<atlka@pg.gda.pl>

- shm memory patch
- optimized a bit sample mixing routines in c and x_86 asm

src/pcm/pcm_direct.c

@@ -101,11 +101,21 @@ int snd_pcm_direct_shm_create_or_connect(snd_pcm_direct_t *dmix)
 {
 	static int snd_pcm_direct_shm_discard(snd_pcm_direct_t *dmix);
 	struct shmid_ds buf;
-	int ret = 0;
+	int tmpid, err;
 	
+retryget:
 	dmix->shmid = shmget(dmix->ipc_key, sizeof(snd_pcm_direct_share_t), IPC_CREAT | 0666);
-	if (dmix->shmid < 0)
-		return -errno;
+	err = -errno;
+	if (dmix->shmid < 0){
+		if (errno == EINVAL)
+		if ((tmpid = shmget(dmix->ipc_key, 0, 0666)) != -1)
+		if (!shmctl(tmpid, IPC_STAT, &buf))
+		if (!buf.shm_nattch)
+	    	/* no users so destroy the segment */
+		if (!shmctl(tmpid, IPC_RMID, NULL))
+		    goto retryget;
+		return err;
+	}
 	dmix->shmptr = shmat(dmix->shmid, 0, 0);
 	if (dmix->shmptr == (void *) -1) {
 		snd_pcm_direct_shm_discard(dmix);
@@ -118,9 +128,9 @@ int snd_pcm_direct_shm_create_or_connect(snd_pcm_direct_t *dmix)
 	}
 	if (buf.shm_nattch == 1) {	/* we're the first user, clear the segment */
 		memset(dmix->shmptr, 0, sizeof(snd_pcm_direct_share_t));
-		ret = 1;
+		return 1;
 	}
-	return ret;
+	return 0;
 }
 
 int snd_pcm_direct_shm_discard(snd_pcm_direct_t *dmix)

src/pcm/pcm_dmix.c

@@ -59,14 +59,26 @@ const char *_snd_module_pcm_dmix = "";
 static int shm_sum_create_or_connect(snd_pcm_direct_t *dmix)
 {
 	static int shm_sum_discard(snd_pcm_direct_t *dmix);
+	struct shmid_ds buf;
+	int tmpid, err;
 	size_t size;
 
 	size = dmix->shmptr->s.channels *
 	       dmix->shmptr->s.buffer_size *
 	       sizeof(signed int);	
+retryshm:
 	dmix->u.dmix.shmid_sum = shmget(dmix->ipc_key + 1, size, IPC_CREAT | 0666);
-	if (dmix->u.dmix.shmid_sum < 0)
-		return -errno;
+	err = -errno;
+	if (dmix->u.dmix.shmid_sum < 0){
+		if (errno == EINVAL)
+		if ((tmpid = shmget(dmix->ipc_key + 1, 0, 0666)) != -1)
+		if (!shmctl(tmpid, IPC_STAT, &buf))
+	    	if (!buf.shm_nattch) 
+		/* no users so destroy the segment */
+		if (!shmctl(tmpid, IPC_RMID, NULL))
+		    goto retryshm;
+		return err;
+	}
 	dmix->u.dmix.sum_buffer = shmat(dmix->u.dmix.shmid_sum, 0, 0);
 	if (dmix->u.dmix.sum_buffer == (void *) -1) {
 		shm_sum_discard(dmix);
@@ -214,7 +226,7 @@ static void mix_areas1(unsigned int size,
 {
 	register signed int sample, old_sample;
 
-	while (size-- > 0) {
+	for (;;) {
 		sample = *src;
 		old_sample = *sum;
 		if (*dst == 0)
@@ -230,6 +242,8 @@ static void mix_areas1(unsigned int size,
 				sample = old_sample;
 			*dst = sample;
 		} while (*sum != old_sample);
+		if (!--size)
+			return;
 		((char *)src) += src_step;
 		((char *)dst) += dst_step;
 		((char *)sum) += sum_step;
@@ -244,7 +258,7 @@ static void mix_areas2(unsigned int size,
 {
 	register signed int sample, old_sample;
 
-	while (size-- > 0) {
+	for (;;) {
 		sample = *src / 256;
 		old_sample = *sum;
 		if (*dst == 0)
@@ -260,6 +274,8 @@ static void mix_areas2(unsigned int size,
 				sample = old_sample * 256;
 			*dst = sample;
 		} while (*sum != old_sample);
+		if (!--size)
+			return;
 		((char *)src) += src_step;
 		((char *)dst) += dst_step;
 		((char *)sum) += sum_step;
@@ -366,9 +382,11 @@ static void snd_pcm_dmix_sync_area(snd_pcm_t *pcm, snd_pcm_uframes_t size)
 	dmix->slave_appl_ptr %= dmix->shmptr->s.boundary;
 	while (size > 0) {
 		transfer = appl_ptr + size > pcm->buffer_size ? pcm->buffer_size - appl_ptr : size;
-		transfer = slave_appl_ptr + transfer > dmix->shmptr->s.buffer_size ? dmix->shmptr->s.buffer_size - slave_appl_ptr : transfer;
-		size -= transfer;
+		if ((transfer = slave_appl_ptr + transfer > dmix->shmptr->s.buffer_size ? dmix->shmptr->s.buffer_size - slave_appl_ptr : transfer))
 		mix_areas(dmix, src_areas, dst_areas, appl_ptr, slave_appl_ptr, transfer);
+		if (transfer >= size)
+			return;
+		size -= transfer;
 		slave_appl_ptr += transfer;
 		slave_appl_ptr %= dmix->shmptr->s.buffer_size;
 		appl_ptr += transfer;

src/pcm/pcm_dmix_i386.h

@@ -51,16 +51,17 @@ static void MIX_AREAS1(unsigned int size,
 		"\tmovl %1, %%edi\n"
 		"\tmovl %2, %%esi\n"
 		"\tmovl %3, %%ebx\n"
+		"\tjmp 2f\n"
+
 
 		/*
-		 * while (size-- > 0) {
+		 * for (;;)
 		 */
-		"\tcmpl $0, %0\n"
-		"jz 6f\n"
-
 		"\t.p2align 4,,15\n"
-
 		"1:"
+		"\tadd %4, %%edi\n"
+		"\tadd %5, %%esi\n"
+		"\tadd %6, %%ebx\n"
 
 		/*
 		 *   sample = *src;
@@ -69,14 +70,17 @@ static void MIX_AREAS1(unsigned int size,
 		 *     sample -= sum_sample;
 		 *   xadd(*sum, sample);
 		 */
-		"\tmovw $0, %%ax\n"
-		"\tmovw $1, %%cx\n"
+
+		"2:"
+		"\txor %%ax, %%ax\n"
+		"\tmovw %%ax, %%cx\n"
+		"\tincw  %%cx\n"
 		"\tmovl (%%ebx), %%edx\n"
 		"\t" LOCK_PREFIX "cmpxchgw %%cx, (%%edi)\n"
 		"\tmovswl (%%esi), %%ecx\n"
-		"\tjnz 2f\n"
+		"\tjnz 3f\n"
 		"\tsubl %%edx, %%ecx\n"
-		"2:"
+		"3:"
 		"\t" LOCK_PREFIX "addl %%ecx, (%%ebx)\n"
 
 		/*
@@ -87,25 +91,22 @@ static void MIX_AREAS1(unsigned int size,
 		 *   } while (v != *sum);
 		 */
 
-		"3:"
+		"4:"
 		"\tmovl (%%ebx), %%ecx\n"
 		"\tcmpl $0x7fff,%%ecx\n"
-		"\tjg 4f\n"
+		"\tjg 5f\n"
 		"\tcmpl $-0x8000,%%ecx\n"
-		"\tjl 5f\n"
+		"\tjl 6f\n"
 		"\tmovw %%cx, (%%edi)\n"
 		"\tcmpl %%ecx, (%%ebx)\n"
-		"\tjnz 3b\n"
+		"\tjnz 4b\n"
 
 		/*
 		 * while (size-- > 0)
 		 */
-		"\tadd %4, %%edi\n"
-		"\tadd %5, %%esi\n"
-		"\tadd %6, %%ebx\n"
 		"\tdecl %0\n"
-		"\tjnz 1b\n"
-		"\tjmp 6f\n"
+		"\tjz  7f\n"
+		"\tjmp 1b\n"
 
 		/*
 		 *  sample > 0x7fff
@@ -113,16 +114,13 @@ static void MIX_AREAS1(unsigned int size,
 
 		"\t.p2align 4,,15\n"
 
-		"4:"
+		"5:"
 		"\tmovw $0x7fff, (%%edi)\n"
 		"\tcmpl %%ecx,(%%ebx)\n"
-		"\tjnz 3b\n"
-		"\tadd %4, %%edi\n"
-		"\tadd %5, %%esi\n"
-		"\tadd %6, %%ebx\n"
+		"\tjnz 4b\n"
 		"\tdecl %0\n"
 		"\tjnz 1b\n"
-		"\tjmp 6f\n"
+		"\tjmp 7f\n"
 
 		/*
 		 *  sample < -0x8000
@@ -130,18 +128,15 @@ static void MIX_AREAS1(unsigned int size,
 
 		"\t.p2align 4,,15\n"
 
-		"5:"
+		"6:"
 		"\tmovw $-0x8000, (%%edi)\n"
 		"\tcmpl %%ecx, (%%ebx)\n"
-		"\tjnz 3b\n"
-		"\tadd %4, %%edi\n"
-		"\tadd %5, %%esi\n"
-		"\tadd %6, %%ebx\n"
+		"\tjnz 4b\n"
 		"\tdecl %0\n"
 		"\tjnz 1b\n"
 		// "\tjmp 6f\n"
 		
-		"6:"
+		"7:"
 
 		: /* no output regs */
 		: "m" (size), "m" (dst), "m" (src), "m" (sum), "m" (dst_step), "m" (src_step), "m" (sum_step)
@@ -174,17 +169,15 @@ static void MIX_AREAS1_MMX(unsigned int size,
 		"\tmovl %1, %%edi\n"
 		"\tmovl %2, %%esi\n"
 		"\tmovl %3, %%ebx\n"
-
-		/*
-		 * while (size-- > 0) {
-		 */
-		"\tcmpl $0, %0\n"
-		"jz 6f\n"
+		"\tjmp  2f\n"
 
 		"\t.p2align 4,,15\n"
-
 		"1:"
+		"\tadd %4, %%edi\n"
+		"\tadd %5, %%esi\n"
+		"\tadd %6, %%ebx\n"
 
+		"2:"
 		/*
 		 *   sample = *src;
 		 *   sum_sample = *sum;
@@ -192,14 +185,15 @@ static void MIX_AREAS1_MMX(unsigned int size,
 		 *     sample -= sum_sample;
 		 *   xadd(*sum, sample);
 		 */
-		"\tmovw $0, %%ax\n"
-		"\tmovw $1, %%cx\n"
+		"\txor  %%ax, %%ax\n"
+		"\tmovw %%ax, %%cx\n"
+		"\tincw %%cx\n"
 		"\tmovl (%%ebx), %%edx\n"
 		"\t" LOCK_PREFIX "cmpxchgw %%cx, (%%edi)\n"
 		"\tmovswl (%%esi), %%ecx\n"
-		"\tjnz 2f\n"
+		"\tjnz 3f\n"
 		"\tsubl %%edx, %%ecx\n"
-		"2:"
+		"3:"
 		"\t" LOCK_PREFIX "addl %%ecx, (%%ebx)\n"
 
 		/*
@@ -210,27 +204,20 @@ static void MIX_AREAS1_MMX(unsigned int size,
 		 *   } while (v != *sum);
 		 */
 
-		"3:"
+		"4:"
 		"\tmovl (%%ebx), %%ecx\n"
 		"\tmovd %%ecx, %%mm0\n"
 		"\tpackssdw %%mm1, %%mm0\n"
 		"\tmovd %%mm0, %%eax\n"
 		"\tmovw %%ax, (%%edi)\n"
 		"\tcmpl %%ecx, (%%ebx)\n"
-		"\tjnz 3b\n"
+		"\tjnz 4b\n"
 
 		/*
 		 * while (size-- > 0)
 		 */
-		"\tadd %4, %%edi\n"
-		"\tadd %5, %%esi\n"
-		"\tadd %6, %%ebx\n"
 		"\tdecl %0\n"
 		"\tjnz 1b\n"
-		"\tjmp 6f\n"
-
-		"6:"
-		
 		"\temms\n"
 
 		: /* no output regs */
@@ -265,12 +252,6 @@ static void MIX_AREAS2(unsigned int size,
 		"\tmovl %2, %%esi\n"
 		"\tmovl %3, %%ebx\n"
 
-		/*
-		 * while (size-- > 0) {
-		 */
-		"\tcmpl $0, %0\n"
-		"jz 6f\n"
-
 		"\t.p2align 4,,15\n"
 
 		"1:"
@@ -334,12 +315,12 @@ static void MIX_AREAS2(unsigned int size,
 		/*
 		 * while (size-- > 0)
 		 */
+		"\tdecl %0\n"
+		"\tjz 6f\n"
 		"\tadd %4, %%edi\n"
 		"\tadd %5, %%esi\n"
 		"\tadd %6, %%ebx\n"
-		"\tdecl %0\n"
-		"\tjnz 1b\n"
-		// "\tjmp 6f\n"
+		"\tjmp 1b\n"
 		
 		"6:"
 		: /* no output regs */

src/pcm/pcm_mmap.c

@@ -368,6 +368,11 @@ int snd_pcm_mmap(snd_pcm_t *pcm)
 						SYSERR("shmat failed");
 						return -errno;
 					}
+					/* automatically remove segment if not used */
+					if (shmctl(id, IPC_RMID, NULL) < 0){
+						SYSERR("shmctl mark remove failed");
+						return -errno;
+					}
 					i->u.shm.area = snd_shm_area_create(id, ptr);
 					if (i->u.shm.area == NULL) {
 						SYSERR("snd_shm_area_create failed");

src/shmarea.c

@@ -56,17 +56,6 @@ struct snd_shm_area *snd_shm_area_share(struct snd_shm_area *area)
 	return area;
 }
 
-static void _x_destroy(struct snd_shm_area *area)
-{
-	struct shmid_ds buf;
-
-	shmdt(area->ptr);
-	if (shmctl(area->shmid, IPC_STAT, &buf) >= 0) {
-		if (buf.shm_nattch == 0)
-			shmctl(area->shmid, IPC_RMID, NULL);
-	}
-}
-
 int snd_shm_area_destroy(struct snd_shm_area *area)
 {
 	if (area == NULL)
@@ -74,7 +63,7 @@ int snd_shm_area_destroy(struct snd_shm_area *area)
 	if (--area->share)
 		return 0;
 	list_del(&area->list);
-	_x_destroy(area);
+	shmdt(area->ptr);
 	free(area);
 	return 0;
 }
@@ -88,6 +77,6 @@ void snd_shm_area_destructor(void)
 
 	list_for_each(pos, &shm_areas) {
 		area = list_entry(pos, struct snd_shm_area, list);
-		_x_destroy(area);
+		shmdt(area->ptr);
 	}
 }