src2 += stride * 4;
-
#define LOAD_LINE(i) \
{ \
vector unsigned char perm##i = vec_lvsl(j##i, src2); \
v_dcThreshold); \
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);
-}
+ {
+ 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
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)AVV(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); \
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)
+ /* 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);
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) {
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);
#undef LOAD_LINE
#define ACCUMULATE_DIFFS(i) \
- vector signed short v_d##i = vec_sub(v_tempBlurredAss##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)
#define OP(i) \
const vector signed short v_temp##i = \
- vec_mladd(v_tempBlurredAss##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); \
#define OP(i) \
const vector signed short v_temp##i = \
- vec_mladd(v_tempBlurredAss##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); \
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);