/dports/misc/ncnn/ncnn-20211208/docs/developer-guide/ |
H A D | armv7-mix-assembly-and-intrinsic.md | 21 float32x4_t _a = vld1q_f32(a); 22 float32x4_t _b = vld1q_f32(b); 23 float32x4_t _c = vld1q_f32(c); 58 float32x4_t _a = vld1q_f32(a); 59 float32x4_t _b = vld1q_f32(b); 60 float32x4_t _c = vld1q_f32(c); 80 float32x4_t _c = vld1q_f32(c); 112 register float32x4_t _a asm("q0") = vld1q_f32(a); 113 register float32x4_t _b asm("q1") = vld1q_f32(b); 114 register float32x4_t _c asm("q2") = vld1q_f32(c);
|
/dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/docs/developer-guide/ |
H A D | armv7-mix-assembly-and-intrinsic.md | 21 float32x4_t _a = vld1q_f32(a); 22 float32x4_t _b = vld1q_f32(b); 23 float32x4_t _c = vld1q_f32(c); 58 float32x4_t _a = vld1q_f32(a); 59 float32x4_t _b = vld1q_f32(b); 60 float32x4_t _c = vld1q_f32(c); 80 float32x4_t _c = vld1q_f32(c); 112 register float32x4_t _a asm("q0") = vld1q_f32(a); 113 register float32x4_t _b asm("q1") = vld1q_f32(b); 114 register float32x4_t _c asm("q2") = vld1q_f32(c);
|
/dports/www/chromium-legacy/chromium-88.0.4324.182/third_party/blink/renderer/modules/webaudio/cpu/arm/ |
H A D | oscillator_kernel_neon.cc | 104 const float32x4_t s1_low = vld1q_f32(sample1_lower); in ProcessKRateVector() 105 const float32x4_t s2_low = vld1q_f32(sample2_lower); in ProcessKRateVector() 106 const float32x4_t s1_high = vld1q_f32(sample1_higher); in ProcessKRateVector() 107 const float32x4_t s2_high = vld1q_f32(sample2_higher); in ProcessKRateVector() 220 vld1q_f32(sample1_higher), v_factor, in ProcessARateVectorKernel() 221 vsubq_f32(vld1q_f32(sample2_higher), vld1q_f32(sample1_higher))); in ProcessARateVectorKernel() 225 vmlaq_f32(vld1q_f32(sample1_lower), v_factor, in ProcessARateVectorKernel() 226 vsubq_f32(vld1q_f32(sample2_lower), vld1q_f32(sample1_lower))); in ProcessARateVectorKernel() 231 vmlaq_f32(sample_higher, vld1q_f32(table_interpolation_factor), in ProcessARateVectorKernel()
|
/dports/games/iortcw/iortcw-1.51c/MP/code/opus-1.2.1/celt/arm/ |
H A D | pitch_neon_intr.c | 204 x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 205 y_f32x4 = vld1q_f32(&y[i]); in celt_inner_prod_neon() 207 x_f32x4 = vld1q_f32(&x[i + 4]); in celt_inner_prod_neon() 208 y_f32x4 = vld1q_f32(&y[i + 4]); in celt_inner_prod_neon() 213 const float32x4_t x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 245 x_f32x4 = vld1q_f32(&x[i]); in dual_inner_prod_neon() 246 y01_f32x4 = vld1q_f32(&y01[i]); in dual_inner_prod_neon() 247 y02_f32x4 = vld1q_f32(&y02[i]); in dual_inner_prod_neon() 250 x_f32x4 = vld1q_f32(&x[i + 4]); in dual_inner_prod_neon() 251 y01_f32x4 = vld1q_f32(&y01[i + 4]); in dual_inner_prod_neon() [all …]
|
/dports/games/iortcw/iortcw-1.51c/SP/code/opus-1.2.1/celt/arm/ |
H A D | pitch_neon_intr.c | 204 x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 205 y_f32x4 = vld1q_f32(&y[i]); in celt_inner_prod_neon() 207 x_f32x4 = vld1q_f32(&x[i + 4]); in celt_inner_prod_neon() 208 y_f32x4 = vld1q_f32(&y[i + 4]); in celt_inner_prod_neon() 213 const float32x4_t x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 245 x_f32x4 = vld1q_f32(&x[i]); in dual_inner_prod_neon() 246 y01_f32x4 = vld1q_f32(&y01[i]); in dual_inner_prod_neon() 247 y02_f32x4 = vld1q_f32(&y02[i]); in dual_inner_prod_neon() 250 x_f32x4 = vld1q_f32(&x[i + 4]); in dual_inner_prod_neon() 251 y01_f32x4 = vld1q_f32(&y01[i + 4]); in dual_inner_prod_neon() [all …]
|
/dports/www/firefox-esr/firefox-91.8.0/media/libopus/celt/arm/ |
H A D | pitch_neon_intr.c | 204 x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 205 y_f32x4 = vld1q_f32(&y[i]); in celt_inner_prod_neon() 207 x_f32x4 = vld1q_f32(&x[i + 4]); in celt_inner_prod_neon() 208 y_f32x4 = vld1q_f32(&y[i + 4]); in celt_inner_prod_neon() 213 const float32x4_t x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 245 x_f32x4 = vld1q_f32(&x[i]); in dual_inner_prod_neon() 246 y01_f32x4 = vld1q_f32(&y01[i]); in dual_inner_prod_neon() 247 y02_f32x4 = vld1q_f32(&y02[i]); in dual_inner_prod_neon() 250 x_f32x4 = vld1q_f32(&x[i + 4]); in dual_inner_prod_neon() 251 y01_f32x4 = vld1q_f32(&y01[i + 4]); in dual_inner_prod_neon() [all …]
|
/dports/www/qt5-webengine/qtwebengine-everywhere-src-5.15.2/src/3rdparty/chromium/third_party/opus/src/celt/arm/ |
H A D | pitch_neon_intr.c | 204 x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 205 y_f32x4 = vld1q_f32(&y[i]); in celt_inner_prod_neon() 207 x_f32x4 = vld1q_f32(&x[i + 4]); in celt_inner_prod_neon() 208 y_f32x4 = vld1q_f32(&y[i + 4]); in celt_inner_prod_neon() 213 const float32x4_t x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 245 x_f32x4 = vld1q_f32(&x[i]); in dual_inner_prod_neon() 246 y01_f32x4 = vld1q_f32(&y01[i]); in dual_inner_prod_neon() 247 y02_f32x4 = vld1q_f32(&y02[i]); in dual_inner_prod_neon() 250 x_f32x4 = vld1q_f32(&x[i + 4]); in dual_inner_prod_neon() 251 y01_f32x4 = vld1q_f32(&y01[i + 4]); in dual_inner_prod_neon() [all …]
|
/dports/www/firefox/firefox-99.0/media/libopus/celt/arm/ |
H A D | pitch_neon_intr.c | 204 x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 205 y_f32x4 = vld1q_f32(&y[i]); in celt_inner_prod_neon() 207 x_f32x4 = vld1q_f32(&x[i + 4]); in celt_inner_prod_neon() 208 y_f32x4 = vld1q_f32(&y[i + 4]); in celt_inner_prod_neon() 213 const float32x4_t x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 245 x_f32x4 = vld1q_f32(&x[i]); in dual_inner_prod_neon() 246 y01_f32x4 = vld1q_f32(&y01[i]); in dual_inner_prod_neon() 247 y02_f32x4 = vld1q_f32(&y02[i]); in dual_inner_prod_neon() 250 x_f32x4 = vld1q_f32(&x[i + 4]); in dual_inner_prod_neon() 251 y01_f32x4 = vld1q_f32(&y01[i + 4]); in dual_inner_prod_neon() [all …]
|
/dports/lang/spidermonkey60/firefox-60.9.0/media/libopus/celt/arm/ |
H A D | pitch_neon_intr.c | 204 x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 205 y_f32x4 = vld1q_f32(&y[i]); in celt_inner_prod_neon() 207 x_f32x4 = vld1q_f32(&x[i + 4]); in celt_inner_prod_neon() 208 y_f32x4 = vld1q_f32(&y[i + 4]); in celt_inner_prod_neon() 213 const float32x4_t x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 245 x_f32x4 = vld1q_f32(&x[i]); in dual_inner_prod_neon() 246 y01_f32x4 = vld1q_f32(&y01[i]); in dual_inner_prod_neon() 247 y02_f32x4 = vld1q_f32(&y02[i]); in dual_inner_prod_neon() 250 x_f32x4 = vld1q_f32(&x[i + 4]); in dual_inner_prod_neon() 251 y01_f32x4 = vld1q_f32(&y01[i + 4]); in dual_inner_prod_neon() [all …]
|
/dports/games/emptyepsilon/SeriousProton-EE-2021.06.23/src/libopus/celt/arm/ |
H A D | pitch_neon_intr.c | 204 x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 205 y_f32x4 = vld1q_f32(&y[i]); in celt_inner_prod_neon() 207 x_f32x4 = vld1q_f32(&x[i + 4]); in celt_inner_prod_neon() 208 y_f32x4 = vld1q_f32(&y[i + 4]); in celt_inner_prod_neon() 213 const float32x4_t x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 245 x_f32x4 = vld1q_f32(&x[i]); in dual_inner_prod_neon() 246 y01_f32x4 = vld1q_f32(&y01[i]); in dual_inner_prod_neon() 247 y02_f32x4 = vld1q_f32(&y02[i]); in dual_inner_prod_neon() 250 x_f32x4 = vld1q_f32(&x[i + 4]); in dual_inner_prod_neon() 251 y01_f32x4 = vld1q_f32(&y01[i + 4]); in dual_inner_prod_neon() [all …]
|
/dports/www/chromium-legacy/chromium-88.0.4324.182/third_party/opus/src/celt/arm/ |
H A D | pitch_neon_intr.c | 204 x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 205 y_f32x4 = vld1q_f32(&y[i]); in celt_inner_prod_neon() 207 x_f32x4 = vld1q_f32(&x[i + 4]); in celt_inner_prod_neon() 208 y_f32x4 = vld1q_f32(&y[i + 4]); in celt_inner_prod_neon() 213 const float32x4_t x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 245 x_f32x4 = vld1q_f32(&x[i]); in dual_inner_prod_neon() 246 y01_f32x4 = vld1q_f32(&y01[i]); in dual_inner_prod_neon() 247 y02_f32x4 = vld1q_f32(&y02[i]); in dual_inner_prod_neon() 250 x_f32x4 = vld1q_f32(&x[i + 4]); in dual_inner_prod_neon() 251 y01_f32x4 = vld1q_f32(&y01[i + 4]); in dual_inner_prod_neon() [all …]
|
/dports/audio/jamulus/jamulus-r3_8_1/libs/opus/celt/arm/ |
H A D | pitch_neon_intr.c | 204 x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 205 y_f32x4 = vld1q_f32(&y[i]); in celt_inner_prod_neon() 207 x_f32x4 = vld1q_f32(&x[i + 4]); in celt_inner_prod_neon() 208 y_f32x4 = vld1q_f32(&y[i + 4]); in celt_inner_prod_neon() 213 const float32x4_t x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 245 x_f32x4 = vld1q_f32(&x[i]); in dual_inner_prod_neon() 246 y01_f32x4 = vld1q_f32(&y01[i]); in dual_inner_prod_neon() 247 y02_f32x4 = vld1q_f32(&y02[i]); in dual_inner_prod_neon() 250 x_f32x4 = vld1q_f32(&x[i + 4]); in dual_inner_prod_neon() 251 y01_f32x4 = vld1q_f32(&y01[i + 4]); in dual_inner_prod_neon() [all …]
|
/dports/audio/opus/opus-1.3.1/celt/arm/ |
H A D | pitch_neon_intr.c | 204 x_f32x4 = vld1q_f32(&x[i]); 205 y_f32x4 = vld1q_f32(&y[i]); 207 x_f32x4 = vld1q_f32(&x[i + 4]); 208 y_f32x4 = vld1q_f32(&y[i + 4]); 213 const float32x4_t x_f32x4 = vld1q_f32(&x[i]); 245 x_f32x4 = vld1q_f32(&x[i]); 246 y01_f32x4 = vld1q_f32(&y01[i]); 247 y02_f32x4 = vld1q_f32(&y02[i]); 250 x_f32x4 = vld1q_f32(&x[i + 4]); 251 y01_f32x4 = vld1q_f32(&y01[i + 4]); [all …]
|
/dports/mail/thunderbird/thunderbird-91.8.0/media/libopus/celt/arm/ |
H A D | pitch_neon_intr.c | 204 x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 205 y_f32x4 = vld1q_f32(&y[i]); in celt_inner_prod_neon() 207 x_f32x4 = vld1q_f32(&x[i + 4]); in celt_inner_prod_neon() 208 y_f32x4 = vld1q_f32(&y[i + 4]); in celt_inner_prod_neon() 213 const float32x4_t x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 245 x_f32x4 = vld1q_f32(&x[i]); in dual_inner_prod_neon() 246 y01_f32x4 = vld1q_f32(&y01[i]); in dual_inner_prod_neon() 247 y02_f32x4 = vld1q_f32(&y02[i]); in dual_inner_prod_neon() 250 x_f32x4 = vld1q_f32(&x[i + 4]); in dual_inner_prod_neon() 251 y01_f32x4 = vld1q_f32(&y01[i + 4]); in dual_inner_prod_neon() [all …]
|
/dports/lang/spidermonkey78/firefox-78.9.0/media/libopus/celt/arm/ |
H A D | pitch_neon_intr.c | 204 x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 205 y_f32x4 = vld1q_f32(&y[i]); in celt_inner_prod_neon() 207 x_f32x4 = vld1q_f32(&x[i + 4]); in celt_inner_prod_neon() 208 y_f32x4 = vld1q_f32(&y[i + 4]); in celt_inner_prod_neon() 213 const float32x4_t x_f32x4 = vld1q_f32(&x[i]); in celt_inner_prod_neon() 245 x_f32x4 = vld1q_f32(&x[i]); in dual_inner_prod_neon() 246 y01_f32x4 = vld1q_f32(&y01[i]); in dual_inner_prod_neon() 247 y02_f32x4 = vld1q_f32(&y02[i]); in dual_inner_prod_neon() 250 x_f32x4 = vld1q_f32(&x[i + 4]); in dual_inner_prod_neon() 251 y01_f32x4 = vld1q_f32(&y01[i + 4]); in dual_inner_prod_neon() [all …]
|
/dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/arm/ |
H A D | convolutiondepthwise_3x3.h | 47 float32x4_t _k012x = vld1q_f32(kernel0); in convdw3x3s1_neon() 48 float32x4_t _k345x = vld1q_f32(kernel0 + 3); in convdw3x3s1_neon() 49 float32x4_t _k678x = vld1q_f32(kernel0 + 6); in convdw3x3s1_neon() 434 float32x4_t _r00 = vld1q_f32(r0); in convdw3x3s1_neon() 435 float32x4_t _r10 = vld1q_f32(r1); in convdw3x3s1_neon() 436 float32x4_t _r20 = vld1q_f32(r2); in convdw3x3s1_neon() 437 float32x4_t _r30 = vld1q_f32(r3); in convdw3x3s1_neon() 783 float32x4_t _r00 = vld1q_f32(r0); in convdw3x3s1_neon() 784 float32x4_t _r10 = vld1q_f32(r1); in convdw3x3s1_neon() 785 float32x4_t _r20 = vld1q_f32(r2); in convdw3x3s1_neon() [all …]
|
H A D | convolution_3x3_pack4to1.h | 2340 float32x4_t _k00_0 = vld1q_f32(k0); in conv3x3s1_pack4to1_neon() 2350 float32x4_t _k00_1 = vld1q_f32(k1); in conv3x3s1_pack4to1_neon() 2756 float32x4_t _k00 = vld1q_f32(k0); in conv3x3s1_pack4to1_neon() 2757 float32x4_t _k01 = vld1q_f32(k0 + 4); in conv3x3s1_pack4to1_neon() 2758 float32x4_t _k02 = vld1q_f32(k0 + 8); in conv3x3s1_pack4to1_neon() 2759 float32x4_t _k10 = vld1q_f32(k0 + 12); in conv3x3s1_pack4to1_neon() 2760 float32x4_t _k11 = vld1q_f32(k0 + 16); in conv3x3s1_pack4to1_neon() 2761 float32x4_t _k12 = vld1q_f32(k0 + 20); in conv3x3s1_pack4to1_neon() 2762 float32x4_t _k20 = vld1q_f32(k0 + 24); in conv3x3s1_pack4to1_neon() 2763 float32x4_t _k21 = vld1q_f32(k0 + 28); in conv3x3s1_pack4to1_neon() [all …]
|
/dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/arm/ |
H A D | convolutiondepthwise_3x3.h | 47 float32x4_t _k012x = vld1q_f32(kernel0); in convdw3x3s1_neon() 48 float32x4_t _k345x = vld1q_f32(kernel0 + 3); in convdw3x3s1_neon() 49 float32x4_t _k678x = vld1q_f32(kernel0 + 6); in convdw3x3s1_neon() 434 float32x4_t _r00 = vld1q_f32(r0); in convdw3x3s1_neon() 435 float32x4_t _r10 = vld1q_f32(r1); in convdw3x3s1_neon() 436 float32x4_t _r20 = vld1q_f32(r2); in convdw3x3s1_neon() 437 float32x4_t _r30 = vld1q_f32(r3); in convdw3x3s1_neon() 783 float32x4_t _r00 = vld1q_f32(r0); in convdw3x3s1_neon() 784 float32x4_t _r10 = vld1q_f32(r1); in convdw3x3s1_neon() 785 float32x4_t _r20 = vld1q_f32(r2); in convdw3x3s1_neon() [all …]
|
H A D | convolution_3x3_pack4to1.h | 2340 float32x4_t _k00_0 = vld1q_f32(k0); in conv3x3s1_pack4to1_neon() 2350 float32x4_t _k00_1 = vld1q_f32(k1); in conv3x3s1_pack4to1_neon() 2756 float32x4_t _k00 = vld1q_f32(k0); in conv3x3s1_pack4to1_neon() 2757 float32x4_t _k01 = vld1q_f32(k0 + 4); in conv3x3s1_pack4to1_neon() 2758 float32x4_t _k02 = vld1q_f32(k0 + 8); in conv3x3s1_pack4to1_neon() 2759 float32x4_t _k10 = vld1q_f32(k0 + 12); in conv3x3s1_pack4to1_neon() 2760 float32x4_t _k11 = vld1q_f32(k0 + 16); in conv3x3s1_pack4to1_neon() 2761 float32x4_t _k12 = vld1q_f32(k0 + 20); in conv3x3s1_pack4to1_neon() 2762 float32x4_t _k20 = vld1q_f32(k0 + 24); in conv3x3s1_pack4to1_neon() 2763 float32x4_t _k21 = vld1q_f32(k0 + 28); in conv3x3s1_pack4to1_neon() [all …]
|
/dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/arm/ |
H A D | convolutiondepthwise_3x3.h | 47 float32x4_t _k012x = vld1q_f32(kernel0); in convdw3x3s1_neon() 48 float32x4_t _k345x = vld1q_f32(kernel0 + 3); in convdw3x3s1_neon() 49 float32x4_t _k678x = vld1q_f32(kernel0 + 6); in convdw3x3s1_neon() 434 float32x4_t _r00 = vld1q_f32(r0); in convdw3x3s1_neon() 435 float32x4_t _r10 = vld1q_f32(r1); in convdw3x3s1_neon() 436 float32x4_t _r20 = vld1q_f32(r2); in convdw3x3s1_neon() 437 float32x4_t _r30 = vld1q_f32(r3); in convdw3x3s1_neon() 783 float32x4_t _r00 = vld1q_f32(r0); in convdw3x3s1_neon() 784 float32x4_t _r10 = vld1q_f32(r1); in convdw3x3s1_neon() 785 float32x4_t _r20 = vld1q_f32(r2); in convdw3x3s1_neon() [all …]
|
H A D | convolution_3x3_pack4to1.h | 2340 float32x4_t _k00_0 = vld1q_f32(k0); in conv3x3s1_pack4to1_neon() 2350 float32x4_t _k00_1 = vld1q_f32(k1); in conv3x3s1_pack4to1_neon() 2756 float32x4_t _k00 = vld1q_f32(k0); in conv3x3s1_pack4to1_neon() 2757 float32x4_t _k01 = vld1q_f32(k0 + 4); in conv3x3s1_pack4to1_neon() 2758 float32x4_t _k02 = vld1q_f32(k0 + 8); in conv3x3s1_pack4to1_neon() 2759 float32x4_t _k10 = vld1q_f32(k0 + 12); in conv3x3s1_pack4to1_neon() 2760 float32x4_t _k11 = vld1q_f32(k0 + 16); in conv3x3s1_pack4to1_neon() 2761 float32x4_t _k12 = vld1q_f32(k0 + 20); in conv3x3s1_pack4to1_neon() 2762 float32x4_t _k20 = vld1q_f32(k0 + 24); in conv3x3s1_pack4to1_neon() 2763 float32x4_t _k21 = vld1q_f32(k0 + 28); in conv3x3s1_pack4to1_neon() [all …]
|
/dports/misc/ncnn/ncnn-20211208/src/layer/arm/ |
H A D | convolutiondepthwise_3x3.h | 47 float32x4_t _k012x = vld1q_f32(kernel0); in convdw3x3s1_neon() 48 float32x4_t _k345x = vld1q_f32(kernel0 + 3); in convdw3x3s1_neon() 49 float32x4_t _k678x = vld1q_f32(kernel0 + 6); in convdw3x3s1_neon() 434 float32x4_t _r00 = vld1q_f32(r0); in convdw3x3s1_neon() 435 float32x4_t _r10 = vld1q_f32(r1); in convdw3x3s1_neon() 436 float32x4_t _r20 = vld1q_f32(r2); in convdw3x3s1_neon() 437 float32x4_t _r30 = vld1q_f32(r3); in convdw3x3s1_neon() 783 float32x4_t _r00 = vld1q_f32(r0); in convdw3x3s1_neon() 784 float32x4_t _r10 = vld1q_f32(r1); in convdw3x3s1_neon() 785 float32x4_t _r20 = vld1q_f32(r2); in convdw3x3s1_neon() [all …]
|
H A D | convolution_3x3_pack4to1.h | 2340 float32x4_t _k00_0 = vld1q_f32(k0); in conv3x3s1_pack4to1_neon() 2350 float32x4_t _k00_1 = vld1q_f32(k1); in conv3x3s1_pack4to1_neon() 2756 float32x4_t _k00 = vld1q_f32(k0); in conv3x3s1_pack4to1_neon() 2757 float32x4_t _k01 = vld1q_f32(k0 + 4); in conv3x3s1_pack4to1_neon() 2758 float32x4_t _k02 = vld1q_f32(k0 + 8); in conv3x3s1_pack4to1_neon() 2759 float32x4_t _k10 = vld1q_f32(k0 + 12); in conv3x3s1_pack4to1_neon() 2760 float32x4_t _k11 = vld1q_f32(k0 + 16); in conv3x3s1_pack4to1_neon() 2761 float32x4_t _k12 = vld1q_f32(k0 + 20); in conv3x3s1_pack4to1_neon() 2762 float32x4_t _k20 = vld1q_f32(k0 + 24); in conv3x3s1_pack4to1_neon() 2763 float32x4_t _k21 = vld1q_f32(k0 + 28); in conv3x3s1_pack4to1_neon() [all …]
|
/dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/arm/ |
H A D | convolutiondepthwise_3x3.h | 47 float32x4_t _k012x = vld1q_f32(kernel0); in convdw3x3s1_neon() 48 float32x4_t _k345x = vld1q_f32(kernel0 + 3); in convdw3x3s1_neon() 49 float32x4_t _k678x = vld1q_f32(kernel0 + 6); in convdw3x3s1_neon() 434 float32x4_t _r00 = vld1q_f32(r0); in convdw3x3s1_neon() 435 float32x4_t _r10 = vld1q_f32(r1); in convdw3x3s1_neon() 436 float32x4_t _r20 = vld1q_f32(r2); in convdw3x3s1_neon() 437 float32x4_t _r30 = vld1q_f32(r3); in convdw3x3s1_neon() 783 float32x4_t _r00 = vld1q_f32(r0); in convdw3x3s1_neon() 784 float32x4_t _r10 = vld1q_f32(r1); in convdw3x3s1_neon() 785 float32x4_t _r20 = vld1q_f32(r2); in convdw3x3s1_neon() [all …]
|
H A D | convolution_3x3_pack4to1.h | 2340 float32x4_t _k00_0 = vld1q_f32(k0); in conv3x3s1_pack4to1_neon() 2350 float32x4_t _k00_1 = vld1q_f32(k1); in conv3x3s1_pack4to1_neon() 2756 float32x4_t _k00 = vld1q_f32(k0); in conv3x3s1_pack4to1_neon() 2757 float32x4_t _k01 = vld1q_f32(k0 + 4); in conv3x3s1_pack4to1_neon() 2758 float32x4_t _k02 = vld1q_f32(k0 + 8); in conv3x3s1_pack4to1_neon() 2759 float32x4_t _k10 = vld1q_f32(k0 + 12); in conv3x3s1_pack4to1_neon() 2760 float32x4_t _k11 = vld1q_f32(k0 + 16); in conv3x3s1_pack4to1_neon() 2761 float32x4_t _k12 = vld1q_f32(k0 + 20); in conv3x3s1_pack4to1_neon() 2762 float32x4_t _k20 = vld1q_f32(k0 + 24); in conv3x3s1_pack4to1_neon() 2763 float32x4_t _k21 = vld1q_f32(k0 + 28); in conv3x3s1_pack4to1_neon() [all …]
|