x86: synth filter float: implement SSE2 version

Timings for Arrandale:
          C    SSE
win32:  2108   334
win64:  1152   322

Factorizing the inner loop with a call/jmp is a >15 cycles cost, even with
the jmp destination being aligned.

Unrolling for ARCH_X86_64 is a 20 cycles gain.

Signed-off-by: Janne Grunau <janne-libav@jannau.net>

1 files changed, 28 insertions, 0 deletions

diff --git a/libavcodec/x86/dcadsp_init.c b/libavcodec/x86/dcadsp_init.c
index c234bd29f5..5f6e8c5a19 100644
--- a/libavcodec/x86/dcadsp_init.c
+++ b/libavcodec/x86/dcadsp_init.c
@@ -49,3 +49,31 @@ av_cold void ff_dcadsp_init_x86(DCADSPContext *s)
         s->int8x8_fmul_int32 = ff_int8x8_fmul_int32_sse4;
     }
 }
+
+void ff_synth_filter_inner_sse2(float *synth_buf_ptr, float synth_buf2[32],
+                                const float window[512],
+                                float out[32], intptr_t offset, float scale);
+
+static void synth_filter_sse2(FFTContext *imdct,
+                              float *synth_buf_ptr, int *synth_buf_offset,
+                              float synth_buf2[32], const float window[512],
+                              float out[32], const float in[32], float scale)
+{
+    float *synth_buf= synth_buf_ptr + *synth_buf_offset;
+
+    imdct->imdct_half(imdct, synth_buf, in);
+
+    ff_synth_filter_inner_sse2(synth_buf, synth_buf2, window,
+                               out, *synth_buf_offset, scale);
+
+    *synth_buf_offset = (*synth_buf_offset - 32) & 511;
+}
+
+av_cold void ff_synth_filter_init_x86(SynthFilterContext *s)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (EXTERNAL_SSE2(cpu_flags)) {
+        s->synth_filter_float = synth_filter_sse2;
+    }
+}