diff options
Diffstat (limited to 'src/arm/64/looprestoration_common.S')
| -rw-r--r-- | src/arm/64/looprestoration_common.S | 28 |
1 files changed, 19 insertions, 9 deletions
diff --git a/src/arm/64/looprestoration_common.S b/src/arm/64/looprestoration_common.S index dc07827..200eb63 100644 --- a/src/arm/64/looprestoration_common.S +++ b/src/arm/64/looprestoration_common.S @@ -328,10 +328,13 @@ function sgr_box5_v_neon, export=1 endfunc // void dav1d_sgr_calc_ab1_neon(int32_t *a, int16_t *b, -// const int w, const int h, const int strength); +// const int w, const int h, const int strength, +// const int bitdepth_max); // void dav1d_sgr_calc_ab2_neon(int32_t *a, int16_t *b, -// const int w, const int h, const int strength); +// const int w, const int h, const int strength, +// const int bitdepth_max); function sgr_calc_ab1_neon, export=1 + clz w9, w5 add x3, x3, #2 // h += 2 movi v31.4s, #9 // n mov x5, #455 @@ -340,6 +343,7 @@ function sgr_calc_ab1_neon, export=1 endfunc function sgr_calc_ab2_neon, export=1 + clz w9, w5 add x3, x3, #3 // h += 3 asr x3, x3, #1 // h /= 2 movi v31.4s, #25 // n @@ -348,14 +352,17 @@ function sgr_calc_ab2_neon, export=1 endfunc function sgr_calc_ab_neon + sub w9, w9, #24 // -bitdepth_min_8 movrel x12, X(sgr_x_by_x) ld1 {v16.16b, v17.16b, v18.16b}, [x12] + dup v6.8h, w9 // -bitdepth_min_8 movi v19.16b, #5 movi v20.8b, #55 // idx of last 5 movi v21.8b, #72 // idx of last 4 movi v22.8b, #101 // idx of last 3 movi v23.8b, #169 // idx of last 2 movi v24.8b, #254 // idx of last 1 + saddl v7.4s, v6.4h, v6.4h // -2*bitdepth_min_8 add x2, x2, #2 // w += 2 add x7, x2, #7 bic x7, x7, #7 // aligned w @@ -373,10 +380,13 @@ function sgr_calc_ab_neon subs x2, x2, #8 ld1 {v0.4s, v1.4s}, [x0] // a ld1 {v2.8h}, [x1] // b + srshl v0.4s, v0.4s, v7.4s + srshl v1.4s, v1.4s, v7.4s + srshl v4.8h, v2.8h, v6.8h mul v0.4s, v0.4s, v31.4s // a * n mul v1.4s, v1.4s, v31.4s // a * n - umull v3.4s, v2.4h, v2.4h // b * b - umull2 v4.4s, v2.8h, v2.8h // b * b + umull v3.4s, v4.4h, v4.4h // b * b + umull2 v4.4s, v4.8h, v4.8h // b * b uqsub v0.4s, v0.4s, v3.4s // imax(a * n - b * b, 0) uqsub v1.4s, v1.4s, v4.4s // imax(a * n - b * b, 0) mul v0.4s, v0.4s, v28.4s // p * s @@ -389,13 +399,13 @@ function sgr_calc_ab_neon cmhi v26.8b, v0.8b, v21.8b // = -1 if sgr_x_by_x[v0] < 4 tbl v1.8b, {v16.16b, v17.16b, v18.16b}, v0.8b cmhi v27.8b, v0.8b, v22.8b // = -1 if sgr_x_by_x[v0] < 3 - cmhi v5.8b, v0.8b, v23.8b // = -1 if sgr_x_by_x[v0] < 2 + cmhi v4.8b, v0.8b, v23.8b // = -1 if sgr_x_by_x[v0] < 2 add v25.8b, v25.8b, v26.8b - cmhi v6.8b, v0.8b, v24.8b // = -1 if sgr_x_by_x[v0] < 1 - add v27.8b, v27.8b, v5.8b - add v6.8b, v6.8b, v19.8b + cmhi v5.8b, v0.8b, v24.8b // = -1 if sgr_x_by_x[v0] < 1 + add v27.8b, v27.8b, v4.8b + add v5.8b, v5.8b, v19.8b add v25.8b, v25.8b, v27.8b - add v1.8b, v1.8b, v6.8b + add v1.8b, v1.8b, v5.8b add v1.8b, v1.8b, v25.8b uxtl v1.8h, v1.8b // x |