summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--simd/arm/common/jidctfst-neon.c12
-rw-r--r--simd/arm/common/jidctint-neon.c34
-rw-r--r--simd/arm/common/jidctred-neon.c8
3 files changed, 32 insertions, 22 deletions
diff --git a/simd/arm/common/jidctfst-neon.c b/simd/arm/common/jidctfst-neon.c
index c926e6d..87806fd 100644
--- a/simd/arm/common/jidctfst-neon.c
+++ b/simd/arm/common/jidctfst-neon.c
@@ -84,8 +84,8 @@ void jsimd_idct_ifast_neon(void *dct_table,
bitmap = vorrq_s16(bitmap, row6);
bitmap = vorrq_s16(bitmap, row7);
- int64_t left_ac_bitmap = vreinterpret_s64_s16(vget_low_s16(bitmap));
- int64_t right_ac_bitmap = vreinterpret_s64_s16(vget_high_s16(bitmap));
+ int64_t left_ac_bitmap = vgetq_lane_s64(vreinterpretq_s64_s16(bitmap), 0);
+ int64_t right_ac_bitmap = vgetq_lane_s64(vreinterpretq_s64_s16(bitmap), 1);
if (left_ac_bitmap == 0 && right_ac_bitmap == 0) {
/* All AC coefficients are zero. */
@@ -405,13 +405,13 @@ void jsimd_idct_ifast_neon(void *dct_table,
vqshrn_n_s16(col7, PASS1_BITS + 3));
/* Clamp to range [0-255]. */
uint8x16_t cols_01 = vreinterpretq_u8_s8(
- vaddq_s8(cols_01_s8, vdupq_n_u8(CENTERJSAMPLE)));
+ vaddq_s8(cols_01_s8, vreinterpretq_s8_u8(vdupq_n_u8(CENTERJSAMPLE))));
uint8x16_t cols_45 = vreinterpretq_u8_s8(
- vaddq_s8(cols_45_s8, vdupq_n_u8(CENTERJSAMPLE)));
+ vaddq_s8(cols_45_s8, vreinterpretq_s8_u8(vdupq_n_u8(CENTERJSAMPLE))));
uint8x16_t cols_23 = vreinterpretq_u8_s8(
- vaddq_s8(cols_23_s8, vdupq_n_u8(CENTERJSAMPLE)));
+ vaddq_s8(cols_23_s8, vreinterpretq_s8_u8(vdupq_n_u8(CENTERJSAMPLE))));
uint8x16_t cols_67 = vreinterpretq_u8_s8(
- vaddq_s8(cols_67_s8, vdupq_n_u8(CENTERJSAMPLE)));
+ vaddq_s8(cols_67_s8, vreinterpretq_s8_u8(vdupq_n_u8(CENTERJSAMPLE))));
/* Transpose block ready for store. */
uint32x4x2_t cols_0415 = vzipq_u32(vreinterpretq_u32_u8(cols_01),
diff --git a/simd/arm/common/jidctint-neon.c b/simd/arm/common/jidctint-neon.c
index 4ee9eb0..0fd4a36 100644
--- a/simd/arm/common/jidctint-neon.c
+++ b/simd/arm/common/jidctint-neon.c
@@ -215,13 +215,13 @@ void jsimd_idct_islow_neon(void *dct_table,
int16x4_t bitmap = vorr_s16(row7, row6);
bitmap = vorr_s16(bitmap, row5);
bitmap = vorr_s16(bitmap, row4);
- int64_t bitmap_rows_4567 = vreinterpret_s64_s16(bitmap);
+ int64_t bitmap_rows_4567 = vget_lane_s64(vreinterpret_s64_s16(bitmap), 0);
if (bitmap_rows_4567 == 0) {
bitmap = vorr_s16(bitmap, row3);
bitmap = vorr_s16(bitmap, row2);
bitmap = vorr_s16(bitmap, row1);
- int64_t left_ac_bitmap = vreinterpret_s64_s16(bitmap);
+ int64_t left_ac_bitmap = vget_lane_s64(vreinterpret_s64_s16(bitmap), 0);
if (left_ac_bitmap == 0) {
int16x4_t dcval = vshl_n_s16(vmul_s16(row0, quant_row0), PASS1_BITS);
@@ -267,18 +267,18 @@ void jsimd_idct_islow_neon(void *dct_table,
bitmap = vorr_s16(row7, row6);
bitmap = vorr_s16(bitmap, row5);
bitmap = vorr_s16(bitmap, row4);
- bitmap_rows_4567 = vreinterpret_s64_s16(bitmap);
+ bitmap_rows_4567 = vget_lane_s64(vreinterpret_s64_s16(bitmap), 0);
bitmap = vorr_s16(bitmap, row3);
bitmap = vorr_s16(bitmap, row2);
bitmap = vorr_s16(bitmap, row1);
- int64_t right_ac_bitmap = vreinterpret_s64_s16(bitmap);
+ int64_t right_ac_bitmap = vget_lane_s64(vreinterpret_s64_s16(bitmap), 0);
/* Initialise to non-zero value: defaults to regular second pass. */
int64_t right_ac_dc_bitmap = 1;
if (right_ac_bitmap == 0) {
bitmap = vorr_s16(bitmap, row0);
- right_ac_dc_bitmap = vreinterpret_s64_s16(bitmap);
+ right_ac_dc_bitmap = vget_lane_s64(vreinterpret_s64_s16(bitmap), 0);
if (right_ac_dc_bitmap != 0) {
int16x4_t dcval = vshl_n_s16(vmul_s16(row0, quant_row0), PASS1_BITS);
@@ -631,10 +631,14 @@ static inline void jsimd_idct_islow_pass2_regular(int16_t *workspace,
int8x8_t cols_46_s8 = vqrshrn_n_s16(cols_46_s16, DESCALE_P2 - 16);
int8x8_t cols_57_s8 = vqrshrn_n_s16(cols_57_s16, DESCALE_P2 - 16);
/* Clamp to range [0-255]. */
- uint8x8_t cols_02_u8 = vadd_u8(cols_02_s8, vdup_n_u8(CENTERJSAMPLE));
- uint8x8_t cols_13_u8 = vadd_u8(cols_13_s8, vdup_n_u8(CENTERJSAMPLE));
- uint8x8_t cols_46_u8 = vadd_u8(cols_46_s8, vdup_n_u8(CENTERJSAMPLE));
- uint8x8_t cols_57_u8 = vadd_u8(cols_57_s8, vdup_n_u8(CENTERJSAMPLE));
+ uint8x8_t cols_02_u8 = vadd_u8(vreinterpret_u8_s8(cols_02_s8),
+ vdup_n_u8(CENTERJSAMPLE));
+ uint8x8_t cols_13_u8 = vadd_u8(vreinterpret_u8_s8(cols_13_s8),
+ vdup_n_u8(CENTERJSAMPLE));
+ uint8x8_t cols_46_u8 = vadd_u8(vreinterpret_u8_s8(cols_46_s8),
+ vdup_n_u8(CENTERJSAMPLE));
+ uint8x8_t cols_57_u8 = vadd_u8(vreinterpret_u8_s8(cols_57_s8),
+ vdup_n_u8(CENTERJSAMPLE));
/* Transpose 4x8 block and store to memory. */
/* Zipping adjacent columns together allows us to store 16-bit elements. */
@@ -723,10 +727,14 @@ static inline void jsimd_idct_islow_pass2_sparse(int16_t *workspace,
int8x8_t cols_46_s8 = vqrshrn_n_s16(cols_46_s16, DESCALE_P2 - 16);
int8x8_t cols_57_s8 = vqrshrn_n_s16(cols_57_s16, DESCALE_P2 - 16);
/* Clamp to range [0-255]. */
- uint8x8_t cols_02_u8 = vadd_u8(cols_02_s8, vdup_n_u8(CENTERJSAMPLE));
- uint8x8_t cols_13_u8 = vadd_u8(cols_13_s8, vdup_n_u8(CENTERJSAMPLE));
- uint8x8_t cols_46_u8 = vadd_u8(cols_46_s8, vdup_n_u8(CENTERJSAMPLE));
- uint8x8_t cols_57_u8 = vadd_u8(cols_57_s8, vdup_n_u8(CENTERJSAMPLE));
+ uint8x8_t cols_02_u8 = vadd_u8(vreinterpret_u8_s8(cols_02_s8),
+ vdup_n_u8(CENTERJSAMPLE));
+ uint8x8_t cols_13_u8 = vadd_u8(vreinterpret_u8_s8(cols_13_s8),
+ vdup_n_u8(CENTERJSAMPLE));
+ uint8x8_t cols_46_u8 = vadd_u8(vreinterpret_u8_s8(cols_46_s8),
+ vdup_n_u8(CENTERJSAMPLE));
+ uint8x8_t cols_57_u8 = vadd_u8(vreinterpret_u8_s8(cols_57_s8),
+ vdup_n_u8(CENTERJSAMPLE));
/* Transpose 4x8 block and store to memory. */
/* Zipping adjacent columns together allow us to store 16-bit elements. */
diff --git a/simd/arm/common/jidctred-neon.c b/simd/arm/common/jidctred-neon.c
index 3c9393b..ed4232c 100644
--- a/simd/arm/common/jidctred-neon.c
+++ b/simd/arm/common/jidctred-neon.c
@@ -218,8 +218,8 @@ void jsimd_idct_4x4_neon(void *dct_table,
bitmap = vorrq_s16(bitmap, row6);
bitmap = vorrq_s16(bitmap, row7);
- int64_t left_ac_bitmap = vreinterpret_s64_s16(vget_low_s16(bitmap));
- int64_t right_ac_bitmap = vreinterpret_s64_s16(vget_high_s16(bitmap));
+ int64_t left_ac_bitmap = vgetq_lane_s64(vreinterpretq_s64_s16(bitmap), 0);
+ int64_t right_ac_bitmap = vgetq_lane_s64(vreinterpretq_s64_s16(bitmap), 1);
/* Load constants for IDCT computation. */
const int16x4x3_t consts = vld1_s16_x3(jsimd_idct_4x4_neon_consts);
@@ -453,7 +453,9 @@ void jsimd_idct_4x4_neon(void *dct_table,
/* Interleaving store completes the transpose. */
uint8x8x2_t output_0123 = vzip_u8(vqmovun_s16(output_cols_02),
vqmovun_s16(output_cols_13));
- uint16x4x2_t output_01_23 = { output_0123.val[0], output_0123.val[1] };
+ uint16x4x2_t output_01_23 = { vreinterpret_u16_u8(output_0123.val[0]),
+ vreinterpret_u16_u8(output_0123.val[1])
+ };
/* Store 4x4 block to memory. */
JSAMPROW outptr0 = output_buf[0] + output_col;
generated by cgit v1.2.3 (git 2.25.1) at 2025-10-03 05:45:31 +0000