diff options
Diffstat (limited to 'sys/i386/isa/sound/mmap_test.c')
-rw-r--r-- | sys/i386/isa/sound/mmap_test.c | 265 |
1 files changed, 265 insertions, 0 deletions
diff --git a/sys/i386/isa/sound/mmap_test.c b/sys/i386/isa/sound/mmap_test.c new file mode 100644 index 0000000..c6afae9 --- /dev/null +++ b/sys/i386/isa/sound/mmap_test.c @@ -0,0 +1,265 @@ +/* + * This is a simple program which demonstrates use of mmapped DMA buffer + * of the sound driver directly from application program. + * + * This sample program works (currently) only with Linux, FreeBSD and BSD/OS + * (FreeBSD and BSD/OS require OSS version 3.8-beta16 or later. + * + * Note! Don't use mmapped DMA buffers (direct audio) unless you have + * very good reasons to do it. Programs using this feature will not + * work with all soundcards. GUS (GF1) is one of them (GUS MAX works). + * + * This program requires version 3.5-beta7 or later of OSS + * (3.8-beta16 or later in FreeBSD and BSD/OS). + */ + +#ifndef lint +static const char rcsid[] = + "$Id: mmap_test.c,v 1.2 1998/01/16 07:20:34 charnier Exp $"; +#endif /* not lint */ + +#include <err.h> +#include <fcntl.h> +#include <stdio.h> +#include <string.h> +#include <unistd.h> +#include <sys/types.h> +#include <sys/mman.h> +#include <machine/soundcard.h> +#include <sys/time.h> + +int +main() +{ + int fd, sz, fsz, tmp, nfrag; + int caps; + + int sd, sl=0, sp; + + unsigned char data[500000], *dp = data; + + caddr_t buf; + struct timeval tim; + + unsigned char *op; + + struct audio_buf_info info; + + int frag = 0xffff000c; /* Max # fragments of 2^13=8k bytes */ + + fd_set writeset; + + close(0); + if ((fd=open("/dev/dsp", O_RDWR, 0))==-1) + err(1, "/dev/dsp"); +/* + * Then setup sampling parameters. Just sampling rate in this case. + */ + + tmp = 8000; + ioctl(fd, SNDCTL_DSP_SPEED, &tmp); + +/* + * Load some test data. + */ + + sl = sp = 0; + if ((sd=open("smpl", O_RDONLY, 0))!=-1) + { + sl = read(sd, data, sizeof(data)); + printf("%d bytes read from file.\n", sl); + close(sd); + } + else warn("smpl"); + + if (ioctl(fd, SNDCTL_DSP_GETCAPS, &caps)==-1) + { + warn("sorry but your sound driver is too old"); + err(1, "/dev/dsp"); + } + +/* + * Check that the device has capability to do this. Currently just + * CS4231 based cards will work. + * + * The application should also check for DSP_CAP_MMAP bit but this + * version of driver doesn't have it yet. + */ +/* ioctl(fd, SNDCTL_DSP_SETSYNCRO, 0); */ + +/* + * You need version 3.5-beta7 or later of the sound driver before next + * two lines compile. There is no point to modify this program to + * compile with older driver versions since they don't have working + * mmap() support. + */ + if (!(caps & DSP_CAP_TRIGGER) || + !(caps & DSP_CAP_MMAP)) + errx(1, "sorry but your soundcard can't do this"); + +/* + * Select the fragment size. This is propably important only when + * the program uses select(). Fragment size defines how often + * select call returns. + */ + + ioctl(fd, SNDCTL_DSP_SETFRAGMENT, &frag); + +/* + * Compute total size of the buffer. It's important to use this value + * in mmap() call. + */ + + if (ioctl(fd, SNDCTL_DSP_GETOSPACE, &info)==-1) + err(1, "GETOSPACE"); + + sz = info.fragstotal * info.fragsize; + fsz = info.fragsize; +/* + * Call mmap(). + * + * IMPORTANT NOTE!!!!!!!!!!! + * + * Full duplex audio devices have separate input and output buffers. + * It is not possible to map both of them at the same mmap() call. The buffer + * is selected based on the prot argument in the following way: + * + * - PROT_READ (alone) selects the input buffer. + * - PROT_WRITE (alone) selects the output buffer. + * - PROT_WRITE|PROT_READ together select the output buffer. This combination + * is required in BSD to make the buffer accessible. With just PROT_WRITE + * every attempt to access the returned buffer will result in segmentation/bus + * error. PROT_READ|PROT_WRITE is also permitted in Linux with OSS version + * 3.8-beta16 and later (earlier versions don't accept it). + * + * Non duplex devices have just one buffer. When an application wants to do both + * input and output it's recommended that the device is closed and re-opened when + * switching between modes. PROT_READ|PROT_WRITE can be used to open the buffer + * for both input and output (with OSS 3.8-beta16 and later) but the result may be + * unpredictable. + */ + + if ((buf=mmap(NULL, sz, PROT_WRITE | PROT_READ, MAP_FILE|MAP_SHARED, fd, 0))==(caddr_t)-1) + err(1, "mmap (write)"); + printf("mmap (out) returned %08x\n", buf); + op=buf; + +/* + * op contains now a pointer to the DMA buffer + */ + +/* + * Then it's time to start the engine. The driver doesn't allow read() and/or + * write() when the buffer is mapped. So the only way to start operation is + * to togle device's enable bits. First set them off. Setting them on enables + * recording and/or playback. + */ + + tmp = 0; + ioctl(fd, SNDCTL_DSP_SETTRIGGER, &tmp); + +/* + * It might be usefull to write some data to the buffer before starting. + */ + + tmp = PCM_ENABLE_OUTPUT; + ioctl(fd, SNDCTL_DSP_SETTRIGGER, &tmp); + +/* + * The machine is up and running now. Use SNDCTL_DSP_GETOPTR to get the + * buffer status. + * + * NOTE! The driver empties each buffer fragmen after they have been + * played. This prevents looping sound if there are some performance problems + * in the application side. For similar reasons it recommended that the + * application uses some amout of play ahead. It can rewrite the unplayed + * data later if necessary. + */ + + nfrag = 0; + while (1) + { + struct count_info count; + int extra; + + FD_ZERO(&writeset); + FD_SET(fd, &writeset); + + tim.tv_sec = 10; + tim.tv_usec= 0; + + select(fd+1, &writeset, &writeset, NULL, NULL); +/* + * SNDCTL_DSP_GETOPTR (and GETIPTR as well) return three items. The + * bytes field returns number of bytes played since start. It can be used + * as a real time clock. + * + * The blocks field returns number of fragment transitions (interrupts) since + * previous GETOPTR call. It can be used as a method to detect underrun + * situations. + * + * The ptr field is the DMA pointer inside the buffer area (in bytes from + * the beginning of total buffer area). + */ + + if (ioctl(fd, SNDCTL_DSP_GETOPTR, &count)==-1) + err(1, "GETOPTR"); + if (count.ptr < 0 ) count.ptr = 0; + nfrag += count.blocks; + + +#ifdef VERBOSE + + printf("\rTotal: %09d, Fragment: %03d, Ptr: %06d", + count.bytes, nfrag, count.ptr); + fflush(stdout); +#endif + +/* + * Caution! This version doesn't check for bounds of the DMA + * memory area. It's possible that the returned pointer value is not aligned + * to fragment boundaries. It may be several samples behind the boundary + * in case there was extra delay between the actual hardware interrupt and + * the time when DSP_GETOPTR was called. + * + * Don't just call memcpy() with length set to 'fragment_size' without + * first checking that the transfer really fits to the buffer area. + * A mistake of just one byte causes seg fault. It may be easiest just + * to align the returned pointer value to fragment boundary before using it. + * + * It would be very good idea to write few extra samples to next fragment + * too. Otherwise several (uninitialized) samples from next fragment + * will get played before your program gets chance to initialize them. + * Take in count the fact thaat there are other processes batling about + * the same CPU. This effect is likely to be very annoying if fragment + * size is decreased too much. + */ + +/* + * Just a minor clarification to the above. The following line alings + * the pointer to fragment boundaries. Note! Don't trust that fragment + * size is always a power of 2. It may not be so in future. + */ + count.ptr = ((count.ptr+16)/fsz )*fsz; +#ifdef VERBOSE + printf(" memcpy(%6d, %4d)", (dp-data), fsz); + fflush(stdout); +#endif + +/* + * Set few bytes in the beginning of next fragment too. + */ + + if ((count.ptr+fsz+16) < sz) /* Last fragment? */ + extra = 16; + else + extra = 0; + memcpy(op+count.ptr, dp, (fsz+extra)); + dp += fsz; + if (dp > (data+sl-fsz)) + dp = data; + + } + + exit(0); +} |