Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(10)

Issues Search
    My Issues | Starred     Open | Closed | All

Unified Diff: source/libvpx/vp9/common/vp9_idct.c

Issue 54923004: libvpx: Pull from upstream (Closed) Base URL: svn://svn.chromium.org/chrome/trunk/deps/third_party/libvpx/
Patch Set: Created 7 years, 2 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View side-by-side diff with in-line comments
Download patch
« no previous file with comments | « source/libvpx/vp9/common/vp9_idct.h ('k') | source/libvpx/vp9/common/vp9_loopfilter.h » ('j') | no next file with comments »
Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
Index: source/libvpx/vp9/common/vp9_idct.c
===================================================================
--- source/libvpx/vp9/common/vp9_idct.c (revision 232232)
+++ source/libvpx/vp9/common/vp9_idct.c (working copy)
@@ -18,20 +18,20 @@
#include "vp9/common/vp9_common.h"
#include "vp9/common/vp9_idct.h"
-void vp9_short_iwalsh4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+void vp9_iwht4x4_16_add_c(const int16_t *input, uint8_t *dest, int stride) {
/* 4-point reversible, orthonormal inverse Walsh-Hadamard in 3.5 adds,
0.5 shifts per pixel. */
int i;
int16_t output[16];
int a1, b1, c1, d1, e1;
- int16_t *ip = input;
+ const int16_t *ip = input;
int16_t *op = output;
for (i = 0; i < 4; i++) {
- a1 = ip[0] >> WHT_UPSCALE_FACTOR;
- c1 = ip[1] >> WHT_UPSCALE_FACTOR;
- d1 = ip[2] >> WHT_UPSCALE_FACTOR;
- b1 = ip[3] >> WHT_UPSCALE_FACTOR;
+ a1 = ip[0] >> UNIT_QUANT_SHIFT;
+ c1 = ip[1] >> UNIT_QUANT_SHIFT;
+ d1 = ip[2] >> UNIT_QUANT_SHIFT;
+ b1 = ip[3] >> UNIT_QUANT_SHIFT;
a1 += c1;
d1 -= b1;
e1 = (a1 - d1) >> 1;
@@ -60,24 +60,24 @@
c1 = e1 - c1;
a1 -= b1;
d1 += c1;
- dest[dest_stride * 0] = clip_pixel(dest[dest_stride * 0] + a1);
- dest[dest_stride * 1] = clip_pixel(dest[dest_stride * 1] + b1);
- dest[dest_stride * 2] = clip_pixel(dest[dest_stride * 2] + c1);
- dest[dest_stride * 3] = clip_pixel(dest[dest_stride * 3] + d1);
+ dest[stride * 0] = clip_pixel(dest[stride * 0] + a1);
+ dest[stride * 1] = clip_pixel(dest[stride * 1] + b1);
+ dest[stride * 2] = clip_pixel(dest[stride * 2] + c1);
+ dest[stride * 3] = clip_pixel(dest[stride * 3] + d1);
ip++;
dest++;
}
}
-void vp9_short_iwalsh4x4_1_add_c(int16_t *in, uint8_t *dest, int dest_stride) {
+void vp9_iwht4x4_1_add_c(const int16_t *in, uint8_t *dest, int dest_stride) {
int i;
int a1, e1;
int16_t tmp[4];
- int16_t *ip = in;
+ const int16_t *ip = in;
int16_t *op = tmp;
- a1 = ip[0] >> WHT_UPSCALE_FACTOR;
+ a1 = ip[0] >> UNIT_QUANT_SHIFT;
e1 = a1 >> 1;
a1 -= e1;
op[0] = a1;
@@ -96,7 +96,7 @@
}
}
-void vp9_idct4_1d_c(int16_t *input, int16_t *output) {
+static void idct4_1d(const int16_t *input, int16_t *output) {
int16_t step[4];
int temp1, temp2;
// stage 1
@@ -116,7 +116,7 @@
output[3] = step[0] - step[3];
}
-void vp9_short_idct4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+void vp9_idct4x4_16_add_c(const int16_t *input, uint8_t *dest, int stride) {
int16_t out[4 * 4];
int16_t *outptr = out;
int i, j;
@@ -124,7 +124,7 @@
// Rows
for (i = 0; i < 4; ++i) {
- vp9_idct4_1d(input, outptr);
+ idct4_1d(input, outptr);
input += 4;
outptr += 4;
}
@@ -133,14 +133,14 @@
for (i = 0; i < 4; ++i) {
for (j = 0; j < 4; ++j)
temp_in[j] = out[j * 4 + i];
- vp9_idct4_1d(temp_in, temp_out);
+ idct4_1d(temp_in, temp_out);
for (j = 0; j < 4; ++j)
- dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
- + dest[j * dest_stride + i]);
+ dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
+ + dest[j * stride + i]);
}
}
-void vp9_short_idct4x4_1_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+void vp9_idct4x4_1_add_c(const int16_t *input, uint8_t *dest, int dest_stride) {
int i;
int a1;
int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
@@ -156,7 +156,7 @@
}
}
-static void idct8_1d(int16_t *input, int16_t *output) {
+static void idct8_1d(const int16_t *input, int16_t *output) {
int16_t step1[8], step2[8];
int temp1, temp2;
// stage 1
@@ -174,7 +174,7 @@
step1[6] = dct_const_round_shift(temp2);
// stage 2 & stage 3 - even half
- vp9_idct4_1d(step1, step1);
+ idct4_1d(step1, step1);
// stage 2 - odd half
step2[4] = step1[4] + step1[5];
@@ -201,7 +201,7 @@
output[7] = step1[0] - step1[7];
}
-void vp9_short_idct8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+void vp9_idct8x8_64_add_c(const int16_t *input, uint8_t *dest, int stride) {
int16_t out[8 * 8];
int16_t *outptr = out;
int i, j;
@@ -220,12 +220,12 @@
temp_in[j] = out[j * 8 + i];
idct8_1d(temp_in, temp_out);
for (j = 0; j < 8; ++j)
- dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
- + dest[j * dest_stride + i]);
+ dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
+ + dest[j * stride + i]);
}
}
-void vp9_short_idct8x8_1_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+void vp9_idct8x8_1_add_c(const int16_t *input, uint8_t *dest, int stride) {
int i, j;
int a1;
int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
@@ -234,11 +234,11 @@
for (j = 0; j < 8; ++j) {
for (i = 0; i < 8; ++i)
dest[i] = clip_pixel(dest[i] + a1);
- dest += dest_stride;
+ dest += stride;
}
}
-static void iadst4_1d(int16_t *input, int16_t *output) {
+static void iadst4_1d(const int16_t *input, int16_t *output) {
int s0, s1, s2, s3, s4, s5, s6, s7;
int x0 = input[0];
@@ -280,13 +280,13 @@
output[3] = dct_const_round_shift(s3);
}
-void vp9_short_iht4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride,
- int tx_type) {
+void vp9_iht4x4_16_add_c(const int16_t *input, uint8_t *dest, int stride,
+ int tx_type) {
const transform_2d IHT_4[] = {
- { vp9_idct4_1d, vp9_idct4_1d }, // DCT_DCT = 0
- { iadst4_1d, vp9_idct4_1d }, // ADST_DCT = 1
- { vp9_idct4_1d, iadst4_1d }, // DCT_ADST = 2
- { iadst4_1d, iadst4_1d } // ADST_ADST = 3
+ { idct4_1d, idct4_1d }, // DCT_DCT = 0
+ { iadst4_1d, idct4_1d }, // ADST_DCT = 1
+ { idct4_1d, iadst4_1d }, // DCT_ADST = 2
+ { iadst4_1d, iadst4_1d } // ADST_ADST = 3
};
int i, j;
@@ -307,11 +307,11 @@
temp_in[j] = out[j * 4 + i];
IHT_4[tx_type].cols(temp_in, temp_out);
for (j = 0; j < 4; ++j)
- dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
- + dest[j * dest_stride + i]);
+ dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
+ + dest[j * stride + i]);
}
}
-static void iadst8_1d(int16_t *input, int16_t *output) {
+static void iadst8_1d(const int16_t *input, int16_t *output) {
int s0, s1, s2, s3, s4, s5, s6, s7;
int x0 = input[7];
@@ -395,8 +395,8 @@
{ iadst8_1d, iadst8_1d } // ADST_ADST = 3
};
-void vp9_short_iht8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride,
- int tx_type) {
+void vp9_iht8x8_64_add_c(const int16_t *input, uint8_t *dest, int stride,
+ int tx_type) {
int i, j;
int16_t out[8 * 8];
int16_t *outptr = out;
@@ -416,12 +416,12 @@
temp_in[j] = out[j * 8 + i];
ht.cols(temp_in, temp_out);
for (j = 0; j < 8; ++j)
- dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
- + dest[j * dest_stride + i]); }
+ dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
+ + dest[j * stride + i]);
+ }
}
-void vp9_short_idct10_8x8_add_c(int16_t *input, uint8_t *dest,
- int dest_stride) {
+void vp9_idct8x8_10_add_c(const int16_t *input, uint8_t *dest, int stride) {
int16_t out[8 * 8] = { 0 };
int16_t *outptr = out;
int i, j;
@@ -441,12 +441,12 @@
temp_in[j] = out[j * 8 + i];
idct8_1d(temp_in, temp_out);
for (j = 0; j < 8; ++j)
- dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
- + dest[j * dest_stride + i]);
+ dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
+ + dest[j * stride + i]);
}
}
-static void idct16_1d(int16_t *input, int16_t *output) {
+static void idct16_1d(const int16_t *input, int16_t *output) {
int16_t step1[16], step2[16];
int temp1, temp2;
@@ -611,7 +611,7 @@
output[15] = step2[0] - step2[15];
}
-void vp9_short_idct16x16_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+void vp9_idct16x16_256_add_c(const int16_t *input, uint8_t *dest, int stride) {
int16_t out[16 * 16];
int16_t *outptr = out;
int i, j;
@@ -630,12 +630,12 @@
temp_in[j] = out[j * 16 + i];
idct16_1d(temp_in, temp_out);
for (j = 0; j < 16; ++j)
- dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
- + dest[j * dest_stride + i]);
+ dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
+ + dest[j * stride + i]);
}
}
-void iadst16_1d(int16_t *input, int16_t *output) {
+static void iadst16_1d(const int16_t *input, int16_t *output) {
int s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11, s12, s13, s14, s15;
int x0 = input[15];
@@ -813,8 +813,8 @@
{ iadst16_1d, iadst16_1d } // ADST_ADST = 3
};
-void vp9_short_iht16x16_add_c(int16_t *input, uint8_t *dest, int dest_stride,
- int tx_type) {
+void vp9_iht16x16_256_add_c(const int16_t *input, uint8_t *dest, int stride,
+ int tx_type) {
int i, j;
int16_t out[16 * 16];
int16_t *outptr = out;
@@ -834,12 +834,11 @@
temp_in[j] = out[j * 16 + i];
ht.cols(temp_in, temp_out);
for (j = 0; j < 16; ++j)
- dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
- + dest[j * dest_stride + i]); }
+ dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
+ + dest[j * stride + i]); }
}
-void vp9_short_idct10_16x16_add_c(int16_t *input, uint8_t *dest,
- int dest_stride) {
+void vp9_idct16x16_10_add_c(const int16_t *input, uint8_t *dest, int stride) {
int16_t out[16 * 16] = { 0 };
int16_t *outptr = out;
int i, j;
@@ -859,13 +858,12 @@
temp_in[j] = out[j*16 + i];
idct16_1d(temp_in, temp_out);
for (j = 0; j < 16; ++j)
- dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
- + dest[j * dest_stride + i]);
+ dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
+ + dest[j * stride + i]);
}
}
-void vp9_short_idct16x16_1_add_c(int16_t *input, uint8_t *dest,
- int dest_stride) {
+void vp9_idct16x16_1_add_c(const int16_t *input, uint8_t *dest, int stride) {
int i, j;
int a1;
int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
@@ -874,11 +872,11 @@
for (j = 0; j < 16; ++j) {
for (i = 0; i < 16; ++i)
dest[i] = clip_pixel(dest[i] + a1);
- dest += dest_stride;
+ dest += stride;
}
}
-static void idct32_1d(int16_t *input, int16_t *output) {
+static void idct32_1d(const int16_t *input, int16_t *output) {
int16_t step1[32], step2[32];
int temp1, temp2;
@@ -1245,7 +1243,7 @@
output[31] = step1[0] - step1[31];
}
-void vp9_short_idct32x32_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+void vp9_idct32x32_1024_add_c(const int16_t *input, uint8_t *dest, int stride) {
int16_t out[32 * 32];
int16_t *outptr = out;
int i, j;
@@ -1253,6 +1251,44 @@
// Rows
for (i = 0; i < 32; ++i) {
+ int16_t zero_coeff[16];
+ for (j = 0; j < 16; ++j)
+ zero_coeff[j] = input[2 * j] | input[2 * j + 1];
+ for (j = 0; j < 8; ++j)
+ zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
+ for (j = 0; j < 4; ++j)
+ zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
+ for (j = 0; j < 2; ++j)
+ zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
+
+ if (zero_coeff[0] | zero_coeff[1])
+ idct32_1d(input, outptr);
+ else
+ vpx_memset(outptr, 0, sizeof(int16_t) * 32);
+ input += 32;
+ outptr += 32;
+ }
+
+ // Columns
+ for (i = 0; i < 32; ++i) {
+ for (j = 0; j < 32; ++j)
+ temp_in[j] = out[j * 32 + i];
+ idct32_1d(temp_in, temp_out);
+ for (j = 0; j < 32; ++j)
+ dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
+ + dest[j * stride + i]);
+ }
+}
+
+void vp9_idct32x32_34_add_c(const int16_t *input, uint8_t *dest, int stride) {
+ int16_t out[32 * 32] = {0};
+ int16_t *outptr = out;
+ int i, j;
+ int16_t temp_in[32], temp_out[32];
+
+ // Rows
+ // only upper-left 8x8 has non-zero coeff
+ for (i = 0; i < 8; ++i) {
idct32_1d(input, outptr);
input += 32;
outptr += 32;
@@ -1264,13 +1300,116 @@
temp_in[j] = out[j * 32 + i];
idct32_1d(temp_in, temp_out);
for (j = 0; j < 32; ++j)
- dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
- + dest[j * dest_stride + i]);
+ dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
+ + dest[j * stride + i]);
}
}
-void vp9_short_idct1_32x32_c(int16_t *input, int16_t *output) {
+void vp9_idct32x32_1_add_c(const int16_t *input, uint8_t *dest, int stride) {
+ int i, j;
+ int a1;
+
int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
out = dct_const_round_shift(out * cospi_16_64);
- output[0] = ROUND_POWER_OF_TWO(out, 6);
+ a1 = ROUND_POWER_OF_TWO(out, 6);
+
+ for (j = 0; j < 32; ++j) {
+ for (i = 0; i < 32; ++i)
+ dest[i] = clip_pixel(dest[i] + a1);
+ dest += stride;
+ }
}
+
+// idct
+void vp9_idct4x4_add(const int16_t *input, uint8_t *dest, int stride, int eob) {
+ if (eob > 1)
+ vp9_idct4x4_16_add(input, dest, stride);
+ else
+ vp9_idct4x4_1_add(input, dest, stride);
+}
+
+
+void vp9_iwht4x4_add(const int16_t *input, uint8_t *dest, int stride, int eob) {
+ if (eob > 1)
+ vp9_iwht4x4_16_add(input, dest, stride);
+ else
+ vp9_iwht4x4_1_add(input, dest, stride);
+}
+
+void vp9_idct8x8_add(const int16_t *input, uint8_t *dest, int stride, int eob) {
+ // If dc is 1, then input[0] is the reconstructed value, do not need
+ // dequantization. Also, when dc is 1, dc is counted in eobs, namely eobs >=1.
+
+ // The calculation can be simplified if there are not many non-zero dct
+ // coefficients. Use eobs to decide what to do.
+ // TODO(yunqingwang): "eobs = 1" case is also handled in vp9_short_idct8x8_c.
+ // Combine that with code here.
+ if (eob) {
+ if (eob == 1)
+ // DC only DCT coefficient
+ vp9_idct8x8_1_add(input, dest, stride);
+ else if (eob <= 10)
+ vp9_idct8x8_10_add(input, dest, stride);
+ else
+ vp9_idct8x8_64_add(input, dest, stride);
+ }
+}
+
+void vp9_idct16x16_add(const int16_t *input, uint8_t *dest, int stride,
+ int eob) {
+ /* The calculation can be simplified if there are not many non-zero dct
+ * coefficients. Use eobs to separate different cases. */
+ if (eob) {
+ if (eob == 1)
+ /* DC only DCT coefficient. */
+ vp9_idct16x16_1_add(input, dest, stride);
+ else if (eob <= 10)
+ vp9_idct16x16_10_add(input, dest, stride);
+ else
+ vp9_idct16x16_256_add(input, dest, stride);
+ }
+}
+
+void vp9_idct32x32_add(const int16_t *input, uint8_t *dest, int stride,
+ int eob) {
+ if (eob) {
+ if (eob == 1)
+ vp9_idct32x32_1_add(input, dest, stride);
+ else if (eob <= 34)
+ // non-zero coeff only in upper-left 8x8
+ vp9_idct32x32_34_add(input, dest, stride);
+ else
+ vp9_idct32x32_1024_add(input, dest, stride);
+ }
+}
+
+// iht
+void vp9_iht4x4_add(TX_TYPE tx_type, const int16_t *input, uint8_t *dest,
+ int stride, int eob) {
+ if (tx_type == DCT_DCT)
+ vp9_idct4x4_add(input, dest, stride, eob);
+ else
+ vp9_iht4x4_16_add(input, dest, stride, tx_type);
+}
+
+void vp9_iht8x8_add(TX_TYPE tx_type, const int16_t *input, uint8_t *dest,
+ int stride, int eob) {
+ if (tx_type == DCT_DCT) {
+ vp9_idct8x8_add(input, dest, stride, eob);
+ } else {
+ if (eob > 0) {
+ vp9_iht8x8_64_add(input, dest, stride, tx_type);
+ }
+ }
+}
+
+void vp9_iht16x16_add(TX_TYPE tx_type, const int16_t *input, uint8_t *dest,
+ int stride, int eob) {
+ if (tx_type == DCT_DCT) {
+ vp9_idct16x16_add(input, dest, stride, eob);
+ } else {
+ if (eob > 0) {
+ vp9_iht16x16_256_add(input, dest, stride, tx_type);
+ }
+ }
+}
« no previous file with comments | « source/libvpx/vp9/common/vp9_idct.h ('k') | source/libvpx/vp9/common/vp9_loopfilter.h » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698