Merge commit '9d5c62ba5b586c80af508b5914934b1c439f6652'

* commit '9d5c62ba5b586c80af508b5914934b1c439f6652':
  lavu/opt: do not filter out the initial sign character except for flags
  eval: treat dB as decibels instead of decibytes
  float_dsp: add vector_dmul_scalar() to multiply a vector of doubles

Conflicts:
	libavutil/eval.c
	tests/ref/fate/eval

Merged-by: Michael Niedermayer <michaelni@gmx.at>

libavutil/eval.c

@@ -94,7 +94,11 @@ 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') {
@@ -448,16 +452,31 @@ 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;
         }
@@ -744,6 +763,8 @@ int main(int argc, char **argv)
         "not(1)",
         "not(NAN)",
         "not(0)",
+        "6.0206dB",
+        "-3.0103dB",
         "pow(0,1.23)",
         "pow(PI,1.23)",
         "PI^1.23",

libavutil/float_dsp.c

@@ -47,11 +47,20 @@ 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);

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;
 
 /**

libavutil/opt.c

@@ -184,10 +184,15 @@ 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 == '-')
+        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;
 

libavutil/x86/float_dsp.asm

@@ -120,3 +120,48 @@ 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 m0, 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

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;
     }
 }

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

tests/ref/fate/eval

@@ -184,6 +184,12 @@ Evaluating 'not(NAN)'
 Evaluating 'not(0)'
 'not(0)' -> 1.000000
 
+Evaluating '6.0206dB'
+'6.0206dB' -> 2.000000
+
+Evaluating '-3.0103dB'
+'-3.0103dB' -> 0.707107
+
 Evaluating 'pow(0,1.23)'
 'pow(0,1.23)' -> 0.000000