libvpx: Pull from upstream

source/config/linux/arm-neon-cpu-detect/vp9_rtcd.h

 #ifndef VP9_RTCD_H_
 #define VP9_RTCD_H_
 
 #ifdef RTCD_C
 #define RTCD_EXTERN
 #else
 #define RTCD_EXTERN extern
 #endif
 
 /*
@@ skipping 15 matching lines @@
 struct yv12_buffer_config;
 
 #ifdef __cplusplus
 extern "C" {
 #endif
 
 unsigned int vp9_avg_4x4_c(const uint8_t *, int p);
 #define vp9_avg_4x4 vp9_avg_4x4_c
 
 unsigned int vp9_avg_8x8_c(const uint8_t *, int p);
-#define vp9_avg_8x8 vp9_avg_8x8_c
+unsigned int vp9_avg_8x8_neon(const uint8_t *, int p);
+RTCD_EXTERN unsigned int (*vp9_avg_8x8)(const uint8_t *, int p);
 
 int64_t vp9_block_error_c(const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz);
 #define vp9_block_error vp9_block_error_c
 
 void vp9_convolve8_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
 void vp9_convolve8_neon(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
 RTCD_EXTERN void (*vp9_convolve8)(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
 
 void vp9_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
 void vp9_convolve8_avg_neon(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
@@ skipping 172 matching lines @@
 
 void vp9_fdct8x8_c(const int16_t *input, tran_low_t *output, int stride);
 void vp9_fdct8x8_neon(const int16_t *input, tran_low_t *output, int stride);
 RTCD_EXTERN void (*vp9_fdct8x8)(const int16_t *input, tran_low_t *output, int stride);
 
 void vp9_fdct8x8_1_c(const int16_t *input, tran_low_t *output, int stride);
 void vp9_fdct8x8_1_neon(const int16_t *input, tran_low_t *output, int stride);
 RTCD_EXTERN void (*vp9_fdct8x8_1)(const int16_t *input, tran_low_t *output, int stride);
 
 void vp9_fdct8x8_quant_c(const int16_t *input, int stride, tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan);
-#define vp9_fdct8x8_quant vp9_fdct8x8_quant_c
+void vp9_fdct8x8_quant_neon(const int16_t *input, int stride, tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan);
+RTCD_EXTERN void (*vp9_fdct8x8_quant)(const int16_t *input, int stride, tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan);
 
 void vp9_fht16x16_c(const int16_t *input, tran_low_t *output, int stride, int tx_type);
 #define vp9_fht16x16 vp9_fht16x16_c
 
 void vp9_fht4x4_c(const int16_t *input, tran_low_t *output, int stride, int tx_type);
 #define vp9_fht4x4 vp9_fht4x4_c
 
 void vp9_fht8x8_c(const int16_t *input, tran_low_t *output, int stride, int tx_type);
 #define vp9_fht8x8 vp9_fht8x8_c
 
@@ skipping 81 matching lines @@
 #define vp9_iht16x16_256_add vp9_iht16x16_256_add_c
 
 void vp9_iht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type);
 void vp9_iht4x4_16_add_neon(const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type);
 RTCD_EXTERN void (*vp9_iht4x4_16_add)(const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type);
 
 void vp9_iht8x8_64_add_c(const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type);
 void vp9_iht8x8_64_add_neon(const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type);
 RTCD_EXTERN void (*vp9_iht8x8_64_add)(const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type);
 
+int16_t vp9_int_pro_col_c(uint8_t const *ref, const int width);
+#define vp9_int_pro_col vp9_int_pro_col_c
+
+void vp9_int_pro_row_c(int16_t *hbuf, uint8_t const *ref, const int ref_stride, const int height);
+#define vp9_int_pro_row vp9_int_pro_row_c
+
 void vp9_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int dest_stride);
 #define vp9_iwht4x4_16_add vp9_iwht4x4_16_add_c
 
 void vp9_iwht4x4_1_add_c(const tran_low_t *input, uint8_t *dest, int dest_stride);
 #define vp9_iwht4x4_1_add vp9_iwht4x4_1_add_c
 
 void vp9_lpf_horizontal_16_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
 void vp9_lpf_horizontal_16_neon(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
 RTCD_EXTERN void (*vp9_lpf_horizontal_16)(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
 
@@ skipping 66 matching lines @@
 unsigned int vp9_sad16x16_neon(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride);
 RTCD_EXTERN unsigned int (*vp9_sad16x16)(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride);
 
 unsigned int vp9_sad16x16_avg_c(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, const uint8_t *second_pred);
 #define vp9_sad16x16_avg vp9_sad16x16_avg_c
 
 void vp9_sad16x16x3_c(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int *sad_array);
 #define vp9_sad16x16x3 vp9_sad16x16x3_c
 
 void vp9_sad16x16x4d_c(const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array);
-#define vp9_sad16x16x4d vp9_sad16x16x4d_c
+void vp9_sad16x16x4d_neon(const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array);
+RTCD_EXTERN void (*vp9_sad16x16x4d)(const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array);
 
 void vp9_sad16x16x8_c(const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int  ref_stride, uint32_t *sad_array);
 #define vp9_sad16x16x8 vp9_sad16x16x8_c
 
 unsigned int vp9_sad16x32_c(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride);
 #define vp9_sad16x32 vp9_sad16x32_c
 
 unsigned int vp9_sad16x32_avg_c(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred);
 #define vp9_sad16x32_avg vp9_sad16x32_avg_c
 
@@ skipping 28 matching lines @@
 unsigned int vp9_sad32x32_neon(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride);
 RTCD_EXTERN unsigned int (*vp9_sad32x32)(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride);
 
 unsigned int vp9_sad32x32_avg_c(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, const uint8_t *second_pred);
 #define vp9_sad32x32_avg vp9_sad32x32_avg_c
 
 void vp9_sad32x32x3_c(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int *sad_array);
 #define vp9_sad32x32x3 vp9_sad32x32x3_c
 
 void vp9_sad32x32x4d_c(const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array);
-#define vp9_sad32x32x4d vp9_sad32x32x4d_c
+void vp9_sad32x32x4d_neon(const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array);
+RTCD_EXTERN void (*vp9_sad32x32x4d)(const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array);
 
 void vp9_sad32x32x8_c(const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int  ref_stride, uint32_t *sad_array);
 #define vp9_sad32x32x8 vp9_sad32x32x8_c
 
 unsigned int vp9_sad32x64_c(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride);
 #define vp9_sad32x64 vp9_sad32x64_c
 
 unsigned int vp9_sad32x64_avg_c(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred);
 #define vp9_sad32x64_avg vp9_sad32x64_avg_c
 
@@ skipping 40 matching lines @@
 unsigned int vp9_sad64x64_neon(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride);
 RTCD_EXTERN unsigned int (*vp9_sad64x64)(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride);
 
 unsigned int vp9_sad64x64_avg_c(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, const uint8_t *second_pred);
 #define vp9_sad64x64_avg vp9_sad64x64_avg_c
 
 void vp9_sad64x64x3_c(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int *sad_array);
 #define vp9_sad64x64x3 vp9_sad64x64x3_c
 
 void vp9_sad64x64x4d_c(const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array);
-#define vp9_sad64x64x4d vp9_sad64x64x4d_c
+void vp9_sad64x64x4d_neon(const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array);
+RTCD_EXTERN void (*vp9_sad64x64x4d)(const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array);
 
 void vp9_sad64x64x8_c(const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int  ref_stride, uint32_t *sad_array);
 #define vp9_sad64x64x8 vp9_sad64x64x8_c
 
 unsigned int vp9_sad8x16_c(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride);
 #define vp9_sad8x16 vp9_sad8x16_c
 
 unsigned int vp9_sad8x16_avg_c(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, const uint8_t *second_pred);
 #define vp9_sad8x16_avg vp9_sad8x16_avg_c
 
@@ skipping 193 matching lines @@
 unsigned int vp9_variance8x16_c(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse);
 #define vp9_variance8x16 vp9_variance8x16_c
 
 unsigned int vp9_variance8x4_c(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse);
 #define vp9_variance8x4 vp9_variance8x4_c
 
 unsigned int vp9_variance8x8_c(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse);
 unsigned int vp9_variance8x8_neon(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse);
 RTCD_EXTERN unsigned int (*vp9_variance8x8)(const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse);
 
+int vp9_vector_sad_c(int16_t const *ref, int16_t const *src, const int width);
+#define vp9_vector_sad vp9_vector_sad_c
+
 void vp9_rtcd(void);
 
 #include "vpx_config.h"
 
 #ifdef RTCD_C
 #include "vpx_ports/arm.h"
 static void setup_rtcd_internal(void)
 {
     int flags = arm_cpu_caps();
 
     (void)flags;
 
+    vp9_avg_8x8 = vp9_avg_8x8_c;
+    if (flags & HAS_NEON) vp9_avg_8x8 = vp9_avg_8x8_neon;
     vp9_convolve8 = vp9_convolve8_c;
     if (flags & HAS_NEON) vp9_convolve8 = vp9_convolve8_neon;
     vp9_convolve8_avg = vp9_convolve8_avg_c;
     if (flags & HAS_NEON) vp9_convolve8_avg = vp9_convolve8_avg_neon;
     vp9_convolve8_avg_horiz = vp9_convolve8_avg_horiz_c;
     if (flags & HAS_NEON) vp9_convolve8_avg_horiz = vp9_convolve8_avg_horiz_neon;
     vp9_convolve8_avg_vert = vp9_convolve8_avg_vert_c;
     if (flags & HAS_NEON) vp9_convolve8_avg_vert = vp9_convolve8_avg_vert_neon;
     vp9_convolve8_horiz = vp9_convolve8_horiz_c;
     if (flags & HAS_NEON) vp9_convolve8_horiz = vp9_convolve8_horiz_neon;
     vp9_convolve8_vert = vp9_convolve8_vert_c;
     if (flags & HAS_NEON) vp9_convolve8_vert = vp9_convolve8_vert_neon;
     vp9_convolve_avg = vp9_convolve_avg_c;
     if (flags & HAS_NEON) vp9_convolve_avg = vp9_convolve_avg_neon;
     vp9_convolve_copy = vp9_convolve_copy_c;
     if (flags & HAS_NEON) vp9_convolve_copy = vp9_convolve_copy_neon;
     vp9_fdct8x8 = vp9_fdct8x8_c;
     if (flags & HAS_NEON) vp9_fdct8x8 = vp9_fdct8x8_neon;
     vp9_fdct8x8_1 = vp9_fdct8x8_1_c;
     if (flags & HAS_NEON) vp9_fdct8x8_1 = vp9_fdct8x8_1_neon;
+    vp9_fdct8x8_quant = vp9_fdct8x8_quant_c;
+    if (flags & HAS_NEON) vp9_fdct8x8_quant = vp9_fdct8x8_quant_neon;
     vp9_get16x16var = vp9_get16x16var_c;
     if (flags & HAS_NEON) vp9_get16x16var = vp9_get16x16var_neon;
     vp9_get8x8var = vp9_get8x8var_c;
     if (flags & HAS_NEON) vp9_get8x8var = vp9_get8x8var_neon;
     vp9_h_predictor_16x16 = vp9_h_predictor_16x16_c;
     if (flags & HAS_NEON) vp9_h_predictor_16x16 = vp9_h_predictor_16x16_neon;
     vp9_h_predictor_32x32 = vp9_h_predictor_32x32_c;
     if (flags & HAS_NEON) vp9_h_predictor_32x32 = vp9_h_predictor_32x32_neon;
     vp9_h_predictor_4x4 = vp9_h_predictor_4x4_c;
     if (flags & HAS_NEON) vp9_h_predictor_4x4 = vp9_h_predictor_4x4_neon;
@@ skipping 44 matching lines @@
     vp9_lpf_vertical_4_dual = vp9_lpf_vertical_4_dual_c;
     if (flags & HAS_NEON) vp9_lpf_vertical_4_dual = vp9_lpf_vertical_4_dual_neon;
     vp9_lpf_vertical_8 = vp9_lpf_vertical_8_c;
     if (flags & HAS_NEON) vp9_lpf_vertical_8 = vp9_lpf_vertical_8_neon;
     vp9_lpf_vertical_8_dual = vp9_lpf_vertical_8_dual_c;
     if (flags & HAS_NEON) vp9_lpf_vertical_8_dual = vp9_lpf_vertical_8_dual_neon;
     vp9_quantize_fp = vp9_quantize_fp_c;
     if (flags & HAS_NEON) vp9_quantize_fp = vp9_quantize_fp_neon;
     vp9_sad16x16 = vp9_sad16x16_c;
     if (flags & HAS_NEON) vp9_sad16x16 = vp9_sad16x16_neon;
+    vp9_sad16x16x4d = vp9_sad16x16x4d_c;
+    if (flags & HAS_NEON) vp9_sad16x16x4d = vp9_sad16x16x4d_neon;
     vp9_sad32x32 = vp9_sad32x32_c;
     if (flags & HAS_NEON) vp9_sad32x32 = vp9_sad32x32_neon;
+    vp9_sad32x32x4d = vp9_sad32x32x4d_c;
+    if (flags & HAS_NEON) vp9_sad32x32x4d = vp9_sad32x32x4d_neon;
     vp9_sad64x64 = vp9_sad64x64_c;
     if (flags & HAS_NEON) vp9_sad64x64 = vp9_sad64x64_neon;
+    vp9_sad64x64x4d = vp9_sad64x64x4d_c;
+    if (flags & HAS_NEON) vp9_sad64x64x4d = vp9_sad64x64x4d_neon;
     vp9_sad8x8 = vp9_sad8x8_c;
     if (flags & HAS_NEON) vp9_sad8x8 = vp9_sad8x8_neon;
     vp9_sub_pixel_variance16x16 = vp9_sub_pixel_variance16x16_c;
     if (flags & HAS_NEON) vp9_sub_pixel_variance16x16 = vp9_sub_pixel_variance16x16_neon;
     vp9_sub_pixel_variance32x32 = vp9_sub_pixel_variance32x32_c;
     if (flags & HAS_NEON) vp9_sub_pixel_variance32x32 = vp9_sub_pixel_variance32x32_neon;
     vp9_sub_pixel_variance64x64 = vp9_sub_pixel_variance64x64_c;
     if (flags & HAS_NEON) vp9_sub_pixel_variance64x64 = vp9_sub_pixel_variance64x64_neon;
     vp9_sub_pixel_variance8x8 = vp9_sub_pixel_variance8x8_c;
     if (flags & HAS_NEON) vp9_sub_pixel_variance8x8 = vp9_sub_pixel_variance8x8_neon;
@@ skipping 28 matching lines @@
     vp9_variance8x8 = vp9_variance8x8_c;
     if (flags & HAS_NEON) vp9_variance8x8 = vp9_variance8x8_neon;
 }
 #endif
 
 #ifdef __cplusplus
 }  // extern "C"
 #endif
 
 #endif