chris@160: /**
chris@160:  * Copyright (c) 2014, 2015, Enzien Audio Ltd.
chris@160:  *
chris@160:  * Permission to use, copy, modify, and/or distribute this software for any
chris@160:  * purpose with or without fee is hereby granted, provided that the above
chris@160:  * copyright notice and this permission notice appear in all copies.
chris@160:  *
chris@160:  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH
chris@160:  * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
chris@160:  * AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT,
chris@160:  * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
chris@160:  * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR
chris@160:  * OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
chris@160:  * PERFORMANCE OF THIS SOFTWARE.
chris@160:  */
chris@160: 
chris@160: #ifndef _SIGNAL_SAMPHOLD_H_
chris@160: #define _SIGNAL_SAMPHOLD_H_
chris@160: 
chris@160: #include "HvBase.h"
chris@160: 
chris@160: typedef struct SignalSamphold {
chris@160:   hv_bufferf_t s;
chris@160: } SignalSamphold;
chris@160: 
chris@160: hv_size_t sSamphold_init(SignalSamphold *o);
chris@160: 
chris@160: static inline void __hv_samphold_f(SignalSamphold *o, hv_bInf_t bIn0, hv_bInf_t bIn1, hv_bOutf_t bOut) {
chris@160: #if HV_SIMD_AVX
chris@160: #warning __hv_samphold_f() not implemented
chris@160: #elif HV_SIMD_SSE
chris@160:   switch (_mm_movemask_ps(bIn1)) {
chris@160:     default:
chris@160:     case 0x0: *bOut = o->s; break;
chris@160:     case 0x1: {
chris@160:       *bOut = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(0,0,0,0));
chris@160:       o->s = *bOut;
chris@160:       break;
chris@160:     }
chris@160:     case 0x2: {
chris@160:       const __m128 x = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(1,1,1,1));
chris@160:       *bOut = _mm_blend_ps(o->s, x, 0xE);
chris@160:       o->s = x;
chris@160:       break;
chris@160:     }
chris@160:     case 0x3: {
chris@160:       const __m128 x = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(1,1,1,1));
chris@160:       *bOut = _mm_blend_ps(bIn0, x, 0xC);
chris@160:       o->s = x;
chris@160:       break;
chris@160:     }
chris@160:     case 0x4: {
chris@160:       const __m128 x = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(2,2,2,2));
chris@160:       *bOut = _mm_blend_ps(o->s, x, 0xC);
chris@160:       o->s = x;
chris@160:       break;
chris@160:     }
chris@160:     case 0x5: {
chris@160:       *bOut = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(2,2,0,0));
chris@160:       o->s = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(2,2,2,2));
chris@160:       break;
chris@160:     }
chris@160:     case 0x6: {
chris@160:       const __m128 x = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(2,2,1,0));
chris@160:       *bOut = _mm_blend_ps(o->s, x, 0xE);
chris@160:       o->s = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(2,2,2,2));
chris@160:       break;
chris@160:     }
chris@160:     case 0x7: {
chris@160:       const __m128 x = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(2,2,2,2));
chris@160:       *bOut = _mm_blend_ps(bIn0, x, 0x8);
chris@160:       o->s = x;
chris@160:       break;
chris@160:     }
chris@160:     case 0x8: {
chris@160:       const __m128 x = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(3,3,3,3));
chris@160:       *bOut = _mm_blend_ps(o->s, x, 0x8);
chris@160:       o->s = x;
chris@160:       break;
chris@160:     }
chris@160:     case 0x9: {
chris@160:       *bOut = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(3,0,0,0));
chris@160:       o->s = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(3,3,3,3));
chris@160:       break;
chris@160:     }
chris@160:     case 0xA: {
chris@160:       const __m128 x = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(3,1,1,0));
chris@160:       *bOut = _mm_blend_ps(o->s, x, 0xE);
chris@160:       o->s = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(3,3,3,3));
chris@160:       break;
chris@160:     }
chris@160:     case 0xB: {
chris@160:       *bOut = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(3,1,1,0));
chris@160:       o->s = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(3,3,3,3));
chris@160:       break;
chris@160:     }
chris@160:     case 0xC: {
chris@160:       *bOut = _mm_blend_ps(o->s, bIn0, 0xC);
chris@160:       o->s = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(3,3,3,3));
chris@160:       break;
chris@160:     }
chris@160:     case 0xD: {
chris@160:       *bOut = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(3,2,0,0));
chris@160:       o->s = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(3,3,3,3));
chris@160:       break;
chris@160:     }
chris@160:     case 0xE: {
chris@160:       *bOut = _mm_blend_ps(o->s, bIn0, 0xE);
chris@160:       o->s = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(3,3,3,3));
chris@160:       break;
chris@160:     }
chris@160:     case 0xF: {
chris@160:       *bOut = bIn0;
chris@160:       o->s = _mm_shuffle_ps(bIn0, bIn0, _MM_SHUFFLE(3,3,3,3));
chris@160:       break;
chris@160:     }
chris@160:   }
chris@160: #elif HV_SIMD_NEON
chris@160:   uint32x4_t mmA = vandq_u32(
chris@160:       vreinterpretq_u32_f32(bIn1), (uint32x4_t) {0x1, 0x2, 0x4, 0x8}); // [0 1 2 3]
chris@160:   uint32x4_t mmB = vextq_u32(mmA, mmA, 2);                             // [2 3 0 1]
chris@160:   uint32x4_t mmC = vorrq_u32(mmA, mmB);                                // [0+2 1+3 0+2 1+3]
chris@160:   uint32x4_t mmD = vextq_u32(mmC, mmC, 3);                             // [1+3 0+2 1+3 0+2]
chris@160:   uint32x4_t mmE = vorrq_u32(mmC, mmD);                                // [0+1+2+3 ...]
chris@160:   uint32_t movemask = vgetq_lane_u32(mmE, 0);
chris@160:   switch (movemask) {
chris@160:     default:
chris@160:     case 0x0: *bOut = o->s; break;
chris@160:     case 0x1: {
chris@160:       *bOut = vdupq_n_f32(vgetq_lane_f32(bIn0,0));
chris@160:       o->s = *bOut;
chris@160:       break;
chris@160:     }
chris@160:     case 0x2: {
chris@160:       const float32x4_t x = vdupq_n_f32(vgetq_lane_f32(bIn0,1));
chris@160:       *bOut = vextq_f32(o->s, x, 3);
chris@160:       o->s = x;
chris@160:       break;
chris@160:     }
chris@160:     case 0x3: {
chris@160:       const float32x4_t x = vdupq_n_f32(vgetq_lane_f32(bIn0,1));
chris@160:       *bOut = vreinterpretq_f32_u32(vorrq_u32(
chris@160:           vandq_u32(vreinterpretq_u32_f32(bIn0), (uint32x4_t) {~0x0, 0x0, 0x0, 0x0}),
chris@160:           vandq_u32(vreinterpretq_u32_f32(x), (uint32x4_t) {0x0, ~0x0, ~0x0, ~0x0})));
chris@160:       o->s = x;
chris@160:       break;
chris@160:     }
chris@160:     case 0x4: {
chris@160:       const float32x4_t x = vdupq_n_f32(vgetq_lane_f32(bIn0,2));
chris@160:       *bOut = vextq_f32(o->s, x, 2);
chris@160:       o->s = x;
chris@160:       break;
chris@160:     }
chris@160:     case 0x5: {
chris@160:       const float32x4_t x = vdupq_n_f32(vgetq_lane_f32(bIn0,0));
chris@160:       const float32x4_t y = vdupq_n_f32(vgetq_lane_f32(bIn0,2));
chris@160:       *bOut = vreinterpretq_f32_u32(vorrq_u32(
chris@160:           vandq_u32(vreinterpretq_u32_f32(x), (uint32x4_t) {~0x0, ~0x0, 0x0, 0x0}),
chris@160:           vandq_u32(vreinterpretq_u32_f32(y), (uint32x4_t) {0x0, 0x0, ~0x0, ~0x0})));
chris@160:       o->s = y;
chris@160:     }
chris@160:     case 0x6: {
chris@160:       const float32x4_t y = vdupq_n_f32(vgetq_lane_f32(bIn0,3));
chris@160:       float32x4_t z = vreinterpretq_f32_u32(vorrq_u32(
chris@160:           vandq_u32(vreinterpretq_u32_f32(o->s), (uint32x4_t) {~0x0, 0x0, 0x0, 0x0}),
chris@160:           vandq_u32(vreinterpretq_u32_f32(bIn0), (uint32x4_t) {0x0, ~0x0, ~0x0, 0x0})));
chris@160:       *bOut = vreinterpretq_f32_u32(vorrq_u32(
chris@160:           vandq_u32(vreinterpretq_u32_f32(z), (uint32x4_t) {~0x0, ~0x0, ~0x0, 0x0}),
chris@160:           vandq_u32(vreinterpretq_u32_f32(y), (uint32x4_t) {0x0, 0x0, 0x0, ~0x0})));
chris@160:       o->s = y;
chris@160:     }
chris@160:     case 0x7: {
chris@160:       const float32x4_t x = vdupq_n_f32(vgetq_lane_f32(bIn0,2));
chris@160:       *bOut = vreinterpretq_f32_u32(vorrq_u32(
chris@160:           vandq_u32(vreinterpretq_u32_f32(bIn0), (uint32x4_t) {~0x0, ~0x0, 0x0, 0x0}),
chris@160:           vandq_u32(vreinterpretq_u32_f32(x), (uint32x4_t) {0x0, 0x0, ~0x0, ~0x0})));
chris@160:       o->s = x;
chris@160:       break;
chris@160:     }
chris@160:     case 0x8: {
chris@160:       const float32x4_t x = vdupq_n_f32(vgetq_lane_f32(bIn0,3));
chris@160:       *bOut = vextq_f32(o->s, x, 1);
chris@160:       o->s = x;
chris@160:       break;
chris@160:     }
chris@160:     case 0x9: {
chris@160:       const float32x4_t x = vdupq_n_f32(vgetq_lane_f32(bIn0,0));
chris@160:       *bOut = vreinterpretq_f32_u32(vorrq_u32(
chris@160:           vandq_u32(vreinterpretq_u32_f32(x), (uint32x4_t) {~0x0, ~0x0, ~0x0, 0x0}),
chris@160:           vandq_u32(vreinterpretq_u32_f32(bIn0), (uint32x4_t) {0x0, 0x0, 0x0, ~0x0})));
chris@160:       o->s = vdupq_n_f32(vgetq_lane_f32(bIn0,3));
chris@160:     }
chris@160:     case 0xA: {
chris@160:       const float32x4_t x = vdupq_n_f32(vgetq_lane_f32(bIn0,1));
chris@160:       const float32x4_t y = vdupq_n_f32(vgetq_lane_f32(bIn0,3));
chris@160:       float32x4_t z = vreinterpretq_f32_u32(vorrq_u32(
chris@160:           vandq_u32(vreinterpretq_u32_f32(o->s), (uint32x4_t) {~0x0, 0x0, 0x0, 0x0}),
chris@160:           vandq_u32(vreinterpretq_u32_f32(x), (uint32x4_t) {0x0, ~0x0, ~0x0, 0x0})));
chris@160:       *bOut = vreinterpretq_f32_u32(vorrq_u32(
chris@160:           vandq_u32(vreinterpretq_u32_f32(z), (uint32x4_t) {~0x0, ~0x0, ~0x0, 0x0}),
chris@160:           vandq_u32(vreinterpretq_u32_f32(y), (uint32x4_t) {0x0, 0x0, 0x0, ~0x0})));
chris@160:       o->s = y;
chris@160:     }
chris@160:     case 0xB: {
chris@160:       const float32x4_t x = vdupq_n_f32(vgetq_lane_f32(bIn0,1));
chris@160:       *bOut = vreinterpretq_f32_u32(vorrq_u32(
chris@160:           vandq_u32(vreinterpretq_u32_f32(bIn0), (uint32x4_t) {~0x0, ~0x0, 0x0, ~0x0}),
chris@160:           vandq_u32(vreinterpretq_u32_f32(x), (uint32x4_t) {0x0, 0x0, ~0x0, 0x0})));
chris@160:       o->s = vdupq_n_f32(vgetq_lane_f32(bIn0,3));
chris@160:       break;
chris@160:     }
chris@160:     case 0xC: {
chris@160:       *bOut = vreinterpretq_f32_u32(vorrq_u32(
chris@160:           vandq_u32(vreinterpretq_u32_f32(o->s), (uint32x4_t) {~0x0, ~0x0, 0x0, 0x0}),
chris@160:           vandq_u32(vreinterpretq_u32_f32(bIn0), (uint32x4_t) {0x0, 0x0, ~0x0, ~0x0})));
chris@160:       o->s = vdupq_n_f32(vgetq_lane_f32(bIn0,3));
chris@160:       break;
chris@160:     }
chris@160:     case 0xD: {
chris@160:       const float32x4_t x = vdupq_n_f32(vgetq_lane_f32(bIn0,0));
chris@160:       *bOut = vreinterpretq_f32_u32(vorrq_u32(
chris@160:           vandq_u32(vreinterpretq_u32_f32(bIn0), (uint32x4_t) {~0x0, 0x0, ~0x0, ~0x0}),
chris@160:           vandq_u32(vreinterpretq_u32_f32(x), (uint32x4_t) {0x0, ~0x0, 0x0, 0x0})));
chris@160:       o->s = vdupq_n_f32(vgetq_lane_f32(bIn0,3));
chris@160:     }
chris@160:     case 0xE: {
chris@160:       *bOut = vreinterpretq_f32_u32(vorrq_u32(
chris@160:           vandq_u32(vreinterpretq_u32_f32(o->s), (uint32x4_t) {~0x0, 0x0, 0x0, 0x0}),
chris@160:           vandq_u32(vreinterpretq_u32_f32(bIn0), (uint32x4_t) {0x0, ~0x0, ~0x0, ~0x0})));
chris@160:       o->s = vdupq_n_f32(vgetq_lane_f32(bIn0,3));
chris@160:       break;
chris@160:     }
chris@160:     case 0xF: {
chris@160:       *bOut = bIn0;
chris@160:       o->s = vdupq_n_f32(vgetq_lane_f32(bIn0,3));
chris@160:       break;
chris@160:     }
chris@160:   }
chris@160: #else // HV_SIMD_NONE
chris@160:   if (bIn1 != 0.0f) o->s = bIn0;
chris@160:   *bOut = o->s;
chris@160: #endif
chris@160: }
chris@160: 
chris@160: void sSamphold_onMessage(HvBase *_c, SignalSamphold *o, int letIndex,
chris@160:     const HvMessage *const m, void *sendMessage);
chris@160: 
chris@160: #endif // _SIGNAL_SAMPHOLD_H_