/* * Copyright (c) 2014 The WebRTC project authors. All Rights Reserved. * * Use of this source code is governed by a BSD-style license * that can be found in the LICENSE file in the root of the source * tree. An additional intellectual property rights grant can be found * in the file PATENTS. All contributing project authors may * be found in the AUTHORS file in the root of the source tree. */ #include #include #include "modules/audio_coding/codecs/isac/fix/source/settings.h" // Filter ar_g_Q0[] and ar_f_Q0[] through an AR filter with coefficients // cth_Q15[] and sth_Q15[]. void WebRtcIsacfix_FilterArLoop(int16_t* ar_g_Q0, // Input samples int16_t* ar_f_Q0, // Input samples int16_t* cth_Q15, // Filter coefficients int16_t* sth_Q15, // Filter coefficients size_t order_coef) { // order of the filter int n = 0; for (n = 0; n < HALF_SUBFRAMELEN - 1; n++) { int count = (int)(order_coef - 1); int offset; #if !defined(MIPS_DSP_R1_LE) int16_t* tmp_cth; int16_t* tmp_sth; int16_t* tmp_arg; int32_t max_q16 = 0x7fff; int32_t min_q16 = 0xffff8000; #endif // Declare variables used as temporary registers. int32_t r0, r1, r2, t0, t1, t2, t_ar; __asm __volatile ( ".set push \n\t" ".set noreorder \n\t" "bltz %[count], 2f \n\t" " lh %[t_ar], 0(%[tmp]) \n\t" // Inner loop "1: \n\t" "sll %[offset], %[count], 1 \n\t" #if defined(MIPS_DSP_R1_LE) "lhx %[r0], %[offset](%[cth_Q15]) \n\t" "lhx %[r1], %[offset](%[sth_Q15]) \n\t" "lhx %[r2], %[offset](%[ar_g_Q0]) \n\t" #else "addu %[tmp_cth], %[cth_Q15], %[offset] \n\t" "addu %[tmp_sth], %[sth_Q15], %[offset] \n\t" "addu %[tmp_arg], %[ar_g_Q0], %[offset] \n\t" "lh %[r0], 0(%[tmp_cth]) \n\t" "lh %[r1], 0(%[tmp_sth]) \n\t" "lh %[r2], 0(%[tmp_arg]) \n\t" #endif "mul %[t0], %[r0], %[t_ar] \n\t" "mul %[t1], %[r1], %[t_ar] \n\t" "mul %[t2], %[r1], %[r2] \n\t" "mul %[r0], %[r0], %[r2] \n\t" "subu %[t0], %[t0], %[t2] \n\t" "addu %[t1], %[t1], %[r0] \n\t" #if defined(MIPS_DSP_R1_LE) "shra_r.w %[t1], %[t1], 15 \n\t" "shra_r.w %[t0], %[t0], 15 \n\t" #else "addiu %[t1], %[t1], 0x4000 \n\t" "sra %[t1], %[t1], 15 \n\t" "addiu %[t0], %[t0], 0x4000 \n\t" "sra %[t0], %[t0], 15 \n\t" #endif "addiu %[offset], %[offset], 2 \n\t" #if defined(MIPS_DSP_R1_LE) "shll_s.w %[t1], %[t1], 16 \n\t" "shll_s.w %[t_ar], %[t0], 16 \n\t" #else "slt %[r0], %[t1], %[max_q16] \n\t" "slt %[r1], %[t0], %[max_q16] \n\t" "movz %[t1], %[max_q16], %[r0] \n\t" "movz %[t0], %[max_q16], %[r1] \n\t" #endif "addu %[offset], %[offset], %[ar_g_Q0] \n\t" #if defined(MIPS_DSP_R1_LE) "sra %[t1], %[t1], 16 \n\t" "sra %[t_ar], %[t_ar], 16 \n\t" #else "slt %[r0], %[t1], %[min_q16] \n\t" "slt %[r1], %[t0], %[min_q16] \n\t" "movn %[t1], %[min_q16], %[r0] \n\t" "movn %[t0], %[min_q16], %[r1] \n\t" "addu %[t_ar], $zero, %[t0] \n\t" #endif "sh %[t1], 0(%[offset]) \n\t" "bgtz %[count], 1b \n\t" " addiu %[count], %[count], -1 \n\t" "2: \n\t" "sh %[t_ar], 0(%[tmp]) \n\t" "sh %[t_ar], 0(%[ar_g_Q0]) \n\t" ".set pop \n\t" : [t_ar] "=&r" (t_ar), [count] "+r" (count), [offset] "=&r" (offset), [r0] "=&r" (r0), [r1] "=&r" (r1), [r2] "=&r" (r2), [t0] "=&r" (t0), #if !defined(MIPS_DSP_R1_LE) [tmp_cth] "=&r" (tmp_cth), [tmp_sth] "=&r" (tmp_sth), [tmp_arg] "=&r" (tmp_arg), #endif [t1] "=&r" (t1), [t2] "=&r" (t2) : [tmp] "r" (&ar_f_Q0[n+1]), [cth_Q15] "r" (cth_Q15), #if !defined(MIPS_DSP_R1_LE) [max_q16] "r" (max_q16), [min_q16] "r" (min_q16), #endif [sth_Q15] "r" (sth_Q15), [ar_g_Q0] "r" (ar_g_Q0) : "memory", "hi", "lo" ); } } // MIPS optimization of the inner loop used for function // WebRtcIsacfix_NormLatticeFilterMa(). It does: // // for 0 <= n < HALF_SUBFRAMELEN - 1: // *ptr2 = input2 * (*ptr2) + input0 * (*ptr0)); // *ptr1 = input1 * (*ptr0) + input0 * (*ptr2); // // Note, function WebRtcIsacfix_FilterMaLoopMIPS and WebRtcIsacfix_FilterMaLoopC // are not bit-exact. The accuracy of the MIPS function is same or better. void WebRtcIsacfix_FilterMaLoopMIPS(int16_t input0, // Filter coefficient int16_t input1, // Filter coefficient int32_t input2, // Inverse coeff (1/input1) int32_t* ptr0, // Sample buffer int32_t* ptr1, // Sample buffer int32_t* ptr2) { // Sample buffer #if defined(MIPS_DSP_R2_LE) // MIPS DSPR2 version. 4 available accumulators allows loop unrolling 4 times. // This variant is not bit-exact with WebRtcIsacfix_FilterMaLoopC, since we // are exploiting 64-bit accumulators. The accuracy of the MIPS DSPR2 function // is same or better. int n = (HALF_SUBFRAMELEN - 1) >> 2; int m = (HALF_SUBFRAMELEN - 1) & 3; int r0, r1, r2, r3; int t0, t1, t2, t3; int s0, s1, s2, s3; __asm __volatile ( ".set push \n\t" ".set noreorder \n\t" "1: \n\t" "lw %[r0], 0(%[ptr0]) \n\t" "lw %[r1], 4(%[ptr0]) \n\t" "lw %[r2], 8(%[ptr0]) \n\t" "lw %[r3], 12(%[ptr0]) \n\t" "mult $ac0, %[r0], %[input0] \n\t" "mult $ac1, %[r1], %[input0] \n\t" "mult $ac2, %[r2], %[input0] \n\t" "mult $ac3, %[r3], %[input0] \n\t" "lw %[t0], 0(%[ptr2]) \n\t" "extr_rs.w %[s0], $ac0, 15 \n\t" "extr_rs.w %[s1], $ac1, 15 \n\t" "extr_rs.w %[s2], $ac2, 15 \n\t" "extr_rs.w %[s3], $ac3, 15 \n\t" "lw %[t1], 4(%[ptr2]) \n\t" "lw %[t2], 8(%[ptr2]) \n\t" "lw %[t3], 12(%[ptr2]) \n\t" "addu %[t0], %[t0], %[s0] \n\t" "addu %[t1], %[t1], %[s1] \n\t" "addu %[t2], %[t2], %[s2] \n\t" "addu %[t3], %[t3], %[s3] \n\t" "mult $ac0, %[t0], %[input2] \n\t" "mult $ac1, %[t1], %[input2] \n\t" "mult $ac2, %[t2], %[input2] \n\t" "mult $ac3, %[t3], %[input2] \n\t" "addiu %[ptr0], %[ptr0], 16 \n\t" "extr_rs.w %[t0], $ac0, 16 \n\t" "extr_rs.w %[t1], $ac1, 16 \n\t" "extr_rs.w %[t2], $ac2, 16 \n\t" "extr_rs.w %[t3], $ac3, 16 \n\t" "addiu %[n], %[n], -1 \n\t" "mult $ac0, %[r0], %[input1] \n\t" "mult $ac1, %[r1], %[input1] \n\t" "mult $ac2, %[r2], %[input1] \n\t" "mult $ac3, %[r3], %[input1] \n\t" "sw %[t0], 0(%[ptr2]) \n\t" "extr_rs.w %[s0], $ac0, 15 \n\t" "extr_rs.w %[s1], $ac1, 15 \n\t" "extr_rs.w %[s2], $ac2, 15 \n\t" "extr_rs.w %[s3], $ac3, 15 \n\t" "sw %[t1], 4(%[ptr2]) \n\t" "sw %[t2], 8(%[ptr2]) \n\t" "sw %[t3], 12(%[ptr2]) \n\t" "mult $ac0, %[t0], %[input0] \n\t" "mult $ac1, %[t1], %[input0] \n\t" "mult $ac2, %[t2], %[input0] \n\t" "mult $ac3, %[t3], %[input0] \n\t" "addiu %[ptr2], %[ptr2], 16 \n\t" "extr_rs.w %[t0], $ac0, 15 \n\t" "extr_rs.w %[t1], $ac1, 15 \n\t" "extr_rs.w %[t2], $ac2, 15 \n\t" "extr_rs.w %[t3], $ac3, 15 \n\t" "addu %[t0], %[t0], %[s0] \n\t" "addu %[t1], %[t1], %[s1] \n\t" "addu %[t2], %[t2], %[s2] \n\t" "addu %[t3], %[t3], %[s3] \n\t" "sw %[t0], 0(%[ptr1]) \n\t" "sw %[t1], 4(%[ptr1]) \n\t" "sw %[t2], 8(%[ptr1]) \n\t" "sw %[t3], 12(%[ptr1]) \n\t" "bgtz %[n], 1b \n\t" " addiu %[ptr1], %[ptr1], 16 \n\t" "beq %[m], %0, 3f \n\t" " nop \n\t" "2: \n\t" "lw %[r0], 0(%[ptr0]) \n\t" "lw %[t0], 0(%[ptr2]) \n\t" "addiu %[ptr0], %[ptr0], 4 \n\t" "mult $ac0, %[r0], %[input0] \n\t" "mult $ac1, %[r0], %[input1] \n\t" "extr_rs.w %[r1], $ac0, 15 \n\t" "extr_rs.w %[t1], $ac1, 15 \n\t" "addu %[t0], %[t0], %[r1] \n\t" "mult $ac0, %[t0], %[input2] \n\t" "extr_rs.w %[t0], $ac0, 16 \n\t" "sw %[t0], 0(%[ptr2]) \n\t" "mult $ac0, %[t0], %[input0] \n\t" "addiu %[ptr2], %[ptr2], 4 \n\t" "addiu %[m], %[m], -1 \n\t" "extr_rs.w %[t0], $ac0, 15 \n\t" "addu %[t0], %[t0], %[t1] \n\t" "sw %[t0], 0(%[ptr1]) \n\t" "bgtz %[m], 2b \n\t" " addiu %[ptr1], %[ptr1], 4 \n\t" "3: \n\t" ".set pop \n\t" : [r0] "=&r" (r0), [r1] "=&r" (r1), [r2] "=&r" (r2), [r3] "=&r" (r3), [t0] "=&r" (t0), [t1] "=&r" (t1), [t2] "=&r" (t2), [t3] "=&r" (t3), [s0] "=&r" (s0), [s1] "=&r" (s1), [s2] "=&r" (s2), [s3] "=&r" (s3), [ptr0] "+r" (ptr0), [ptr1] "+r" (ptr1), [m] "+r" (m), [ptr2] "+r" (ptr2), [n] "+r" (n) : [input0] "r" (input0), [input1] "r" (input1), [input2] "r" (input2) : "memory", "hi", "lo", "$ac1hi", "$ac1lo", "$ac2hi", "$ac2lo", "$ac3hi", "$ac3lo" ); #else // Non-DSPR2 version of the function. Avoiding the accumulator usage due to // large latencies. This variant is bit-exact with C code. int n = HALF_SUBFRAMELEN - 1; int32_t t16a, t16b; int32_t r0, r1, r2, r3, r4; __asm __volatile ( ".set push \n\t" ".set noreorder \n\t" "sra %[t16a], %[input2], 16 \n\t" "andi %[t16b], %[input2], 0xFFFF \n\t" #if defined(MIPS32R2_LE) "seh %[t16b], %[t16b] \n\t" "seh %[input0], %[input0] \n\t" "seh %[input1], %[input1] \n\t" #else "sll %[t16b], %[t16b], 16 \n\t" "sra %[t16b], %[t16b], 16 \n\t" "sll %[input0], %[input0], 16 \n\t" "sra %[input0], %[input0], 16 \n\t" "sll %[input1], %[input1], 16 \n\t" "sra %[input1], %[input1], 16 \n\t" #endif "addiu %[r0], %[t16a], 1 \n\t" "slt %[r1], %[t16b], $zero \n\t" "movn %[t16a], %[r0], %[r1] \n\t" "1: \n\t" "lw %[r0], 0(%[ptr0]) \n\t" "lw %[r1], 0(%[ptr2]) \n\t" "addiu %[ptr0], %[ptr0], 4 \n\t" "sra %[r2], %[r0], 16 \n\t" "andi %[r0], %[r0], 0xFFFF \n\t" "mul %[r3], %[r2], %[input0] \n\t" "mul %[r4], %[r0], %[input0] \n\t" "mul %[r2], %[r2], %[input1] \n\t" "mul %[r0], %[r0], %[input1] \n\t" "addiu %[ptr2], %[ptr2], 4 \n\t" "sll %[r3], %[r3], 1 \n\t" "sra %[r4], %[r4], 1 \n\t" "addiu %[r4], %[r4], 0x2000 \n\t" "sra %[r4], %[r4], 14 \n\t" "addu %[r3], %[r3], %[r4] \n\t" "addu %[r1], %[r1], %[r3] \n\t" "sra %[r3], %[r1], 16 \n\t" "andi %[r4], %[r1], 0xFFFF \n\t" "sra %[r4], %[r4], 1 \n\t" "mul %[r1], %[r1], %[t16a] \n\t" "mul %[r3], %[r3], %[t16b] \n\t" "mul %[r4], %[r4], %[t16b] \n\t" "sll %[r2], %[r2], 1 \n\t" "sra %[r0], %[r0], 1 \n\t" "addiu %[r0], %[r0], 0x2000 \n\t" "sra %[r0], %[r0], 14 \n\t" "addu %[r0], %[r0], %[r2] \n\t" "addiu %[n], %[n], -1 \n\t" "addu %[r1], %[r1], %[r3] \n\t" "addiu %[r4], %[r4], 0x4000 \n\t" "sra %[r4], %[r4], 15 \n\t" "addu %[r1], %[r1], %[r4] \n\t" "sra %[r2], %[r1], 16 \n\t" "andi %[r3], %[r1], 0xFFFF \n\t" "mul %[r3], %[r3], %[input0] \n\t" "mul %[r2], %[r2], %[input0] \n\t" "sw %[r1], -4(%[ptr2]) \n\t" "sra %[r3], %[r3], 1 \n\t" "addiu %[r3], %[r3], 0x2000 \n\t" "sra %[r3], %[r3], 14 \n\t" "addu %[r0], %[r0], %[r3] \n\t" "sll %[r2], %[r2], 1 \n\t" "addu %[r0], %[r0], %[r2] \n\t" "sw %[r0], 0(%[ptr1]) \n\t" "bgtz %[n], 1b \n\t" " addiu %[ptr1], %[ptr1], 4 \n\t" ".set pop \n\t" : [t16a] "=&r" (t16a), [t16b] "=&r" (t16b), [r0] "=&r" (r0), [r1] "=&r" (r1), [r2] "=&r" (r2), [r3] "=&r" (r3), [r4] "=&r" (r4), [ptr0] "+r" (ptr0), [ptr1] "+r" (ptr1), [ptr2] "+r" (ptr2), [n] "+r" (n) : [input0] "r" (input0), [input1] "r" (input1), [input2] "r" (input2) : "hi", "lo", "memory" ); #endif }