* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
-#include "avutil.h"
+#include "libavutil/avutil.h"
#define ALTIVEC_TRANSPOSE_8x8_SHORT(src_a,src_b,src_c,src_d,src_e,src_f,src_g,src_h) \
do { \
vector by assuming (stride % 16) == 0, unfortunately
this is not always true.
*/
- DECLARE_ALIGNED(16, short, data[8]);
+ short data_0 = ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
+ DECLARE_ALIGNED(16, short, data)[8] =
+ {
+ data_0,
+ data_0 * 2 + 1,
+ c->QP * 2,
+ c->QP * 4
+ };
int numEq;
uint8_t *src2 = src;
vector signed short v_dcOffset;
const vector signed int zero = vec_splat_s32(0);
const vector signed short mask = vec_splat_s16(1);
vector signed int v_numEq = vec_splat_s32(0);
-
- data[0] = ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
- data[1] = data[0] * 2 + 1;
- data[2] = c->QP * 2;
- data[3] = c->QP * 4;
vector signed short v_data = vec_ld(0, data);
+ vector signed short v_srcAss0, v_srcAss1, v_srcAss2, v_srcAss3,
+ v_srcAss4, v_srcAss5, v_srcAss6, v_srcAss7;
+//FIXME avoid this mess if possible
+ register int j0 = 0,
+ j1 = stride,
+ j2 = 2 * stride,
+ j3 = 3 * stride,
+ j4 = 4 * stride,
+ j5 = 5 * stride,
+ j6 = 6 * stride,
+ j7 = 7 * stride;
+ vector unsigned char v_srcA0, v_srcA1, v_srcA2, v_srcA3,
+ v_srcA4, v_srcA5, v_srcA6, v_srcA7;
+
v_dcOffset = vec_splat(v_data, 0);
v_dcThreshold = (vector unsigned short)vec_splat(v_data, 1);
v2QP = vec_splat(v_data, 2);
src2 += stride * 4;
- vector signed short v_srcAss0, v_srcAss1, v_srcAss2, v_srcAss3, v_srcAss4, v_srcAss5, v_srcAss6, v_srcAss7;
-
#define LOAD_LINE(i) \
- register int j##i = i * stride; \
+ { \
vector unsigned char perm##i = vec_lvsl(j##i, src2); \
- const vector unsigned char v_srcA1##i = vec_ld(j##i, src2); \
vector unsigned char v_srcA2##i; \
+ vector unsigned char v_srcA1##i = vec_ld(j##i, src2); \
if (two_vectors) \
v_srcA2##i = vec_ld(j##i + 16, src2); \
- const vector unsigned char v_srcA##i = \
+ v_srcA##i = \
vec_perm(v_srcA1##i, v_srcA2##i, perm##i); \
v_srcAss##i = \
(vector signed short)vec_mergeh((vector signed char)zero, \
- (vector signed char)v_srcA##i)
+ (vector signed char)v_srcA##i); }
#define LOAD_LINE_ALIGNED(i) \
- register int j##i = i * stride; \
- const vector unsigned char v_srcA##i = vec_ld(j##i, src2); \
+ v_srcA##i = vec_ld(j##i, src2); \
v_srcAss##i = \
(vector signed short)vec_mergeh((vector signed char)zero, \
(vector signed char)v_srcA##i)
const vector signed short v_comp##i = \
(vector signed short)vec_cmplt((vector unsigned short)v_sum##i, \
v_dcThreshold); \
- const vector signed short v_part##i = vec_and(mask, v_comp##i); \
- v_numEq = vec_sum4s(v_part##i, v_numEq);
-
- ITER(0, 1);
- ITER(1, 2);
- ITER(2, 3);
- ITER(3, 4);
- ITER(4, 5);
- ITER(5, 6);
- ITER(6, 7);
+ const vector signed short v_part##i = vec_and(mask, v_comp##i);
+
+ {
+ ITER(0, 1)
+ ITER(1, 2)
+ ITER(2, 3)
+ ITER(3, 4)
+ ITER(4, 5)
+ ITER(5, 6)
+ ITER(6, 7)
+
+ v_numEq = vec_sum4s(v_part0, v_numEq);
+ v_numEq = vec_sum4s(v_part1, v_numEq);
+ v_numEq = vec_sum4s(v_part2, v_numEq);
+ v_numEq = vec_sum4s(v_part3, v_numEq);
+ v_numEq = vec_sum4s(v_part4, v_numEq);
+ v_numEq = vec_sum4s(v_part5, v_numEq);
+ v_numEq = vec_sum4s(v_part6, v_numEq);
+ }
+
#undef ITER
v_numEq = vec_sums(v_numEq, zero);
if (numEq > c->ppMode.flatnessThreshold){
const vector unsigned char mmoP1 = (const vector unsigned char)
- AVV(0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f,
- 0x00, 0x01, 0x12, 0x13, 0x08, 0x09, 0x1A, 0x1B);
+ {0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f,
+ 0x00, 0x01, 0x12, 0x13, 0x08, 0x09, 0x1A, 0x1B};
const vector unsigned char mmoP2 = (const vector unsigned char)
- AVV(0x04, 0x05, 0x16, 0x17, 0x0C, 0x0D, 0x1E, 0x1F,
- 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f);
+ {0x04, 0x05, 0x16, 0x17, 0x0C, 0x0D, 0x1E, 0x1F,
+ 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f};
const vector unsigned char mmoP = (const vector unsigned char)
vec_lvsl(8, (unsigned char*)0);
const vector signed int zero = vec_splat_s32(0);
const int properStride = (stride % 16);
const int srcAlign = ((unsigned long)src2 % 16);
- DECLARE_ALIGNED(16, short, qp[8]);
- qp[0] = c->QP;
+ DECLARE_ALIGNED(16, short, qp)[8] = {c->QP};
vector signed short vqp = vec_ld(0, qp);
+ vector signed short vb0, vb1, vb2, vb3, vb4, vb5, vb6, vb7, vb8, vb9;
+ vector unsigned char vbA0, av_uninit(vbA1), av_uninit(vbA2), av_uninit(vbA3), av_uninit(vbA4), av_uninit(vbA5), av_uninit(vbA6), av_uninit(vbA7), av_uninit(vbA8), vbA9;
+ vector unsigned char vbB0, av_uninit(vbB1), av_uninit(vbB2), av_uninit(vbB3), av_uninit(vbB4), av_uninit(vbB5), av_uninit(vbB6), av_uninit(vbB7), av_uninit(vbB8), vbB9;
+ vector unsigned char vbT0, vbT1, vbT2, vbT3, vbT4, vbT5, vbT6, vbT7, vbT8, vbT9;
+ vector unsigned char perml0, perml1, perml2, perml3, perml4,
+ perml5, perml6, perml7, perml8, perml9;
+ register int j0 = 0,
+ j1 = stride,
+ j2 = 2 * stride,
+ j3 = 3 * stride,
+ j4 = 4 * stride,
+ j5 = 5 * stride,
+ j6 = 6 * stride,
+ j7 = 7 * stride,
+ j8 = 8 * stride,
+ j9 = 9 * stride;
+
vqp = vec_splat(vqp, 0);
src2 += stride*3;
- vector signed short vb0, vb1, vb2, vb3, vb4, vb5, vb6, vb7, vb8, vb9;
- vector unsigned char vbA0, vbA1, vbA2, vbA3, vbA4, vbA5, vbA6, vbA7, vbA8, vbA9;
- vector unsigned char vbB0, vbB1, vbB2, vbB3, vbB4, vbB5, vbB6, vbB7, vbB8, vbB9;
- vector unsigned char vbT0, vbT1, vbT2, vbT3, vbT4, vbT5, vbT6, vbT7, vbT8, vbT9;
-
#define LOAD_LINE(i) \
- const vector unsigned char perml##i = \
- vec_lvsl(i * stride, src2); \
+ perml##i = vec_lvsl(i * stride, src2); \
vbA##i = vec_ld(i * stride, src2); \
vbB##i = vec_ld(i * stride + 16, src2); \
vbT##i = vec_perm(vbA##i, vbB##i, perml##i); \
(vector unsigned char)vbT##i)
#define LOAD_LINE_ALIGNED(i) \
- register int j##i = i * stride; \
vbT##i = vec_ld(j##i, src2); \
vb##i = \
(vector signed short)vec_mergeh((vector signed char)zero, \
/* Special-casing the aligned case is worthwhile, as all calls from
* the (transposed) horizontable deblocks will be aligned, in addition
* to the naturally aligned vertical deblocks. */
- if (properStride && srcAlign) {
+ if (properStride && srcAlign) {
LOAD_LINE_ALIGNED(0);
LOAD_LINE_ALIGNED(1);
LOAD_LINE_ALIGNED(2);
LOAD_LINE_ALIGNED(7);
LOAD_LINE_ALIGNED(8);
LOAD_LINE_ALIGNED(9);
- } else {
+ } else {
LOAD_LINE(0);
LOAD_LINE(1);
LOAD_LINE(2);
LOAD_LINE(7);
LOAD_LINE(8);
LOAD_LINE(9);
- }
+ }
#undef LOAD_LINE
#undef LOAD_LINE_ALIGNED
-
- const vector unsigned short v_2 = vec_splat_u16(2);
- const vector unsigned short v_4 = vec_splat_u16(4);
-
- const vector signed short v_diff01 = vec_sub(vb0, vb1);
- const vector unsigned short v_cmp01 =
- (const vector unsigned short) vec_cmplt(vec_abs(v_diff01), vqp);
- const vector signed short v_first = vec_sel(vb1, vb0, v_cmp01);
- const vector signed short v_diff89 = vec_sub(vb8, vb9);
- const vector unsigned short v_cmp89 =
- (const vector unsigned short) vec_cmplt(vec_abs(v_diff89), vqp);
- const vector signed short v_last = vec_sel(vb8, vb9, v_cmp89);
-
- const vector signed short temp01 = vec_mladd(v_first, (vector signed short)v_4, vb1);
- const vector signed short temp02 = vec_add(vb2, vb3);
- const vector signed short temp03 = vec_add(temp01, (vector signed short)v_4);
- const vector signed short v_sumsB0 = vec_add(temp02, temp03);
-
- const vector signed short temp11 = vec_sub(v_sumsB0, v_first);
- const vector signed short v_sumsB1 = vec_add(temp11, vb4);
-
- const vector signed short temp21 = vec_sub(v_sumsB1, v_first);
- const vector signed short v_sumsB2 = vec_add(temp21, vb5);
-
- const vector signed short temp31 = vec_sub(v_sumsB2, v_first);
- const vector signed short v_sumsB3 = vec_add(temp31, vb6);
-
- const vector signed short temp41 = vec_sub(v_sumsB3, v_first);
- const vector signed short v_sumsB4 = vec_add(temp41, vb7);
-
- const vector signed short temp51 = vec_sub(v_sumsB4, vb1);
- const vector signed short v_sumsB5 = vec_add(temp51, vb8);
-
- const vector signed short temp61 = vec_sub(v_sumsB5, vb2);
- const vector signed short v_sumsB6 = vec_add(temp61, v_last);
-
- const vector signed short temp71 = vec_sub(v_sumsB6, vb3);
- const vector signed short v_sumsB7 = vec_add(temp71, v_last);
-
- const vector signed short temp81 = vec_sub(v_sumsB7, vb4);
- const vector signed short v_sumsB8 = vec_add(temp81, v_last);
-
- const vector signed short temp91 = vec_sub(v_sumsB8, vb5);
- const vector signed short v_sumsB9 = vec_add(temp91, v_last);
-
-#define COMPUTE_VR(i, j, k) \
- const vector signed short temps1##i = \
- vec_add(v_sumsB##i, v_sumsB##k); \
- const vector signed short temps2##i = \
- vec_mladd(vb##j, (vector signed short)v_2, temps1##i); \
- const vector signed short vr##j = vec_sra(temps2##i, v_4)
-
- COMPUTE_VR(0, 1, 2);
- COMPUTE_VR(1, 2, 3);
- COMPUTE_VR(2, 3, 4);
- COMPUTE_VR(3, 4, 5);
- COMPUTE_VR(4, 5, 6);
- COMPUTE_VR(5, 6, 7);
- COMPUTE_VR(6, 7, 8);
- COMPUTE_VR(7, 8, 9);
-
- const vector signed char neg1 = vec_splat_s8(-1);
- const vector unsigned char permHH = (const vector unsigned char)AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
- 0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F);
+ {
+ const vector unsigned short v_2 = vec_splat_u16(2);
+ const vector unsigned short v_4 = vec_splat_u16(4);
+
+ const vector signed short v_diff01 = vec_sub(vb0, vb1);
+ const vector unsigned short v_cmp01 =
+ (const vector unsigned short) vec_cmplt(vec_abs(v_diff01), vqp);
+ const vector signed short v_first = vec_sel(vb1, vb0, v_cmp01);
+ const vector signed short v_diff89 = vec_sub(vb8, vb9);
+ const vector unsigned short v_cmp89 =
+ (const vector unsigned short) vec_cmplt(vec_abs(v_diff89), vqp);
+ const vector signed short v_last = vec_sel(vb8, vb9, v_cmp89);
+
+ const vector signed short temp01 = vec_mladd(v_first, (vector signed short)v_4, vb1);
+ const vector signed short temp02 = vec_add(vb2, vb3);
+ const vector signed short temp03 = vec_add(temp01, (vector signed short)v_4);
+ const vector signed short v_sumsB0 = vec_add(temp02, temp03);
+
+ const vector signed short temp11 = vec_sub(v_sumsB0, v_first);
+ const vector signed short v_sumsB1 = vec_add(temp11, vb4);
+
+ const vector signed short temp21 = vec_sub(v_sumsB1, v_first);
+ const vector signed short v_sumsB2 = vec_add(temp21, vb5);
+
+ const vector signed short temp31 = vec_sub(v_sumsB2, v_first);
+ const vector signed short v_sumsB3 = vec_add(temp31, vb6);
+
+ const vector signed short temp41 = vec_sub(v_sumsB3, v_first);
+ const vector signed short v_sumsB4 = vec_add(temp41, vb7);
+
+ const vector signed short temp51 = vec_sub(v_sumsB4, vb1);
+ const vector signed short v_sumsB5 = vec_add(temp51, vb8);
+
+ const vector signed short temp61 = vec_sub(v_sumsB5, vb2);
+ const vector signed short v_sumsB6 = vec_add(temp61, v_last);
+
+ const vector signed short temp71 = vec_sub(v_sumsB6, vb3);
+ const vector signed short v_sumsB7 = vec_add(temp71, v_last);
+
+ const vector signed short temp81 = vec_sub(v_sumsB7, vb4);
+ const vector signed short v_sumsB8 = vec_add(temp81, v_last);
+
+ const vector signed short temp91 = vec_sub(v_sumsB8, vb5);
+ const vector signed short v_sumsB9 = vec_add(temp91, v_last);
+
+ #define COMPUTE_VR(i, j, k) \
+ const vector signed short temps1##i = \
+ vec_add(v_sumsB##i, v_sumsB##k); \
+ const vector signed short temps2##i = \
+ vec_mladd(vb##j, (vector signed short)v_2, temps1##i); \
+ const vector signed short vr##j = vec_sra(temps2##i, v_4)
+
+ COMPUTE_VR(0, 1, 2);
+ COMPUTE_VR(1, 2, 3);
+ COMPUTE_VR(2, 3, 4);
+ COMPUTE_VR(3, 4, 5);
+ COMPUTE_VR(4, 5, 6);
+ COMPUTE_VR(5, 6, 7);
+ COMPUTE_VR(6, 7, 8);
+ COMPUTE_VR(7, 8, 9);
+
+ const vector signed char neg1 = vec_splat_s8(-1);
+ const vector unsigned char permHH = (const vector unsigned char){0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
+ 0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F};
#define PACK_AND_STORE(i) \
- const vector unsigned char perms##i = \
+{ const vector unsigned char perms##i = \
vec_lvsr(i * stride, src2); \
const vector unsigned char vf##i = \
vec_packsu(vr##i, (vector signed short)zero); \
const vector unsigned char svB##i = \
vec_sel(vg2##i, vbB##i, mask##i); \
vec_st(svA##i, i * stride, src2); \
- vec_st(svB##i, i * stride + 16, src2)
+ vec_st(svB##i, i * stride + 16, src2);}
#define PACK_AND_STORE_ALIGNED(i) \
- const vector unsigned char vf##i = \
+{ const vector unsigned char vf##i = \
vec_packsu(vr##i, (vector signed short)zero); \
const vector unsigned char vg##i = \
vec_perm(vf##i, vbT##i, permHH); \
- vec_st(vg##i, i * stride, src2)
-
- /* Special-casing the aligned case is worthwhile, as all calls from
- * the (transposed) horizontable deblocks will be aligned, in addition
- * to the naturally aligned vertical deblocks. */
- if (properStride && srcAlign) {
- PACK_AND_STORE_ALIGNED(1);
- PACK_AND_STORE_ALIGNED(2);
- PACK_AND_STORE_ALIGNED(3);
- PACK_AND_STORE_ALIGNED(4);
- PACK_AND_STORE_ALIGNED(5);
- PACK_AND_STORE_ALIGNED(6);
- PACK_AND_STORE_ALIGNED(7);
- PACK_AND_STORE_ALIGNED(8);
- } else {
- PACK_AND_STORE(1);
- PACK_AND_STORE(2);
- PACK_AND_STORE(3);
- PACK_AND_STORE(4);
- PACK_AND_STORE(5);
- PACK_AND_STORE(6);
- PACK_AND_STORE(7);
- PACK_AND_STORE(8);
+ vec_st(vg##i, i * stride, src2);}
+
+ /* Special-casing the aligned case is worthwhile, as all calls from
+ * the (transposed) horizontable deblocks will be aligned, in addition
+ * to the naturally aligned vertical deblocks. */
+ if (properStride && srcAlign) {
+ PACK_AND_STORE_ALIGNED(1)
+ PACK_AND_STORE_ALIGNED(2)
+ PACK_AND_STORE_ALIGNED(3)
+ PACK_AND_STORE_ALIGNED(4)
+ PACK_AND_STORE_ALIGNED(5)
+ PACK_AND_STORE_ALIGNED(6)
+ PACK_AND_STORE_ALIGNED(7)
+ PACK_AND_STORE_ALIGNED(8)
+ } else {
+ PACK_AND_STORE(1)
+ PACK_AND_STORE(2)
+ PACK_AND_STORE(3)
+ PACK_AND_STORE(4)
+ PACK_AND_STORE(5)
+ PACK_AND_STORE(6)
+ PACK_AND_STORE(7)
+ PACK_AND_STORE(8)
+ }
+ #undef PACK_AND_STORE
+ #undef PACK_AND_STORE_ALIGNED
}
-#undef PACK_AND_STORE
-#undef PACK_AND_STORE_ALIGNED
}
can be removed by assuming proper alignment of
src & stride :-(
*/
- uint8_t *src2 = src;
+ uint8_t *src2 = src + stride*3;
const vector signed int zero = vec_splat_s32(0);
- DECLARE_ALIGNED(16, short, qp[8]);
- qp[0] = 8*c->QP;
- vector signed short vqp = vec_ld(0, qp);
- vqp = vec_splat(vqp, 0);
+ DECLARE_ALIGNED(16, short, qp)[8] = {8*c->QP};
+ vector signed short vqp = vec_splat(
+ (vector signed short)vec_ld(0, qp), 0);
#define LOAD_LINE(i) \
const vector unsigned char perm##i = \
(vector signed short)vec_mergeh((vector unsigned char)zero, \
(vector unsigned char)vbT##i)
- src2 += stride*3;
-
LOAD_LINE(1);
LOAD_LINE(2);
LOAD_LINE(3);
const vector unsigned char st5 = vec_packsu(vb5plusd, (vector signed short)zero);
const vector signed char neg1 = vec_splat_s8(-1);
- const vector unsigned char permHH = (const vector unsigned char)AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
- 0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F);
+ const vector unsigned char permHH = (const vector unsigned char){0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
+ 0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F};
#define STORE(i) \
- const vector unsigned char perms##i = \
+{ const vector unsigned char perms##i = \
vec_lvsr(i * stride, src2); \
const vector unsigned char vg##i = \
vec_perm(st##i, vbT##i, permHH); \
const vector unsigned char svB##i = \
vec_sel(vg2##i, vbB##i, mask##i); \
vec_st(svA##i, i * stride, src2); \
- vec_st(svB##i, i * stride + 16, src2)
+ vec_st(svB##i, i * stride + 16, src2);}
- STORE(4);
- STORE(5);
+ STORE(4)
+ STORE(5)
}
static inline void dering_altivec(uint8_t src[], int stride, PPContext *c) {
src & stride :-(
*/
uint8_t *srcCopy = src;
- DECLARE_ALIGNED(16, uint8_t, dt[16]);
+ DECLARE_ALIGNED(16, uint8_t, dt)[16];
const vector signed int zero = vec_splat_s32(0);
vector unsigned char v_dt;
dt[0] = deringThreshold;
v_dt = vec_splat(vec_ld(0, dt), 0);
-#define LOAD_LINE(i) \
- const vector unsigned char perm##i = \
- vec_lvsl(i * stride, srcCopy); \
- vector unsigned char sA##i = vec_ld(i * stride, srcCopy); \
- vector unsigned char sB##i = vec_ld(i * stride + 16, srcCopy); \
+#define LOAD_LINE(i) \
+ const vector unsigned char perm##i = \
+ vec_lvsl(i * stride, srcCopy); \
+ vector unsigned char sA##i = vec_ld(i * stride, srcCopy); \
+ vector unsigned char sB##i = vec_ld(i * stride + 16, srcCopy); \
vector unsigned char src##i = vec_perm(sA##i, sB##i, perm##i)
LOAD_LINE(0);
vector unsigned char v_avg;
{
const vector unsigned char trunc_perm = (vector unsigned char)
- AVV(0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08,
- 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18);
+ {0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08,
+ 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18};
const vector unsigned char trunc_src12 = vec_perm(src1, src2, trunc_perm);
const vector unsigned char trunc_src34 = vec_perm(src3, src4, trunc_perm);
const vector unsigned char trunc_src56 = vec_perm(src5, src6, trunc_perm);
v_avg = vec_avg(v_min, v_max);
}
- DECLARE_ALIGNED(16, signed int, S[8]);
+ DECLARE_ALIGNED(16, signed int, S)[8];
{
const vector unsigned short mask1 = (vector unsigned short)
- AVV(0x0001, 0x0002, 0x0004, 0x0008,
- 0x0010, 0x0020, 0x0040, 0x0080);
+ {0x0001, 0x0002, 0x0004, 0x0008,
+ 0x0010, 0x0020, 0x0040, 0x0080};
const vector unsigned short mask2 = (vector unsigned short)
- AVV(0x0100, 0x0200, 0x0000, 0x0000,
- 0x0000, 0x0000, 0x0000, 0x0000);
+ {0x0100, 0x0200, 0x0000, 0x0000,
+ 0x0000, 0x0000, 0x0000, 0x0000};
const vector unsigned int vuint32_16 = vec_sl(vec_splat_u32(1), vec_splat_u32(4));
const vector unsigned int vuint32_1 = vec_splat_u32(1);
/* I'm not sure the following is actually faster
than straight, unvectorized C code :-( */
- DECLARE_ALIGNED(16, int, tQP2[4]);
+ DECLARE_ALIGNED(16, int, tQP2)[4];
tQP2[0]= c->QP/2 + 1;
vector signed int vQP2 = vec_ld(0, tQP2);
vQP2 = vec_splat(vQP2, 0);
const vector unsigned int vuint32_4 = vec_splat_u32(4);
const vector unsigned char permA1 = (vector unsigned char)
- AVV(0x00, 0x01, 0x02, 0x10, 0x11, 0x12, 0x1F, 0x1F,
- 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F);
+ {0x00, 0x01, 0x02, 0x10, 0x11, 0x12, 0x1F, 0x1F,
+ 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F};
const vector unsigned char permA2 = (vector unsigned char)
- AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x10, 0x11,
- 0x12, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F);
+ {0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x10, 0x11,
+ 0x12, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F};
const vector unsigned char permA1inc = (vector unsigned char)
- AVV(0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00,
- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
+ {0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00};
const vector unsigned char permA2inc = (vector unsigned char)
- AVV(0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x01,
- 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
+ {0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x01,
+ 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00};
const vector unsigned char magic = (vector unsigned char)
- AVV(0x01, 0x02, 0x01, 0x02, 0x04, 0x02, 0x01, 0x02,
- 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
+ {0x01, 0x02, 0x01, 0x02, 0x04, 0x02, 0x01, 0x02,
+ 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00};
const vector unsigned char extractPerm = (vector unsigned char)
- AVV(0x10, 0x10, 0x10, 0x01, 0x10, 0x10, 0x10, 0x01,
- 0x10, 0x10, 0x10, 0x01, 0x10, 0x10, 0x10, 0x01);
+ {0x10, 0x10, 0x10, 0x01, 0x10, 0x10, 0x10, 0x01,
+ 0x10, 0x10, 0x10, 0x01, 0x10, 0x10, 0x10, 0x01};
const vector unsigned char extractPermInc = (vector unsigned char)
- AVV(0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01,
- 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01);
+ {0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01,
+ 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01};
const vector unsigned char identity = vec_lvsl(0,(unsigned char *)0);
const vector unsigned char tenRight = (vector unsigned char)
- AVV(0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
+ {0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00};
const vector unsigned char eightLeft = (vector unsigned char)
- AVV(0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x08);
+ {0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x08};
#define F_INIT(i) \
#define do_a_deblock_altivec(a...) do_a_deblock_C(a)
static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride,
- uint8_t *tempBlured, uint32_t *tempBluredPast, int *maxNoise)
+ uint8_t *tempBlurred, uint32_t *tempBlurredPast, int *maxNoise)
{
const vector signed int zero = vec_splat_s32(0);
const vector signed short vsint16_1 = vec_splat_s16(1);
vector signed int v_sysdp = zero;
int d, sysd, i;
- tempBluredPast[127]= maxNoise[0];
- tempBluredPast[128]= maxNoise[1];
- tempBluredPast[129]= maxNoise[2];
+ tempBlurredPast[127]= maxNoise[0];
+ tempBlurredPast[128]= maxNoise[1];
+ tempBlurredPast[129]= maxNoise[2];
#define LOAD_LINE(src, i) \
register int j##src##i = i * stride; \
LOAD_LINE(src, 6);
LOAD_LINE(src, 7);
- LOAD_LINE(tempBlured, 0);
- LOAD_LINE(tempBlured, 1);
- LOAD_LINE(tempBlured, 2);
- LOAD_LINE(tempBlured, 3);
- LOAD_LINE(tempBlured, 4);
- LOAD_LINE(tempBlured, 5);
- LOAD_LINE(tempBlured, 6);
- LOAD_LINE(tempBlured, 7);
+ LOAD_LINE(tempBlurred, 0);
+ LOAD_LINE(tempBlurred, 1);
+ LOAD_LINE(tempBlurred, 2);
+ LOAD_LINE(tempBlurred, 3);
+ LOAD_LINE(tempBlurred, 4);
+ LOAD_LINE(tempBlurred, 5);
+ LOAD_LINE(tempBlurred, 6);
+ LOAD_LINE(tempBlurred, 7);
#undef LOAD_LINE
#define ACCUMULATE_DIFFS(i) \
- vector signed short v_d##i = vec_sub(v_tempBluredAss##i, \
+ vector signed short v_d##i = vec_sub(v_tempBlurredAss##i, \
v_srcAss##i); \
v_dp = vec_msums(v_d##i, v_d##i, v_dp); \
v_sysdp = vec_msums(v_d##i, vsint16_1, v_sysdp)
i = d;
d = (4*d
- +(*(tempBluredPast-256))
- +(*(tempBluredPast-1))+ (*(tempBluredPast+1))
- +(*(tempBluredPast+256))
+ +(*(tempBlurredPast-256))
+ +(*(tempBlurredPast-1))+ (*(tempBlurredPast+1))
+ +(*(tempBlurredPast+256))
+4)>>3;
- *tempBluredPast=i;
+ *tempBlurredPast=i;
if (d > maxNoise[1]) {
if (d < maxNoise[2]) {
-#define OP(i) v_tempBluredAss##i = vec_avg(v_tempBluredAss##i, v_srcAss##i);
+#define OP(i) v_tempBlurredAss##i = vec_avg(v_tempBlurredAss##i, v_srcAss##i);
OP(0);
OP(1);
OP(7);
#undef OP
} else {
-#define OP(i) v_tempBluredAss##i = v_srcAss##i;
+#define OP(i) v_tempBlurredAss##i = v_srcAss##i;
OP(0);
OP(1);
#define OP(i) \
const vector signed short v_temp##i = \
- vec_mladd(v_tempBluredAss##i, \
+ vec_mladd(v_tempBlurredAss##i, \
vsint16_7, v_srcAss##i); \
const vector signed short v_temp2##i = \
vec_add(v_temp##i, vsint16_4); \
- v_tempBluredAss##i = vec_sr(v_temp2##i, vuint16_3)
+ v_tempBlurredAss##i = vec_sr(v_temp2##i, vuint16_3)
OP(0);
OP(1);
#define OP(i) \
const vector signed short v_temp##i = \
- vec_mladd(v_tempBluredAss##i, \
+ vec_mladd(v_tempBlurredAss##i, \
vsint16_3, v_srcAss##i); \
const vector signed short v_temp2##i = \
vec_add(v_temp##i, vsint16_2); \
- v_tempBluredAss##i = vec_sr(v_temp2##i, (vector unsigned short)vsint16_2)
+ v_tempBlurredAss##i = vec_sr(v_temp2##i, (vector unsigned short)vsint16_2)
OP(0);
OP(1);
}
const vector signed char neg1 = vec_splat_s8(-1);
- const vector unsigned char permHH = (const vector unsigned char)AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
- 0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F);
+ const vector unsigned char permHH = (const vector unsigned char){0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
+ 0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F};
#define PACK_AND_STORE(src, i) \
const vector unsigned char perms##src##i = \
vec_lvsr(i * stride, src); \
const vector unsigned char vf##src##i = \
- vec_packsu(v_tempBluredAss##i, (vector signed short)zero); \
+ vec_packsu(v_tempBlurredAss##i, (vector signed short)zero); \
const vector unsigned char vg##src##i = \
vec_perm(vf##src##i, v_##src##A##i, permHH); \
const vector unsigned char mask##src##i = \
PACK_AND_STORE(src, 5);
PACK_AND_STORE(src, 6);
PACK_AND_STORE(src, 7);
- PACK_AND_STORE(tempBlured, 0);
- PACK_AND_STORE(tempBlured, 1);
- PACK_AND_STORE(tempBlured, 2);
- PACK_AND_STORE(tempBlured, 3);
- PACK_AND_STORE(tempBlured, 4);
- PACK_AND_STORE(tempBlured, 5);
- PACK_AND_STORE(tempBlured, 6);
- PACK_AND_STORE(tempBlured, 7);
+ PACK_AND_STORE(tempBlurred, 0);
+ PACK_AND_STORE(tempBlurred, 1);
+ PACK_AND_STORE(tempBlurred, 2);
+ PACK_AND_STORE(tempBlurred, 3);
+ PACK_AND_STORE(tempBlurred, 4);
+ PACK_AND_STORE(tempBlurred, 5);
+ PACK_AND_STORE(tempBlurred, 6);
+ PACK_AND_STORE(tempBlurred, 7);
#undef PACK_AND_STORE
}
vector unsigned char tempO = vec_mergeh(src7, zero);
vector unsigned char tempP = vec_mergel(src7, zero);
- vector unsigned char temp0 = vec_mergeh(tempA, tempI);
- vector unsigned char temp1 = vec_mergel(tempA, tempI);
- vector unsigned char temp2 = vec_mergeh(tempB, tempJ);
- vector unsigned char temp3 = vec_mergel(tempB, tempJ);
- vector unsigned char temp4 = vec_mergeh(tempC, tempK);
- vector unsigned char temp5 = vec_mergel(tempC, tempK);
- vector unsigned char temp6 = vec_mergeh(tempD, tempL);
- vector unsigned char temp7 = vec_mergel(tempD, tempL);
- vector unsigned char temp8 = vec_mergeh(tempE, tempM);
- vector unsigned char temp9 = vec_mergel(tempE, tempM);
+ vector unsigned char temp0 = vec_mergeh(tempA, tempI);
+ vector unsigned char temp1 = vec_mergel(tempA, tempI);
+ vector unsigned char temp2 = vec_mergeh(tempB, tempJ);
+ vector unsigned char temp3 = vec_mergel(tempB, tempJ);
+ vector unsigned char temp4 = vec_mergeh(tempC, tempK);
+ vector unsigned char temp5 = vec_mergel(tempC, tempK);
+ vector unsigned char temp6 = vec_mergeh(tempD, tempL);
+ vector unsigned char temp7 = vec_mergel(tempD, tempL);
+ vector unsigned char temp8 = vec_mergeh(tempE, tempM);
+ vector unsigned char temp9 = vec_mergel(tempE, tempM);
vector unsigned char temp10 = vec_mergeh(tempF, tempN);
vector unsigned char temp11 = vec_mergel(tempF, tempN);
vector unsigned char temp12 = vec_mergeh(tempG, tempO);
tempO = vec_mergeh(temp7, temp15);
tempP = vec_mergel(temp7, temp15);
- temp0 = vec_mergeh(tempA, tempI);
- temp1 = vec_mergel(tempA, tempI);
- temp2 = vec_mergeh(tempB, tempJ);
- temp3 = vec_mergel(tempB, tempJ);
- temp4 = vec_mergeh(tempC, tempK);
- temp5 = vec_mergel(tempC, tempK);
- temp6 = vec_mergeh(tempD, tempL);
- temp7 = vec_mergel(tempD, tempL);
- temp8 = vec_mergeh(tempE, tempM);
- temp9 = vec_mergel(tempE, tempM);
+ temp0 = vec_mergeh(tempA, tempI);
+ temp1 = vec_mergel(tempA, tempI);
+ temp2 = vec_mergeh(tempB, tempJ);
+ temp3 = vec_mergel(tempB, tempJ);
+ temp4 = vec_mergeh(tempC, tempK);
+ temp5 = vec_mergel(tempC, tempK);
+ temp6 = vec_mergeh(tempD, tempL);
+ temp7 = vec_mergel(tempD, tempL);
+ temp8 = vec_mergeh(tempE, tempM);
+ temp9 = vec_mergel(tempE, tempM);
temp10 = vec_mergeh(tempF, tempN);
temp11 = vec_mergel(tempF, tempN);
temp12 = vec_mergeh(tempG, tempO);
temp14 = vec_mergeh(tempH, tempP);
temp15 = vec_mergel(tempH, tempP);
- vec_st(temp0, 0, dst);
- vec_st(temp1, 16, dst);
- vec_st(temp2, 32, dst);
- vec_st(temp3, 48, dst);
- vec_st(temp4, 64, dst);
- vec_st(temp5, 80, dst);
- vec_st(temp6, 96, dst);
- vec_st(temp7, 112, dst);
- vec_st(temp8, 128, dst);
- vec_st(temp9, 144, dst);
+ vec_st(temp0, 0, dst);
+ vec_st(temp1, 16, dst);
+ vec_st(temp2, 32, dst);
+ vec_st(temp3, 48, dst);
+ vec_st(temp4, 64, dst);
+ vec_st(temp5, 80, dst);
+ vec_st(temp6, 96, dst);
+ vec_st(temp7, 112, dst);
+ vec_st(temp8, 128, dst);
+ vec_st(temp9, 144, dst);
vec_st(temp10, 160, dst);
vec_st(temp11, 176, dst);
vec_st(temp12, 192, dst);