; - if (*val == '+' || *val == '-') - cmd = *(val++); + i = 0; + if (*val == '+' || *val == '-') { + if (o->type == AV_OPT_TYPE_FLAGS) + cmd = *(val++); + else if (!notfirst) + buf[i++] = *val; + } - for (i = 0; i < sizeof(buf) - 1 && val[i] && val[i] != '+' && val[i] != '-'; i++) + for (; i < sizeof(buf) - 1 && val[i] && val[i] != '+' && val[i] != '-'; i++) buf[i] = val[i]; buf[i] = 0;