chris@162: /**
chris@162:  * Copyright (c) 2014, 2015, Enzien Audio Ltd.
chris@162:  *
chris@162:  * Permission to use, copy, modify, and/or distribute this software for any
chris@162:  * purpose with or without fee is hereby granted, provided that the above
chris@162:  * copyright notice and this permission notice appear in all copies.
chris@162:  *
chris@162:  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH
chris@162:  * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
chris@162:  * AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT,
chris@162:  * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
chris@162:  * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR
chris@162:  * OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
chris@162:  * PERFORMANCE OF THIS SOFTWARE.
chris@162:  */
chris@162: 
chris@162: #ifndef _HEAVY_SIGNAL_TABREAD_H_
chris@162: #define _HEAVY_SIGNAL_TABREAD_H_
chris@162: 
chris@162: #include "HvBase.h"
chris@162: #include "HvTable.h"
chris@162: 
chris@162: typedef struct SignalTabread {
chris@162:   HvTable *table; // the table to read
chris@162:   hv_uint32_t head;
chris@162:   bool forceAlignedLoads; // false by default, true if using __hv_tabread_f
chris@162: } SignalTabread;
chris@162: 
chris@162: // random access to a table
chris@162: hv_size_t sTabread_init(SignalTabread *o, HvTable *table, bool forceAlignedLoads);
chris@162: 
chris@162: 
chris@162: 
chris@162: #if HV_APPLE
chris@162: #pragma mark - Tabread - Random Access
chris@162: #endif
chris@162: 
chris@162: static inline void __hv_tabread_if(SignalTabread *o, hv_bIni_t bIn, hv_bOutf_t bOut) {
chris@162:   const float *const b = hTable_getBuffer(o->table);
chris@162: #if HV_SIMD_AVX
chris@162:   hv_assert((int) (bIn[0] & 0xFFFFFFFFL) >= 0 && (int) (bIn[0] & 0xFFFFFFFFL) < hTable_getAllocated(o->table));
chris@162:   hv_assert((int) (bIn[0] >> 32) >= 0 && (int) ((bIn[0] & ~0xFFFFFFFFL) >> 32) < hTable_getAllocated(o->table));
chris@162:   hv_assert((int) (bIn[1] & 0xFFFFFFFFL) >= 0 && (int) (bIn[1] & 0xFFFFFFFFL) < hTable_getAllocated(o->table));
chris@162:   hv_assert((int) (bIn[1] >> 32) >= 0 && (int) ((bIn[1] & ~0xFFFFFFFFL) >> 32) < hTable_getAllocated(o->table));
chris@162:   hv_assert((int) (bIn[2] & 0xFFFFFFFFL) >= 0 && (int) (bIn[2] & 0xFFFFFFFFL) < hTable_getAllocated(o->table));
chris@162:   hv_assert((int) (bIn[2] >> 32) >= 0 && (int) ((bIn[2] & ~0xFFFFFFFFL) >> 32) < hTable_getAllocated(o->table));
chris@162:   hv_assert((int) (bIn[3] & 0xFFFFFFFFL) >= 0 && (int) (bIn[3] & 0xFFFFFFFFL) < hTable_getAllocated(o->table));
chris@162:   hv_assert((int) (bIn[3] >> 32) >= 0 && (int) ((bIn[3] & ~0xFFFFFFFFL) >> 32) < hTable_getAllocated(o->table));
chris@162: 
chris@162:   *bOut = _mm256_set_ps(
chris@162:       b[(int) (bIn[3] >> 32)],
chris@162:       b[(int) (bIn[3] & 0xFFFFFFFFL)],
chris@162:       b[(int) (bIn[2] >> 32)],
chris@162:       b[(int) (bIn[2] & 0xFFFFFFFFL)],
chris@162:       b[(int) (bIn[1] >> 32)],
chris@162:       b[(int) (bIn[1] & 0xFFFFFFFFL)],
chris@162:       b[(int) (bIn[0] >> 32)],
chris@162:       b[(int) (bIn[0] & 0xFFFFFFFFL)]);
chris@162: #elif HV_SIMD_SSE
chris@162:   hv_assert((int) (bIn[0] & 0xFFFFFFFFL) >= 0 && (int) (bIn[0] & 0xFFFFFFFFL) < hTable_getAllocated(o->table));
chris@162:   hv_assert((int) (bIn[0] >> 32) >= 0 && (int) (bIn[0] >> 32) < hTable_getAllocated(o->table));
chris@162:   hv_assert((int) (bIn[1] & 0xFFFFFFFFL) >= 0 && (int) (bIn[1] & 0xFFFFFFFFL) < hTable_getAllocated(o->table));
chris@162:   hv_assert((int) (bIn[1] >> 32) >= 0 && (int) (bIn[1] >> 32) < hTable_getAllocated(o->table));
chris@162: 
chris@162:   *bOut = _mm_set_ps(
chris@162:       b[(int) (bIn[1] >> 32)],
chris@162:       b[(int) (bIn[1] & 0xFFFFFFFFL)],
chris@162:       b[(int) (bIn[0] >> 32)],
chris@162:       b[(int) (bIn[0] & 0xFFFFFFFFL)]);
chris@162: #elif HV_SIMD_NEON
chris@162:   hv_assert((bIn[0] >= 0) && (bIn[0] < hTable_getAllocated(o->table)));
chris@162:   hv_assert((bIn[1] >= 0) && (bIn[1] < hTable_getAllocated(o->table)));
chris@162:   hv_assert((bIn[2] >= 0) && (bIn[2] < hTable_getAllocated(o->table)));
chris@162:   hv_assert((bIn[3] >= 0) && (bIn[3] < hTable_getAllocated(o->table)));
chris@162: 
chris@162:   *bOut = (float32x4_t) {b[bIn[0]], b[bIn[1]], b[bIn[2]], b[bIn[3]]};
chris@162: #else // HV_SIMD_NONE
chris@162:   hv_assert(bIn >= 0 && ((hv_uint32_t) bIn < hTable_getAllocated(o->table)));
chris@162: 
chris@162:   *bOut = b[bIn];
chris@162: #endif
chris@162: }
chris@162: 
chris@162: 
chris@162: 
chris@162: #if HV_APPLE
chris@162: #pragma mark - Tabread - Linear Access
chris@162: #endif
chris@162: 
chris@162: // this tabread never stops reading. It is mainly intended for linear reads that loop around a table.
chris@162: static inline void __hv_tabread_f(SignalTabread *o, hv_bOutf_t bOut) {
chris@162:   hv_assert((o->head + HV_N_SIMD) <= hTable_getAllocated(o->table)); // assert that we always read within the table bounds
chris@162:   hv_uint32_t head = o->head;
chris@162: #if HV_SIMD_AVX
chris@162:   *bOut = _mm256_load_ps(hTable_getBuffer(o->table) + head);
chris@162: #elif HV_SIMD_SSE
chris@162:   *bOut = _mm_load_ps(hTable_getBuffer(o->table) + head);
chris@162: #elif HV_SIMD_NEON
chris@162:   *bOut = vld1q_f32(hTable_getBuffer(o->table) + head);
chris@162: #else // HV_SIMD_NONE
chris@162:   *bOut = *(hTable_getBuffer(o->table) + head);
chris@162: #endif
chris@162:   o->head = head + HV_N_SIMD;
chris@162: }
chris@162: 
chris@162: // unaligned linear tabread, as above
chris@162: static inline void __hv_tabreadu_f(SignalTabread *o, hv_bOutf_t bOut) {
chris@162:   hv_assert((o->head + HV_N_SIMD) <= hTable_getAllocated(o->table)); // assert that we always read within the table bounds
chris@162:   hv_uint32_t head = o->head;
chris@162: #if HV_SIMD_AVX
chris@162:   *bOut = _mm256_loadu_ps(hTable_getBuffer(o->table) + head);
chris@162: #elif HV_SIMD_SSE
chris@162:   *bOut = _mm_loadu_ps(hTable_getBuffer(o->table) + head);
chris@162: #elif HV_SIMD_NEON
chris@162:   *bOut = vld1q_f32(hTable_getBuffer(o->table) + head);
chris@162: #else // HV_SIMD_NONE
chris@162:   *bOut = *(hTable_getBuffer(o->table) + head);
chris@162: #endif
chris@162:   o->head = head + HV_N_SIMD;
chris@162: }
chris@162: 
chris@162: // this tabread can be instructed to stop. It is mainly intended for linear reads that only process a portion of a buffer.
chris@162: static inline void __hv_tabread_stoppable_f(SignalTabread *o, hv_bOutf_t bOut) {
chris@162: #if HV_SIMD_AVX
chris@162:   if (o->head == ~0x0) {
chris@162:     *bOut = _mm256_setzero_ps();
chris@162:   } else {
chris@162:     *bOut = _mm256_load_ps(hTable_getBuffer(o->table) + o->head);
chris@162:     o->head += HV_N_SIMD;
chris@162:   }
chris@162: #elif HV_SIMD_SSE
chris@162:   if (o->head == ~0x0) {
chris@162:     *bOut = _mm_setzero_ps();
chris@162:   } else {
chris@162:     *bOut = _mm_load_ps(hTable_getBuffer(o->table) + o->head);
chris@162:     o->head += HV_N_SIMD;
chris@162:   }
chris@162: #elif HV_SIMD_NEON
chris@162:   if (o->head == ~0x0) {
chris@162:     *bOut = vdupq_n_f32(0.0f);
chris@162:   } else {
chris@162:     *bOut = vld1q_f32(hTable_getBuffer(o->table) + o->head);
chris@162:     o->head += HV_N_SIMD;
chris@162:   }
chris@162: #else // HV_SIMD_NONE
chris@162:   if (o->head == ~0x0) {
chris@162:     *bOut = 0.0f;
chris@162:   } else {
chris@162:     *bOut = *(hTable_getBuffer(o->table) + o->head);
chris@162:     o->head += HV_N_SIMD;
chris@162:   }
chris@162: #endif
chris@162: }
chris@162: 
chris@162: void sTabread_onMessage(HvBase *_c, SignalTabread *o, int letIn, const HvMessage *const m);
chris@162: 
chris@162: 
chris@162: 
chris@162: #if HV_APPLE
chris@162: #pragma mark - Tabhead
chris@162: #endif
chris@162: 
chris@162: typedef struct SignalTabhead {
chris@162:   HvTable *table;
chris@162: } SignalTabhead;
chris@162: 
chris@162: hv_size_t sTabhead_init(SignalTabhead *o, HvTable *table);
chris@162: 
chris@162: static inline void __hv_tabhead_f(SignalTabhead *o, hv_bOutf_t bOut) {
chris@162: #if HV_SIMD_AVX
chris@162:   *bOut = _mm256_set1_ps((float) hTable_getHead(o->table));
chris@162: #elif HV_SIMD_SSE
chris@162:   *bOut = _mm_set1_ps((float) hTable_getHead(o->table));
chris@162: #elif HV_SIMD_NEON
chris@162:   *bOut = vdupq_n_f32((float32_t) hTable_getHead(o->table));
chris@162: #else // HV_SIMD_NONE
chris@162:   *bOut = (float) hTable_getHead(o->table);
chris@162: #endif
chris@162: }
chris@162: 
chris@162: void sTabhead_onMessage(HvBase *_c, SignalTabhead *o, const HvMessage *const m);
chris@162: 
chris@162: #endif // _HEAVY_SIGNAL_TABREAD_H_