Home
last modified time | relevance | path

Searched refs:vld1_dup_u32 (Results 1 – 25 of 31) sorted by relevance

12

/aosp_15_r20/external/XNNPACK/src/u32-filterbank-accumulate/
H A Dneon.c.in36 const uint32x2_t vi = vld1_dup_u32(input); input += 1;
37 const uint16x4_t vw = vreinterpret_u16_u32(vld1_dup_u32((const void*) weights)); weights += 2;
59 const uint32x2_t vi = vld1_dup_u32(input); input += 1;
60 … const uint16x4_t vw = vreinterpret_u16_u32(vld1_dup_u32((const void*) weights)); weights += 2;
67 const uint32x2_t vi = vld1_dup_u32(input); input += 1;
68 … const uint16x4_t vw = vreinterpret_u16_u32(vld1_dup_u32((const void*) weights)); weights += 2;
/aosp_15_r20/external/XNNPACK/src/u32-filterbank-accumulate/gen/
H A Dneon-x1.c39 const uint32x2_t vi = vld1_dup_u32(input); input += 1; in xnn_u32_filterbank_accumulate_ukernel__neon_x1()
40 const uint16x4_t vw = vreinterpret_u16_u32(vld1_dup_u32((const void*) weights)); weights += 2; in xnn_u32_filterbank_accumulate_ukernel__neon_x1()
52 const uint32x2_t vi = vld1_dup_u32(input); input += 1; in xnn_u32_filterbank_accumulate_ukernel__neon_x1()
53 const uint16x4_t vw = vreinterpret_u16_u32(vld1_dup_u32((const void*) weights)); weights += 2; in xnn_u32_filterbank_accumulate_ukernel__neon_x1()
H A Dneon-x2.c39 const uint32x2_t vi = vld1_dup_u32(input); input += 1; in xnn_u32_filterbank_accumulate_ukernel__neon_x2()
40 const uint16x4_t vw = vreinterpret_u16_u32(vld1_dup_u32((const void*) weights)); weights += 2; in xnn_u32_filterbank_accumulate_ukernel__neon_x2()
61 const uint32x2_t vi = vld1_dup_u32(input); input += 1; in xnn_u32_filterbank_accumulate_ukernel__neon_x2()
62 const uint16x4_t vw = vreinterpret_u16_u32(vld1_dup_u32((const void*) weights)); weights += 2; in xnn_u32_filterbank_accumulate_ukernel__neon_x2()
/aosp_15_r20/external/pytorch/aten/src/ATen/native/quantized/cpu/qnnpack/src/q8gemm/
H A D4x-sumrows-neon.c80 vld1_dup_u32(__builtin_assume_aligned((const uint32_t*)a0, 1)))))); in pytorch_q8sumrows_ukernel_4x__neon()
85 vld1_dup_u32(__builtin_assume_aligned((const uint32_t*)a1, 1)))))); in pytorch_q8sumrows_ukernel_4x__neon()
90 vld1_dup_u32(__builtin_assume_aligned((const uint32_t*)a2, 1)))))); in pytorch_q8sumrows_ukernel_4x__neon()
95 vld1_dup_u32(__builtin_assume_aligned((const uint32_t*)a3, 1)))))); in pytorch_q8sumrows_ukernel_4x__neon()
H A D6x4-neon.c246 const uint8x8_t vb0123c0 = vreinterpret_u8_u32(vld1_dup_u32(w)); in pytorch_q8gemm_ukernel_6x4__neon()
265 const uint8x8_t vb0123c1 = vreinterpret_u8_u32(vld1_dup_u32(w)); in pytorch_q8gemm_ukernel_6x4__neon()
284 const uint8x8_t vb0123c2 = vreinterpret_u8_u32(vld1_dup_u32(w)); in pytorch_q8gemm_ukernel_6x4__neon()
303 const uint8x8_t vb0123c3 = vreinterpret_u8_u32(vld1_dup_u32(w)); in pytorch_q8gemm_ukernel_6x4__neon()
322 const uint8x8_t vb0123c4 = vreinterpret_u8_u32(vld1_dup_u32(w)); in pytorch_q8gemm_ukernel_6x4__neon()
341 const uint8x8_t vb0123c5 = vreinterpret_u8_u32(vld1_dup_u32(w)); in pytorch_q8gemm_ukernel_6x4__neon()
360 const uint8x8_t vb0123c6 = vreinterpret_u8_u32(vld1_dup_u32(w)); in pytorch_q8gemm_ukernel_6x4__neon()
/aosp_15_r20/external/XNNPACK/src/x32-zip/
H A Dx3-neon.c43 uint32x2_t vxy = vld1_dup_u32(x); in xnn_x32_zip_x3_ukernel__neon()
44 const uint32x2_t vz = vld1_dup_u32(z); in xnn_x32_zip_x3_ukernel__neon()
H A Dxm-neon.c86 const uint32x2_t vx = vld1_dup_u32(x); in xnn_x32_zip_xm_ukernel__neon()
87 const uint32x2_t vz = vld1_dup_u32(z); in xnn_x32_zip_xm_ukernel__neon()
H A Dx2-neon.c40 uint32x2_t vxy = vld1_dup_u32(x); in xnn_x32_zip_x2_ukernel__neon()
/aosp_15_r20/external/XNNPACK/src/x32-packx/
H A Dx4-neon-st4.c48 const uint32x2_t vx00 = vld1_dup_u32(x0); x0 += 1; in xnn_x32_packx_ukernel_4x__neon_st4()
49 const uint32x2_t vx22 = vld1_dup_u32(x2); x2 += 1; in xnn_x32_packx_ukernel_4x__neon_st4()
/aosp_15_r20/external/libaom/aom_dsp/arm/
H A Dloopfilter_neon.c930 p3q3 = vreinterpret_u8_u32(vld1_dup_u32((uint32_t *)(src - 4 * stride))); in aom_lpf_horizontal_8_neon()
931 p2q2 = vreinterpret_u8_u32(vld1_dup_u32((uint32_t *)(src - 3 * stride))); in aom_lpf_horizontal_8_neon()
932 p1q1 = vreinterpret_u8_u32(vld1_dup_u32((uint32_t *)(src - 2 * stride))); in aom_lpf_horizontal_8_neon()
933 p0q0 = vreinterpret_u8_u32(vld1_dup_u32((uint32_t *)(src - 1 * stride))); in aom_lpf_horizontal_8_neon()
978 p2q2 = vreinterpret_u8_u32(vld1_dup_u32((uint32_t *)(src - 3 * stride))); in aom_lpf_horizontal_6_neon()
979 p1q1 = vreinterpret_u8_u32(vld1_dup_u32((uint32_t *)(src - 2 * stride))); in aom_lpf_horizontal_6_neon()
980 p0q0 = vreinterpret_u8_u32(vld1_dup_u32((uint32_t *)(src - 1 * stride))); in aom_lpf_horizontal_6_neon()
/aosp_15_r20/external/libhevc/encoder/arm/
H A Dihevce_coarse_layer_sad_neon.c161 a[i] = vld1_dup_u32((uint32_t *)pu1_inp); in hme_store_4x4_sads_high_speed_neon()
312 a[i] = vld1_dup_u32((uint32_t *)pu1_inp); in hme_store_4x4_sads_high_quality_neon()
/aosp_15_r20/external/libaom/av1/common/arm/
H A Dcfl_neon.c30 return vreinterpret_u8_u32(vld1_dup_u32((const uint32_t *)ptr)); in vldh_dup_u8()
/aosp_15_r20/external/clang/test/CodeGen/
H A Daarch64-neon-ldst-one.c184 return vld1_dup_u32(a); in test_vld1_dup_u32()
H A Darm_neon_intrinsics.c4466 return vld1_dup_u32(a); in test_vld1_dup_u32()
/aosp_15_r20/external/eigen/Eigen/src/Core/arch/NEON/
H A DPacketMath.h1817 { return vld1_dup_u32(from); }
1819 { return vcombine_u32(vld1_dup_u32(from), vld1_dup_u32(from+1)); }
2091 Packet2ui res = vld1_dup_u32(from);
/aosp_15_r20/external/llvm/test/CodeGen/AArch64/
H A Darm64-ld1.ll912 ; Add rdar://13098923 test case: vld1_dup_u32 doesn't generate ld1r.2s
/aosp_15_r20/external/neon_2_sse/
H A DNEON_2_SSE.h1289 _NEON2SSESTORAGE uint32x2_t vld1_dup_u32(__transfersize(1) uint32_t const * ptr); // VLD1.32 {d0[]}…
9511 _NEON2SSESTORAGE uint32x2_t vld1_dup_u32(__transfersize(1) uint32_t const * ptr); // VLD1.32 {d0[]}…
9512 _NEON2SSE_INLINE _NEON2SSE_PERFORMANCE_WARNING(uint32x2_t vld1_dup_u32(__transfersize(1) uint32_t c… in _NEON2SSE_PERFORMANCE_WARNING()
9537 #define vld1_dup_s32(ptr) vld1_dup_u32((uint32_t*)ptr)
/aosp_15_r20/prebuilts/rust/linux-x86/1.81.0.u1/lib/rustlib/src/rust/library/stdarch/crates/core_arch/src/arm_shared/neon/
H A Dmod.rs2223 pub unsafe fn vld1_dup_u32(ptr: *const u32) -> uint32x2_t { in vld1_dup_u32() function
10535 let r: u32x2 = transmute(vld1_dup_u32(&elem)); in test_vld1_dup_u32()
/aosp_15_r20/prebuilts/rust/linux-musl-x86/1.81.0/lib/rustlib/src/rust/library/stdarch/crates/core_arch/src/arm_shared/neon/
H A Dmod.rs2223 pub unsafe fn vld1_dup_u32(ptr: *const u32) -> uint32x2_t { in vld1_dup_u32() function
10535 let r: u32x2 = transmute(vld1_dup_u32(&elem)); in test_vld1_dup_u32()
/aosp_15_r20/prebuilts/rust/linux-musl-x86/1.80.1/lib/rustlib/src/rust/library/stdarch/crates/core_arch/src/arm_shared/neon/
H A Dmod.rs2223 pub unsafe fn vld1_dup_u32(ptr: *const u32) -> uint32x2_t { in vld1_dup_u32() function
10535 let r: u32x2 = transmute(vld1_dup_u32(&elem)); in test_vld1_dup_u32()
/aosp_15_r20/prebuilts/rust/linux-x86/1.81.0/lib/rustlib/src/rust/library/stdarch/crates/core_arch/src/arm_shared/neon/
H A Dmod.rs2223 pub unsafe fn vld1_dup_u32(ptr: *const u32) -> uint32x2_t { in vld1_dup_u32() function
10535 let r: u32x2 = transmute(vld1_dup_u32(&elem)); in test_vld1_dup_u32()
/aosp_15_r20/prebuilts/rust/linux-x86/1.80.1/lib/rustlib/src/rust/library/stdarch/crates/core_arch/src/arm_shared/neon/
H A Dmod.rs2223 pub unsafe fn vld1_dup_u32(ptr: *const u32) -> uint32x2_t { in vld1_dup_u32() function
10535 let r: u32x2 = transmute(vld1_dup_u32(&elem)); in test_vld1_dup_u32()
/aosp_15_r20/out/soong/.intermediates/external/clang/clang-gen-arm-neon/gen/clang/Basic/
Darm_neon.h8029 #define vld1_dup_u32(__p0) __extension__ ({ \ macro
8035 #define vld1_dup_u32(__p0) __extension__ ({ \ macro
/aosp_15_r20/prebuilts/clang/host/linux-x86/clang-3289846/lib64/clang/3.8/include/
Darm_neon.h8029 #define vld1_dup_u32(__p0) __extension__ ({ \ macro
8035 #define vld1_dup_u32(__p0) __extension__ ({ \ macro
/aosp_15_r20/prebuilts/sdk/renderscript/clang-include/
H A Darm_neon.h8117 #define vld1_dup_u32(__p0) __extension__ ({ \ macro
8123 #define vld1_dup_u32(__p0) __extension__ ({ \ macro

12