Prefer scalar arithmetic and use BSL in Neon z1 predictors
The whole calculation of 'shift' can be done in scalar rather than
vector which is generally cheaper.
The refactoring of ORR(AND(a, b), BIC(c, b)) => BSL(a, b, c) is already
done by compilers automatically, so this is just cleaning up the code.
Change-Id: I4f82bc4d8e07947261ae78c9b96c8cbdcd71c21b
diff --git a/aom_dsp/arm/intrapred_neon.c b/aom_dsp/arm/intrapred_neon.c
index 7eb9e7d..f66f048 100644
--- a/aom_dsp/arm/intrapred_neon.c
+++ b/aom_dsp/arm/intrapred_neon.c
@@ -1171,8 +1171,6 @@
a16 = vdupq_n_u16(16);
a_mbase_x = vdup_n_u8(above[max_base_x]);
uint16x8_t v_32 = vdupq_n_u16(32);
- int16x8_t v_upsample_above = vdupq_n_s16(upsample_above);
- uint16x8_t c3f = vdupq_n_u16(0x3f);
int x = dx;
for (int r = 0; r < W; r++) {
@@ -1193,12 +1191,11 @@
if (upsample_above) {
v_tmp_a0_128 = vld2_u8(above + base);
- shift = vshrq_n_u16(
- vandq_u16(vshlq_u16(vdupq_n_u16(x), v_upsample_above), c3f), 1);
+ shift = vdupq_n_u16(((x << upsample_above) & 0x3f) >> 1);
} else {
v_tmp_a0_128.val[0] = vld1_u8(above + base);
v_tmp_a0_128.val[1] = vld1_u8(above + base + 1);
- shift = vshrq_n_u16(vandq_u16(vdupq_n_u16(x), c3f), 1);
+ shift = vdupq_n_u16((x & 0x3f) >> 1);
}
a0 = vmovl_u8(v_tmp_a0_128.val[0]);
a1 = vmovl_u8(v_tmp_a0_128.val[1]);
@@ -1207,8 +1204,7 @@
res = vmlaq_u16(a32, diff, shift);
uint8x8_t mask = vld1_u8(BaseMask[base_max_diff]);
- dst[r] =
- vorr_u8(vand_u8(mask, vshrn_n_u16(res, 5)), vbic_u8(a_mbase_x, mask));
+ dst[r] = vbsl_u8(mask, vshrn_n_u16(res, 5), a_mbase_x);
x += dx;
}
@@ -1255,15 +1251,11 @@
uint8x16x2_t a0, a1;
uint16x8x2_t diff, a32;
- uint16x8_t a16, c3f;
- uint8x16_t a_mbase_x;
- a16 = vdupq_n_u16(16);
- a_mbase_x = vdupq_n_u8(above[max_base_x]);
- c3f = vdupq_n_u16(0x3f);
+ uint16x8_t a16 = vdupq_n_u16(16);
+ uint8x16_t a_mbase_x = vdupq_n_u8(above[max_base_x]);
uint16x8_t v_32 = vdupq_n_u16(32);
uint8x16_t v_zero = vdupq_n_u8(0);
- int16x8_t v_upsample_above = vdupq_n_s16(upsample_above);
int x = dx;
for (int r = 0; r < W; r++) {
@@ -1286,12 +1278,11 @@
uint8x8x2_t v_tmp_a0_128 = vld2_u8(above + base);
a0_128 = vcombine_u8(v_tmp_a0_128.val[0], v_tmp_a0_128.val[1]);
a1_128 = vextq_u8(a0_128, v_zero, 8);
- shift = vshrq_n_u16(
- vandq_u16(vshlq_u16(vdupq_n_u16(x), v_upsample_above), c3f), 1);
+ shift = vdupq_n_u16(((x << upsample_above) & 0x3f) >> 1);
} else {
a0_128 = vld1q_u8(above + base);
a1_128 = vld1q_u8(above + base + 1);
- shift = vshrq_n_u16(vandq_u16(vdupq_n_u16(x), c3f), 1);
+ shift = vdupq_n_u16((x & 0x3f) >> 1);
}
a0 = vzipq_u8(a0_128, v_zero);
a1 = vzipq_u8(a1_128, v_zero);
@@ -1309,7 +1300,7 @@
vcombine_u8(vshrn_n_u16(res.val[0], 5), vshrn_n_u16(res.val[1], 5));
uint8x16_t mask = vld1q_u8(BaseMask[base_max_diff]);
- dst[r] = vorrq_u8(vandq_u8(mask, v_temp), vbicq_u8(a_mbase_x, mask));
+ dst[r] = vbslq_u8(mask, v_temp, a_mbase_x);
x += dx;
}
@@ -1344,11 +1335,9 @@
uint8x16_t a_mbase_x;
uint8x16x2_t a0, a1;
uint16x8x2_t diff, a32;
- uint16x8_t a16, c3f;
a_mbase_x = vdupq_n_u8(above[max_base_x]);
- a16 = vdupq_n_u16(16);
- c3f = vdupq_n_u16(0x3f);
+ uint16x8_t a16 = vdupq_n_u16(16);
uint16x8_t v_32 = vdupq_n_u16(32);
uint8x16_t v_zero = vdupq_n_u8(0);
@@ -1369,7 +1358,7 @@
}
if (base_max_diff > 32) base_max_diff = 32;
- uint16x8_t shift = vshrq_n_u16(vandq_u16(vdupq_n_u16(x), c3f), 1);
+ uint16x8_t shift = vdupq_n_u16((x & 0x3f) >> 1);
for (int j = 0, jj = 0; j < 32; j += 16, jj++) {
int mdiff = base_max_diff - j;
@@ -1402,10 +1391,8 @@
mask.val[0] = vld1q_u8(BaseMask[base_max_diff]);
mask.val[1] = vld1q_u8(BaseMask[base_max_diff] + 16);
- dstvec[r].val[0] = vorrq_u8(vandq_u8(mask.val[0], res16[0]),
- vbicq_u8(a_mbase_x, mask.val[0]));
- dstvec[r].val[1] = vorrq_u8(vandq_u8(mask.val[1], res16[1]),
- vbicq_u8(a_mbase_x, mask.val[1]));
+ dstvec[r].val[0] = vbslq_u8(mask.val[0], res16[0], a_mbase_x);
+ dstvec[r].val[1] = vbslq_u8(mask.val[1], res16[1], a_mbase_x);
x += dx;
}
}
@@ -1439,13 +1426,10 @@
uint8x16x2_t a0, a1;
uint16x8x2_t a32, diff;
- uint16x8_t a16, c3f;
- uint8x16_t a_mbase_x, max_base_x128, mask128;
- a16 = vdupq_n_u16(16);
- a_mbase_x = vdupq_n_u8(above[max_base_x]);
- max_base_x128 = vdupq_n_u8(max_base_x);
- c3f = vdupq_n_u16(0x3f);
+ uint16x8_t a16 = vdupq_n_u16(16);
+ uint8x16_t a_mbase_x = vdupq_n_u8(above[max_base_x]);
+ uint8x16_t max_base_x128 = vdupq_n_u8(max_base_x);
uint16x8_t v_32 = vdupq_n_u16(32);
uint8x16_t v_zero = vdupq_n_u8(0);
uint8x16_t step = vdupq_n_u8(16);
@@ -1466,8 +1450,8 @@
return;
}
- uint16x8_t shift = vshrq_n_u16(vandq_u16(vdupq_n_u16(x), c3f), 1);
- uint8x16_t a0_128, a1_128, res128;
+ uint16x8_t shift = vdupq_n_u16((x & 0x3f) >> 1);
+ uint8x16_t a0_128, a1_128, res128, mask128;
uint8x16_t base_inc128 =
vaddq_u8(vdupq_n_u8(base), vcombine_u8(vcreate_u8(0x0706050403020100),
vcreate_u8(0x0F0E0D0C0B0A0908)));
@@ -1497,8 +1481,7 @@
vcombine_u8(vshrn_n_u16(res.val[0], 5), vshrn_n_u16(res.val[1], 5));
mask128 = vcgtq_u8(vqsubq_u8(max_base_x128, base_inc128), v_zero);
- res128 =
- vorrq_u8(vandq_u8(mask128, v_temp), vbicq_u8(a_mbase_x, mask128));
+ res128 = vbslq_u8(mask128, v_temp, a_mbase_x);
vst1q_u8(dst + j, res128);
base_inc128 = vaddq_u8(base_inc128, step);
