x86: Use *_FAST/*_SLOW CPU feature detection macros where appropriate

[ffmpeg.git] / libavcodec / x86 / ac3dsp_init.c

diff --git a/libavcodec/x86/ac3dsp_init.c b/libavcodec/x86/ac3dsp_init.c
index ca10864..89044f4 100644 (file)
--- a/libavcodec/x86/ac3dsp_init.c
+++ b/libavcodec/x86/ac3dsp_init.c
@@ -1,5 +1,5 @@
 /*
 /*

- * x86-optimized AC-3 DSP utils
+ * x86-optimized AC-3 DSP functions

  * Copyright (c) 2011 Justin Ruggles
  *
  * This file is part of Libav.
  * Copyright (c) 2011 Justin Ruggles
  *
  * This file is part of Libav.


@@ -19,10 +19,10 @@
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 

+#include "libavutil/attributes.h"

 #include "libavutil/mem.h"
 #include "libavutil/x86/asm.h"
 #include "libavutil/x86/cpu.h"
 #include "libavutil/mem.h"
 #include "libavutil/x86/asm.h"
 #include "libavutil/x86/cpu.h"

-#include "dsputil_x86.h"

 #include "libavcodec/ac3.h"
 #include "libavcodec/ac3dsp.h"
 
 #include "libavcodec/ac3.h"
 #include "libavcodec/ac3dsp.h"
 


@@ -47,10 +47,22 @@ void ff_float_to_fixed24_sse2 (int32_t *dst, const float *src, unsigned int len)
 
 int ff_ac3_compute_mantissa_size_sse2(uint16_t mant_cnt[6][16]);
 
 
 int ff_ac3_compute_mantissa_size_sse2(uint16_t mant_cnt[6][16]);
 

-void ff_ac3_extract_exponents_3dnow(uint8_t *exp, int32_t *coef, int nb_coefs);

 void ff_ac3_extract_exponents_sse2 (uint8_t *exp, int32_t *coef, int nb_coefs);
 void ff_ac3_extract_exponents_ssse3(uint8_t *exp, int32_t *coef, int nb_coefs);
 
 void ff_ac3_extract_exponents_sse2 (uint8_t *exp, int32_t *coef, int nb_coefs);
 void ff_ac3_extract_exponents_ssse3(uint8_t *exp, int32_t *coef, int nb_coefs);
 

+void ff_apply_window_int16_round_mmxext(int16_t *output, const int16_t *input,
+                                        const int16_t *window, unsigned int len);
+void ff_apply_window_int16_round_sse2(int16_t *output, const int16_t *input,
+                                      const int16_t *window, unsigned int len);
+void ff_apply_window_int16_mmxext(int16_t *output, const int16_t *input,
+                                  const int16_t *window, unsigned int len);
+void ff_apply_window_int16_sse2(int16_t *output, const int16_t *input,
+                                const int16_t *window, unsigned int len);
+void ff_apply_window_int16_ssse3(int16_t *output, const int16_t *input,
+                                 const int16_t *window, unsigned int len);
+void ff_apply_window_int16_ssse3_atom(int16_t *output, const int16_t *input,
+                                      const int16_t *window, unsigned int len);
+

 #if HAVE_SSE_INLINE && HAVE_7REGS
 
 #define IF1(x) x
 #if HAVE_SSE_INLINE && HAVE_7REGS
 
 #define IF1(x) x


@@ -196,6 +208,11 @@ av_cold void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact)
     if (EXTERNAL_MMXEXT(cpu_flags)) {
         c->ac3_exponent_min = ff_ac3_exponent_min_mmxext;
         c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_mmxext;
     if (EXTERNAL_MMXEXT(cpu_flags)) {
         c->ac3_exponent_min = ff_ac3_exponent_min_mmxext;
         c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_mmxext;

+        if (bit_exact) {
+            c->apply_window_int16 = ff_apply_window_int16_mmxext;
+        } else {
+            c->apply_window_int16 = ff_apply_window_int16_round_mmxext;
+        }

     }
     if (EXTERNAL_SSE(cpu_flags)) {
         c->float_to_fixed24 = ff_float_to_fixed24_sse;
     }
     if (EXTERNAL_SSE(cpu_flags)) {
         c->float_to_fixed24 = ff_float_to_fixed24_sse;


@@ -206,15 +223,26 @@ av_cold void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact)
         c->float_to_fixed24 = ff_float_to_fixed24_sse2;
         c->compute_mantissa_size = ff_ac3_compute_mantissa_size_sse2;
         c->extract_exponents = ff_ac3_extract_exponents_sse2;
         c->float_to_fixed24 = ff_float_to_fixed24_sse2;
         c->compute_mantissa_size = ff_ac3_compute_mantissa_size_sse2;
         c->extract_exponents = ff_ac3_extract_exponents_sse2;

-        if (!(cpu_flags & AV_CPU_FLAG_SSE2SLOW)) {
-            c->ac3_lshift_int16 = ff_ac3_lshift_int16_sse2;
-            c->ac3_rshift_int32 = ff_ac3_rshift_int32_sse2;
+        if (bit_exact) {
+            c->apply_window_int16 = ff_apply_window_int16_sse2;

         }
     }
         }
     }

+
+    if (EXTERNAL_SSE2_FAST(cpu_flags)) {
+        c->ac3_lshift_int16 = ff_ac3_lshift_int16_sse2;
+        c->ac3_rshift_int32 = ff_ac3_rshift_int32_sse2;
+        if (!bit_exact) {
+            c->apply_window_int16 = ff_apply_window_int16_round_sse2;
+        }
+    }
+

     if (EXTERNAL_SSSE3(cpu_flags)) {
         c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_ssse3;
     if (EXTERNAL_SSSE3(cpu_flags)) {
         c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_ssse3;

-        if (!(cpu_flags & AV_CPU_FLAG_ATOM)) {
+        if (cpu_flags & AV_CPU_FLAG_ATOM) {
+            c->apply_window_int16 = ff_apply_window_int16_ssse3_atom;
+        } else {

             c->extract_exponents = ff_ac3_extract_exponents_ssse3;
             c->extract_exponents = ff_ac3_extract_exponents_ssse3;

+            c->apply_window_int16 = ff_apply_window_int16_ssse3;

         }
     }
 
         }
     }