Use common transpose for vp9_iht8x8_add_neon

Change-Id: I354b5d22130d76b0eceda0748db1f871f58fa372

diff --git a/vp9/common/arm/neon/vp9_iht8x8_add_neon.c b/vp9/common/arm/neon/vp9_iht8x8_add_neon.c
index 39bed46..0e3febf 100644 (file)
--- a/vp9/common/arm/neon/vp9_iht8x8_add_neon.c
+++ b/vp9/common/arm/neon/vp9_iht8x8_add_neon.c
@@ -14,6 +14,7 @@
 #include "./vp9_rtcd.h"
 #include "./vpx_config.h"
 #include "vp9/common/vp9_common.h"
+#include "vpx_dsp/arm/transpose_neon.h"
 
 static int16_t cospi_2_64 = 16305;
 static int16_t cospi_4_64 = 16069;
@@ -31,70 +32,6 @@ static int16_t cospi_26_64 = 4756;
 static int16_t cospi_28_64 = 3196;
 static int16_t cospi_30_64 = 1606;
 
-static INLINE void TRANSPOSE8X8(int16x8_t *q8s16, int16x8_t *q9s16,
-                                int16x8_t *q10s16, int16x8_t *q11s16,
-                                int16x8_t *q12s16, int16x8_t *q13s16,
-                                int16x8_t *q14s16, int16x8_t *q15s16) {
-  int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16;
-  int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16;
-  int32x4x2_t q0x2s32, q1x2s32, q2x2s32, q3x2s32;
-  int16x8x2_t q0x2s16, q1x2s16, q2x2s16, q3x2s16;
-
-  d16s16 = vget_low_s16(*q8s16);
-  d17s16 = vget_high_s16(*q8s16);
-  d18s16 = vget_low_s16(*q9s16);
-  d19s16 = vget_high_s16(*q9s16);
-  d20s16 = vget_low_s16(*q10s16);
-  d21s16 = vget_high_s16(*q10s16);
-  d22s16 = vget_low_s16(*q11s16);
-  d23s16 = vget_high_s16(*q11s16);
-  d24s16 = vget_low_s16(*q12s16);
-  d25s16 = vget_high_s16(*q12s16);
-  d26s16 = vget_low_s16(*q13s16);
-  d27s16 = vget_high_s16(*q13s16);
-  d28s16 = vget_low_s16(*q14s16);
-  d29s16 = vget_high_s16(*q14s16);
-  d30s16 = vget_low_s16(*q15s16);
-  d31s16 = vget_high_s16(*q15s16);
-
-  *q8s16 = vcombine_s16(d16s16, d24s16);   // vswp d17, d24
-  *q9s16 = vcombine_s16(d18s16, d26s16);   // vswp d19, d26
-  *q10s16 = vcombine_s16(d20s16, d28s16);  // vswp d21, d28
-  *q11s16 = vcombine_s16(d22s16, d30s16);  // vswp d23, d30
-  *q12s16 = vcombine_s16(d17s16, d25s16);
-  *q13s16 = vcombine_s16(d19s16, d27s16);
-  *q14s16 = vcombine_s16(d21s16, d29s16);
-  *q15s16 = vcombine_s16(d23s16, d31s16);
-
-  q0x2s32 =
-      vtrnq_s32(vreinterpretq_s32_s16(*q8s16), vreinterpretq_s32_s16(*q10s16));
-  q1x2s32 =
-      vtrnq_s32(vreinterpretq_s32_s16(*q9s16), vreinterpretq_s32_s16(*q11s16));
-  q2x2s32 =
-      vtrnq_s32(vreinterpretq_s32_s16(*q12s16), vreinterpretq_s32_s16(*q14s16));
-  q3x2s32 =
-      vtrnq_s32(vreinterpretq_s32_s16(*q13s16), vreinterpretq_s32_s16(*q15s16));
-
-  q0x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[0]),   // q8
-                      vreinterpretq_s16_s32(q1x2s32.val[0]));  // q9
-  q1x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[1]),   // q10
-                      vreinterpretq_s16_s32(q1x2s32.val[1]));  // q11
-  q2x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[0]),   // q12
-                      vreinterpretq_s16_s32(q3x2s32.val[0]));  // q13
-  q3x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[1]),   // q14
-                      vreinterpretq_s16_s32(q3x2s32.val[1]));  // q15
-
-  *q8s16 = q0x2s16.val[0];
-  *q9s16 = q0x2s16.val[1];
-  *q10s16 = q1x2s16.val[0];
-  *q11s16 = q1x2s16.val[1];
-  *q12s16 = q2x2s16.val[0];
-  *q13s16 = q2x2s16.val[1];
-  *q14s16 = q3x2s16.val[0];
-  *q15s16 = q3x2s16.val[1];
-  return;
-}
-
 static INLINE void IDCT8x8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
                               int16x8_t *q10s16, int16x8_t *q11s16,
                               int16x8_t *q12s16, int16x8_t *q13s16,
@@ -489,8 +426,8 @@ void vp9_iht8x8_64_add_neon(const tran_low_t *input, uint8_t *dest,
   q14s16 = vld1q_s16(input + 8 * 6);
   q15s16 = vld1q_s16(input + 8 * 7);
 
-  TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
-               &q15s16);
+  transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
+                    &q15s16);
 
   switch (tx_type) {
     case 0:  // idct_idct is not supported. Fall back to C
@@ -506,8 +443,8 @@ void vp9_iht8x8_64_add_neon(const tran_low_t *input, uint8_t *dest,
                  &q15s16);
 
       // transpose the matrix
-      TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
-                   &q15s16);
+      transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16,
+                        &q14s16, &q15s16);
 
       // generate IADST constants
       // GENERATE_IADST_CONSTANTS
@@ -525,8 +462,8 @@ void vp9_iht8x8_64_add_neon(const tran_low_t *input, uint8_t *dest,
                   &q15s16);
 
       // transpose the matrix
-      TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
-                   &q15s16);
+      transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16,
+                        &q14s16, &q15s16);
 
       // generate IDCT constants
       // GENERATE_IDCT_CONSTANTS
@@ -544,8 +481,8 @@ void vp9_iht8x8_64_add_neon(const tran_low_t *input, uint8_t *dest,
                   &q15s16);
 
       // transpose the matrix
-      TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
-                   &q15s16);
+      transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16,
+                        &q14s16, &q15s16);
 
       // then transform columns
       IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,