/external/XNNPACK/src/f32-vrnd/ |
D | vrndne-neon.c.in | 37 float32x4_t vrndabsx${ABC[N:N+4]} = vaddq_f32(vabsx${ABC[N:N+4]}, vmagic_number); 43 vrndabsx${ABC[N:N+4]} = vsubq_f32(vrndabsx${ABC[N:N+4]}, vmagic_number); 46 …t32x4_t vy${ABC[N:N+4]} = vbslq_f32(vrndmask${ABC[N:N+4]}, vx${ABC[N:N+4]}, vrndabsx${ABC[N:N+4]}); 56 float32x4_t vrndabsx = vaddq_f32(vabsx, vmagic_number); variable 58 vrndabsx = vsubq_f32(vrndabsx, vmagic_number); 59 const float32x4_t vy = vbslq_f32(vrndmask, vx, vrndabsx); 66 float32x4_t vrndabsx = vaddq_f32(vabsx, vmagic_number); variable 68 vrndabsx = vsubq_f32(vrndabsx, vmagic_number); 69 const float32x4_t vy = vbslq_f32(vrndmask, vx, vrndabsx);
|
D | vrndz-wasmsimd-addsub.c.in | 44 …const v128_t vrndabsx${ABC[N:N+4]} = wasm_f32x4_sub(wasm_f32x4_add(vabsx${ABC[N:N+4]}, vmagic_numb… 47 …justment${ABC[N:N+4]} = wasm_v128_and(wasm_f32x4_lt(vabsx${ABC[N:N+4]}, vrndabsx${ABC[N:N+4]}), vo… 50 …const v128_t vflrabsx${ABC[N:N+4]} = wasm_f32x4_sub(vrndabsx${ABC[N:N+4]}, vadjustment${ABC[N:N+4]… 66 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); variable 67 const v128_t vadjustment = wasm_v128_and(wasm_f32x4_lt(vabsx, vrndabsx), vone); 68 const v128_t vflrabsx = wasm_f32x4_sub(vrndabsx, vadjustment); 79 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); variable 80 const v128_t vadjustment = wasm_v128_and(wasm_f32x4_lt(vabsx, vrndabsx), vone); 81 const v128_t vflrabsx = wasm_f32x4_sub(vrndabsx, vadjustment);
|
D | vrndne-wasmsimd-addsub.c.in | 43 …const v128_t vrndabsx${ABC[N:N+4]} = wasm_f32x4_sub(wasm_f32x4_add(vabsx${ABC[N:N+4]}, vmagic_numb… 46 …const v128_t vy${ABC[N:N+4]} = wasm_v128_bitselect(vx${ABC[N:N+4]}, vrndabsx${ABC[N:N+4]}, vrndmas… 59 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); variable 60 const v128_t vy = wasm_v128_bitselect(vx, vrndabsx, vrndmask); 70 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); variable 71 v128_t vy = wasm_v128_bitselect(vx, vrndabsx, vrndmask);
|
D | vrndd-wasmsimd-addsub.c.in | 44 …const v128_t vrndabsx${ABC[N:N+4]} = wasm_f32x4_sub(wasm_f32x4_add(vabsx${ABC[N:N+4]}, vmagic_numb… 47 …const v128_t vrndx${ABC[N:N+4]} = wasm_v128_bitselect(vx${ABC[N:N+4]}, vrndabsx${ABC[N:N+4]}, vrnd… 63 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); variable 64 const v128_t vrndx = wasm_v128_bitselect(vx, vrndabsx, vrndmask); 75 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); variable 76 const v128_t vrndx = wasm_v128_bitselect(vx, vrndabsx, vrndmask);
|
D | vrndu-wasmsimd-addsub.c.in | 44 …const v128_t vrndabsx${ABC[N:N+4]} = wasm_f32x4_sub(wasm_f32x4_add(vabsx${ABC[N:N+4]}, vmagic_numb… 47 …const v128_t vrndx${ABC[N:N+4]} = wasm_v128_bitselect(vx${ABC[N:N+4]}, vrndabsx${ABC[N:N+4]}, vrnd… 69 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); variable 70 const v128_t vrndx = wasm_v128_bitselect(vx, vrndabsx, vrndmask); 83 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); variable 84 const v128_t vrndx = wasm_v128_bitselect(vx, vrndabsx, vrndmask);
|
/external/XNNPACK/src/f32-vrnd/gen/ |
D | vrndne-neon-x8.c | 57 float32x4_t vrndabsx = vaddq_f32(vabsx, vmagic_number); in xnn_f32_vrndne_ukernel__neon_x8() local 59 vrndabsx = vsubq_f32(vrndabsx, vmagic_number); in xnn_f32_vrndne_ukernel__neon_x8() 60 const float32x4_t vy = vbslq_f32(vrndmask, vx, vrndabsx); in xnn_f32_vrndne_ukernel__neon_x8() 67 float32x4_t vrndabsx = vaddq_f32(vabsx, vmagic_number); in xnn_f32_vrndne_ukernel__neon_x8() local 69 vrndabsx = vsubq_f32(vrndabsx, vmagic_number); in xnn_f32_vrndne_ukernel__neon_x8() 70 const float32x4_t vy = vbslq_f32(vrndmask, vx, vrndabsx); in xnn_f32_vrndne_ukernel__neon_x8()
|
D | vrndz-wasmsimd-addsub-x4.c | 37 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); in xnn_f32_vrndz_ukernel__wasmsimd_addsub_x4() local 38 const v128_t vadjustment = wasm_v128_and(wasm_f32x4_lt(vabsx, vrndabsx), vone); in xnn_f32_vrndz_ukernel__wasmsimd_addsub_x4() 39 const v128_t vflrabsx = wasm_f32x4_sub(vrndabsx, vadjustment); in xnn_f32_vrndz_ukernel__wasmsimd_addsub_x4() 50 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); in xnn_f32_vrndz_ukernel__wasmsimd_addsub_x4() local 51 const v128_t vadjustment = wasm_v128_and(wasm_f32x4_lt(vabsx, vrndabsx), vone); in xnn_f32_vrndz_ukernel__wasmsimd_addsub_x4() 52 const v128_t vflrabsx = wasm_f32x4_sub(vrndabsx, vadjustment); in xnn_f32_vrndz_ukernel__wasmsimd_addsub_x4()
|
D | vrndz-wasmsimd-addsub-x8.c | 64 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); in xnn_f32_vrndz_ukernel__wasmsimd_addsub_x8() local 65 const v128_t vadjustment = wasm_v128_and(wasm_f32x4_lt(vabsx, vrndabsx), vone); in xnn_f32_vrndz_ukernel__wasmsimd_addsub_x8() 66 const v128_t vflrabsx = wasm_f32x4_sub(vrndabsx, vadjustment); in xnn_f32_vrndz_ukernel__wasmsimd_addsub_x8() 77 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); in xnn_f32_vrndz_ukernel__wasmsimd_addsub_x8() local 78 const v128_t vadjustment = wasm_v128_and(wasm_f32x4_lt(vabsx, vrndabsx), vone); in xnn_f32_vrndz_ukernel__wasmsimd_addsub_x8() 79 const v128_t vflrabsx = wasm_f32x4_sub(vrndabsx, vadjustment); in xnn_f32_vrndz_ukernel__wasmsimd_addsub_x8()
|
D | vrndne-wasmsimd-addsub-x4.c | 36 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); in xnn_f32_vrndne_ukernel__wasmsimd_addsub_x4() local 37 const v128_t vy = wasm_v128_bitselect(vx, vrndabsx, vrndmask); in xnn_f32_vrndne_ukernel__wasmsimd_addsub_x4() 47 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); in xnn_f32_vrndne_ukernel__wasmsimd_addsub_x4() local 48 v128_t vy = wasm_v128_bitselect(vx, vrndabsx, vrndmask); in xnn_f32_vrndne_ukernel__wasmsimd_addsub_x4()
|
D | vrndne-neon-x4.c | 49 float32x4_t vrndabsx = vaddq_f32(vabsx, vmagic_number); in xnn_f32_vrndne_ukernel__neon_x4() local 51 vrndabsx = vsubq_f32(vrndabsx, vmagic_number); in xnn_f32_vrndne_ukernel__neon_x4() 52 const float32x4_t vy = vbslq_f32(vrndmask, vx, vrndabsx); in xnn_f32_vrndne_ukernel__neon_x4()
|
D | vrndd-wasmsimd-addsub-x4.c | 37 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); in xnn_f32_vrndd_ukernel__wasmsimd_addsub_x4() local 38 const v128_t vrndx = wasm_v128_bitselect(vx, vrndabsx, vrndmask); in xnn_f32_vrndd_ukernel__wasmsimd_addsub_x4() 49 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); in xnn_f32_vrndd_ukernel__wasmsimd_addsub_x4() local 50 const v128_t vrndx = wasm_v128_bitselect(vx, vrndabsx, vrndmask); in xnn_f32_vrndd_ukernel__wasmsimd_addsub_x4()
|
D | vrndu-wasmsimd-addsub-x4.c | 37 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); in xnn_f32_vrndu_ukernel__wasmsimd_addsub_x4() local 38 const v128_t vrndx = wasm_v128_bitselect(vx, vrndabsx, vrndmask); in xnn_f32_vrndu_ukernel__wasmsimd_addsub_x4() 51 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); in xnn_f32_vrndu_ukernel__wasmsimd_addsub_x4() local 52 const v128_t vrndx = wasm_v128_bitselect(vx, vrndabsx, vrndmask); in xnn_f32_vrndu_ukernel__wasmsimd_addsub_x4()
|
D | vrndne-wasmsimd-addsub-x8.c | 57 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); in xnn_f32_vrndne_ukernel__wasmsimd_addsub_x8() local 58 const v128_t vy = wasm_v128_bitselect(vx, vrndabsx, vrndmask); in xnn_f32_vrndne_ukernel__wasmsimd_addsub_x8() 68 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); in xnn_f32_vrndne_ukernel__wasmsimd_addsub_x8() local 69 v128_t vy = wasm_v128_bitselect(vx, vrndabsx, vrndmask); in xnn_f32_vrndne_ukernel__wasmsimd_addsub_x8()
|
D | vrndd-wasmsimd-addsub-x8.c | 61 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); in xnn_f32_vrndd_ukernel__wasmsimd_addsub_x8() local 62 const v128_t vrndx = wasm_v128_bitselect(vx, vrndabsx, vrndmask); in xnn_f32_vrndd_ukernel__wasmsimd_addsub_x8() 73 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); in xnn_f32_vrndd_ukernel__wasmsimd_addsub_x8() local 74 const v128_t vrndx = wasm_v128_bitselect(vx, vrndabsx, vrndmask); in xnn_f32_vrndd_ukernel__wasmsimd_addsub_x8()
|
D | vrndu-wasmsimd-addsub-x8.c | 67 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); in xnn_f32_vrndu_ukernel__wasmsimd_addsub_x8() local 68 const v128_t vrndx = wasm_v128_bitselect(vx, vrndabsx, vrndmask); in xnn_f32_vrndu_ukernel__wasmsimd_addsub_x8() 81 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); in xnn_f32_vrndu_ukernel__wasmsimd_addsub_x8() local 82 const v128_t vrndx = wasm_v128_bitselect(vx, vrndabsx, vrndmask); in xnn_f32_vrndu_ukernel__wasmsimd_addsub_x8()
|
/external/XNNPACK/src/math/ |
D | roundz-scalar-addsub.c | 39 const float vrndabsx = (vabsx + vmagic_number) - vmagic_number; in xnn_math_f32_roundz__scalar_addsub() local 43 const float vflrabsx = XNN_UNPREDICTABLE(vrndabsx <= vabsx) ? vrndabsx : vrndabsx - vone; in xnn_math_f32_roundz__scalar_addsub()
|
D | roundz-neon-addsub.c | 48 const float32x4_t vrndabsx = vsubq_f32(vaddq_f32(vabsx, vmagic_number), vmagic_number); in xnn_math_f32_roundz__neon_addsub() local 52 …const float32x4_t vadjustment = vreinterpretq_f32_u32(vandq_u32(vone, vcgtq_f32(vrndabsx, vabsx))); in xnn_math_f32_roundz__neon_addsub() 55 const float32x4_t vflrabsx = vsubq_f32(vrndabsx, vadjustment); in xnn_math_f32_roundz__neon_addsub()
|
D | roundz-wasmsimd-addsub.c | 48 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); in xnn_math_f32_roundz__wasmsimd_addsub() local 52 const v128_t vadjustment = wasm_v128_and(wasm_f32x4_gt(vrndabsx, vabsx), vone); in xnn_math_f32_roundz__wasmsimd_addsub() 55 const v128_t vflrabsx = wasm_f32x4_sub(vrndabsx, vadjustment); in xnn_math_f32_roundz__wasmsimd_addsub()
|
D | roundz-sse-addsub.c | 48 const __m128 vrndabsx = _mm_sub_ps(_mm_add_ps(vabsx, vmagic_number), vmagic_number); in xnn_math_f32_roundz__sse_addsub() local 52 const __m128 vadjustment = _mm_and_ps(vone, _mm_cmpgt_ps(vrndabsx, vabsx)); in xnn_math_f32_roundz__sse_addsub() 55 const __m128 vflrabsx = _mm_sub_ps(vrndabsx, vadjustment); in xnn_math_f32_roundz__sse_addsub()
|
D | roundne-scalar-addsub.c | 36 const float vrndabsx = (vabsx + vmagic_number) - vmagic_number; in xnn_math_f32_roundne__scalar_addsub() local 42 const float vabsy = XNN_UNPREDICTABLE(vabsx >= vmagic_number) ? vabsx : vrndabsx; in xnn_math_f32_roundne__scalar_addsub()
|
D | roundd-scalar-addsub.c | 44 const float vrndabsx = XNN_UNPREDICTABLE(vabsx >= vmagic_number) ? vabsx : vprerndabsx; in xnn_math_f32_roundd__scalar_addsub() local 46 const float vrndx = copysignf(vrndabsx, vx); in xnn_math_f32_roundd__scalar_addsub()
|
D | roundu-scalar-addsub.c | 44 const float vrndabsx = XNN_UNPREDICTABLE(vabsx >= vmagic_number) ? vabsx : vprerndabsx; in xnn_math_f32_roundu__scalar_addsub() local 46 const float vrndx = copysignf(vrndabsx, vx); in xnn_math_f32_roundu__scalar_addsub()
|
D | roundne-neon-addsub.c | 45 const float32x4_t vrndabsx = vsubq_f32(vaddq_f32(vabsx, vmagic_number), vmagic_number); in xnn_math_f32_roundne__neon_addsub() local 51 const float32x4_t vy = vbslq_f32(vrndmask, vx, vrndabsx); in xnn_math_f32_roundne__neon_addsub()
|
D | roundne-wasmsimd-addsub.c | 45 const v128_t vrndabsx = wasm_f32x4_sub(wasm_f32x4_add(vabsx, vmagic_number), vmagic_number); in xnn_math_f32_roundne__wasmsimd_addsub() local 51 const v128_t vy = wasm_v128_bitselect(vx, vrndabsx, vrndmask); in xnn_math_f32_roundne__wasmsimd_addsub()
|
D | roundne-sse-addsub.c | 45 const __m128 vrndabsx = _mm_sub_ps(_mm_add_ps(vabsx, vmagic_number), vmagic_number); in xnn_math_f32_roundne__sse_addsub() local 51 const __m128 vy = _mm_or_ps(_mm_and_ps(vrndabsx, vrndmask), _mm_andnot_ps(vrndmask, vx)); in xnn_math_f32_roundne__sse_addsub()
|