Merge pull request #79 from Themaister/neon

Merging of neon branch
This commit is contained in:
Squarepusher 2012-12-02 20:15:48 -08:00
commit d03b6afeb8
8 changed files with 91 additions and 8 deletions

View File

@ -1,7 +1,9 @@
RARCH_VERSION = "0.9.8-beta3"
LOCAL_PATH := $(call my-dir)
PERF_TEST := 0
PERF_TEST := 1
HAVE_OPENSL := 1
WANT_NEON := 1
HAVE_SINC := 1
include $(CLEAR_VARS)
@ -15,10 +17,22 @@ LOCAL_CFLAGS += -DANDROID_X86 -DHAVE_SSSE3
endif
ifeq ($(TARGET_ARCH_ABI),armeabi-v7a)
#LOCAL_CFLAGS += -mfpu=neon
ifeq ($(WANT_NEON),1)
LOCAL_CFLAGS += -DWANT_NEON
endif
ifeq ($(HAVE_SINC),1)
ifeq ($(WANT_NEON),1)
LOCAL_SRC_FILES += ../../../audio/sinc_neon.S.neon
endif
LOCAL_CFLAGS += -DHAVE_SINC
endif
LOCAL_CFLAGS += -DANDROID_ARM_V7
endif
ifeq ($(TARGET_ARCH),mips)
LOCAL_CFLAGS += -DANDROID_MIPS -D__mips__ -D__MIPSEL__
endif
@ -27,7 +41,7 @@ LOCAL_MODULE := retroarch-activity
RARCH_PATH := ../../..
LIBXML_PATH := ../../../deps/libxml2
LOCAL_SRC_FILES = $(RARCH_PATH)/console/griffin/griffin.c
LOCAL_SRC_FILES += $(RARCH_PATH)/console/griffin/griffin.c
ifeq ($(PERF_TEST), 1)

View File

@ -52,6 +52,7 @@ static inline float hermite_kernel(float mu1, float a, float b, float c, float d
rarch_resampler_t *resampler_new(void)
{
RARCH_LOG("Hermite resampler [C]\n");
return (rarch_resampler_t*)calloc(1, sizeof(rarch_resampler_t));
}

View File

@ -109,8 +109,6 @@ static void *sl_init(const char *device, unsigned rate, unsigned latency)
{
(void)device;
g_settings.audio.out_rate = 44100;
SLDataFormat_PCM fmt_pcm = {0};
SLDataSource audio_src = {0};
SLDataSink audio_sink = {0};

View File

@ -123,6 +123,8 @@ rarch_resampler_t *resampler_new(void)
RARCH_LOG("Sinc resampler [AVX]\n");
#elif defined(__SSE__)
RARCH_LOG("Sinc resampler [SSE]\n");
#elif defined(WANT_NEON)
RARCH_LOG("Sinc resampler [NEON]\n");
#else
RARCH_LOG("Sinc resampler [C]\n");
#endif
@ -208,6 +210,18 @@ static void process_sinc(rarch_resampler_t *resamp, float *out_buffer)
// movehl { X, R, X, L } == { X, R, X, R }
_mm_store_ss(out_buffer + 1, _mm_movehl_ps(sum, sum));
}
#elif defined(WANT_NEON)
void process_sinc_neon_asm(float *out, const float *left, const float *right, const float *coeff);
static void process_sinc(rarch_resampler_t *resamp, float *out_buffer)
{
const float *buffer_l = resamp->buffer_l + resamp->ptr;
const float *buffer_r = resamp->buffer_r + resamp->ptr;
unsigned phase = resamp->time >> SUBPHASE_BITS;
const float *phase_table = resamp->phase_table[phase];
process_sinc_neon_asm(out_buffer, buffer_l, buffer_r, phase_table);
}
#else // Plain ol' C99
static void process_sinc(rarch_resampler_t *resamp, float *out_buffer)
{

52
audio/sinc_neon.S Normal file
View File

@ -0,0 +1,52 @@
/* RetroArch - A frontend for libretro.
* Copyright (C) 2010-2012 - Hans-Kristian Arntzen
*
* RetroArch is free software: you can redistribute it and/or modify it under the terms
* of the GNU General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* RetroArch is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with RetroArch.
* If not, see <http://www.gnu.org/licenses/>.
*/
.arm
.align 4
.global process_sinc_neon_asm
# void process_sinc_neon(float *out, const float *left, const float *right, const float *coeff)
# Hardcoded to 16 taps.
process_sinc_neon_asm:
# Left
vld1.f32 {q0-q1}, [r1]!
vld1.f32 {q2-q3}, [r1]!
# Right
vld1.f32 {q4-q5}, [r2]!
vld1.f32 {q6-q7}, [r2]!
# Coeff
vld1.f32 {q8-q9}, [r3, :128]!
vld1.f32 {q10-q11}, [r3, :128]!
# Left
vmul.f32 q0, q0, q8
vmul.f32 q1, q1, q9
vmla.f32 q0, q2, q10
vmla.f32 q1, q3, q11
# Right
vmul.f32 q4, q4, q8
vmul.f32 q5, q5, q9
vmla.f32 q4, q6, q10
vmla.f32 q5, q7, q11
# Add everything together
vadd.f32 q0, q0, q1
vadd.f32 q4, q4, q5
vadd.f32 d0, d0, d1
vadd.f32 d8, d8, d9
vpadd.f32 d0, d0, d8
vst1.f32 d0, [r0]
bx lr

View File

@ -247,7 +247,11 @@ FIFO BUFFER
/*============================================================
AUDIO HERMITE
============================================================ */
#ifdef HAVE_SINC
#include "../../audio/sinc.c"
#else
#include "../../audio/hermite.c"
#endif
/*============================================================
RSOUND

View File

@ -1175,7 +1175,7 @@ static bool gl_frame(void *data, const void *frame, unsigned width, unsigned hei
if (msg)
gl_render_msg(gl, msg);
#if !defined(RARCH_CONSOLE) && !defined(ANDROID)
#if !defined(RARCH_CONSOLE)
context_update_window_title_func(false);
#endif

View File

@ -95,7 +95,7 @@ static bool gen_filter_bilinear(struct scaler_ctx *ctx)
return true;
}
static inline double sinc(double phase)
static inline double filter_sinc(double phase)
{
if (fabs(phase) < 0.0001)
return 1.0;
@ -116,7 +116,7 @@ static void gen_filter_sinc_sub(struct scaler_filter *filter, int len, int pos,
{
double sinc_phase = M_PI * ((double)((sinc_size << 15) + (pos & 0xffff)) / 0x10000 - j);
double lanczos_phase = sinc_phase / ((sinc_size >> 1));
int16_t sinc_val = FILTER_UNITY * sinc(sinc_phase * phase_mul) * sinc(lanczos_phase) * phase_mul;
int16_t sinc_val = FILTER_UNITY * filter_sinc(sinc_phase * phase_mul) * filter_sinc(lanczos_phase) * phase_mul;
//sinc_sum += sinc_val;
filter->filter[i * sinc_size + j] = sinc_val;