x86/float_dsp: add ff_vector_dmul_{sse2,avx}

~3x to 5x faster.
Signed-off-by: James Almer <jamrial@gmail.com>

libavutil/x86/float_dsp.asm

@@ -58,6 +58,39 @@ INIT_YMM avx
 VECTOR_FMUL
 %endif
 
+;-----------------------------------------------------------------------------
+; void vector_dmul(double *dst, const double *src0, const double *src1, int len)
+;-----------------------------------------------------------------------------
+%macro VECTOR_DMUL 0
+cglobal vector_dmul, 4,4,4, dst, src0, src1, len
+    lea       lend, [lenq*8 - mmsize*4]
+ALIGN 16
+.loop:
+    movaps    m0,     [src0q + lenq + 0*mmsize]
+    movaps    m1,     [src0q + lenq + 1*mmsize]
+    movaps    m2,     [src0q + lenq + 2*mmsize]
+    movaps    m3,     [src0q + lenq + 3*mmsize]
+    mulpd     m0, m0, [src1q + lenq + 0*mmsize]
+    mulpd     m1, m1, [src1q + lenq + 1*mmsize]
+    mulpd     m2, m2, [src1q + lenq + 2*mmsize]
+    mulpd     m3, m3, [src1q + lenq + 3*mmsize]
+    movaps    [dstq + lenq + 0*mmsize], m0
+    movaps    [dstq + lenq + 1*mmsize], m1
+    movaps    [dstq + lenq + 2*mmsize], m2
+    movaps    [dstq + lenq + 3*mmsize], m3
+
+    sub       lenq, mmsize*4
+    jge       .loop
+    RET
+%endmacro
+
+INIT_XMM sse2
+VECTOR_DMUL
+%if HAVE_AVX_EXTERNAL
+INIT_YMM avx
+VECTOR_DMUL
+%endif
+
 ;------------------------------------------------------------------------------
 ; void ff_vector_fmac_scalar(float *dst, const float *src, float mul, int len)
 ;------------------------------------------------------------------------------

libavutil/x86/float_dsp_init.c

@@ -29,6 +29,11 @@ void ff_vector_fmul_sse(float *dst, const float *src0, const float *src1,
 void ff_vector_fmul_avx(float *dst, const float *src0, const float *src1,
                         int len);
 
+void ff_vector_dmul_sse2(double *dst, const double *src0, const double *src1,
+                         int len);
+void ff_vector_dmul_avx(double *dst, const double *src0, const double *src1,
+                        int len);
+
 void ff_vector_fmac_scalar_sse(float *dst, const float *src, float mul,
                                int len);
 void ff_vector_fmac_scalar_avx(float *dst, const float *src, float mul,
@@ -92,11 +97,13 @@ av_cold void ff_float_dsp_init_x86(AVFloatDSPContext *fdsp)
         fdsp->butterflies_float   = ff_butterflies_float_sse;
     }
     if (EXTERNAL_SSE2(cpu_flags)) {
+        fdsp->vector_dmul = ff_vector_dmul_sse2;
         fdsp->vector_dmac_scalar = ff_vector_dmac_scalar_sse2;
         fdsp->vector_dmul_scalar = ff_vector_dmul_scalar_sse2;
     }
     if (EXTERNAL_AVX_FAST(cpu_flags)) {
         fdsp->vector_fmul = ff_vector_fmul_avx;
+        fdsp->vector_dmul = ff_vector_dmul_avx;
         fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_avx;
         fdsp->vector_dmul_scalar = ff_vector_dmul_scalar_avx;
         fdsp->vector_dmac_scalar = ff_vector_dmac_scalar_avx;