dmix: add S24_3LE support

Add to the dmix plugin support for the S24_3LE sample format which is
used by 24-bit USB devices.

The optimized assembler version uses only 23 bits for sample data so
that the lowest bit can be used for synchronization because there is no
24-bit cmpxchg instruction.

src/pcm/pcm_direct.c

@@ -800,6 +800,7 @@ int snd_pcm_direct_initialize_slave(snd_pcm_direct_t *dmix, snd_pcm_t *spcm, str
 			case SND_PCM_FORMAT_S32_BE:
 			case SND_PCM_FORMAT_S16_LE:
 			case SND_PCM_FORMAT_S16_BE:
+			case SND_PCM_FORMAT_S24_3LE:
 				break;
 			default:
 				SNDERR("invalid format");

src/pcm/pcm_direct.h

@@ -34,6 +34,11 @@ typedef void (mix_areas2_t)(unsigned int size,
 			volatile signed int *sum, size_t dst_step,
 			size_t src_step, size_t sum_step);
 
+typedef void (mix_areas3_t)(unsigned int size,
+			volatile unsigned char *dst, unsigned char *src,
+			volatile signed int *sum, size_t dst_step,
+			size_t src_step, size_t sum_step);
+
 struct slave_params {
 	snd_pcm_format_t format;
 	int rate;
@@ -120,6 +125,7 @@ struct snd_pcm_direct {
 			signed int *sum_buffer;		/* shared sum buffer */
 			mix_areas1_t *mix_areas1;
 			mix_areas2_t *mix_areas2;
+			mix_areas3_t *mix_areas3;
 		} dmix;
 		struct {
 		} dsnoop;

src/pcm/pcm_dmix.c

@@ -188,7 +188,8 @@ static void mix_areas(snd_pcm_direct_t *dmix,
 			sum = dmix->u.dmix.sum_buffer + channels * dst_ofs + chn;
 			dmix->u.dmix.mix_areas1(size, dst, src, sum, dst_step, src_step, channels * sizeof(signed int));
 		}
-	} else {
+	} else if (dmix->shmptr->s.format == SND_PCM_FORMAT_S32_LE ||
+		   dmix->shmptr->s.format == SND_PCM_FORMAT_S32_BE) {
 		signed int *src;
 		volatile signed int *dst;
 		if (dmix->interleaved) {
@@ -216,6 +217,32 @@ static void mix_areas(snd_pcm_direct_t *dmix,
 			sum = dmix->u.dmix.sum_buffer + channels * dst_ofs + chn;
 			dmix->u.dmix.mix_areas2(size, dst, src, sum, dst_step, src_step, channels * sizeof(signed int));
 		}
+	} else { /* SND_PCM_FORMAT_S24_3LE */
+		unsigned char *src;
+		volatile unsigned char *dst;
+		if (dmix->interleaved) {
+			/*
+			 * process all areas in one loop
+			 * it optimizes the memory accesses for this case
+			 */
+			dmix->u.dmix.mix_areas3(size * channels,
+					((char *)dst_areas[0].addr) + 3 * dst_ofs * channels,
+					((char *)src_areas[0].addr) + 3 * src_ofs * channels,
+					dmix->u.dmix.sum_buffer + (dst_ofs * channels),
+					3, 3, sizeof(signed int));
+			return;
+		}
+		for (chn = 0; chn < channels; chn++) {
+			dchn = dmix->bindings ? dmix->bindings[chn] : chn;
+			if (dchn >= dmix->shmptr->s.channels)
+				continue;
+			src_step = src_areas[chn].step / 8;
+			dst_step = dst_areas[dchn].step / 8;
+			src = (unsigned char *)(((char *)src_areas[chn].addr + src_areas[chn].first / 8) + (src_ofs * src_step));
+			dst = (unsigned char *)(((char *)dst_areas[dchn].addr + dst_areas[dchn].first / 8) + (dst_ofs * dst_step));
+			sum = dmix->u.dmix.sum_buffer + channels * dst_ofs + chn;
+			dmix->u.dmix.mix_areas3(size, dst, src, sum, dst_step, src_step, channels * sizeof(signed int));
+		}
 	}
 }
 

src/pcm/pcm_dmix_generic.c

@@ -121,7 +121,8 @@ static void mix_select_callbacks(snd_pcm_direct_t *dmix)
 /* non-concurrent version, supporting both endians */
 static unsigned long long dmix_supported_format =
 	(1ULL << SND_PCM_FORMAT_S16_LE) | (1ULL << SND_PCM_FORMAT_S32_LE) |
-	(1ULL << SND_PCM_FORMAT_S16_BE) | (1ULL << SND_PCM_FORMAT_S32_BE);
+	(1ULL << SND_PCM_FORMAT_S16_BE) | (1ULL << SND_PCM_FORMAT_S32_BE) |
+	(1ULL << SND_PCM_FORMAT_S24_3LE);
 
 #include <byteswap.h>
 
@@ -245,6 +246,37 @@ static void mix_areas2_swap(unsigned int size,
 	}
 }
 
+/* always little endian */
+static void mix_areas3(unsigned int size,
+		       volatile unsigned char *dst, unsigned char *src,
+		       volatile signed int *sum, size_t dst_step,
+		       size_t src_step, size_t sum_step)
+{
+	register signed int sample;
+
+	for (;;) {
+		sample = src[0] | (src[1] << 8) | (((signed char *)src)[2] << 16);
+		if (!(dst[0] | dst[1] | dst[2])) {
+			*sum = sample;
+		} else {
+			sample += *sum;
+			*sum = sample;
+			if (sample > 0x7fffff)
+				sample = 0x7fffff;
+			else if (sample < -0x800000)
+				sample = -0x800000;
+		}
+		dst[0] = sample;
+		dst[1] = sample >> 8;
+		dst[2] = sample >> 16;
+		if (!--size)
+			return;
+		dst += dst_step;
+		src += src_step;
+		sum = (signed int *) ((char *)sum + sum_step);
+	}
+}
+
 
 static void mix_select_callbacks(snd_pcm_direct_t *dmix)
 {
@@ -255,6 +287,7 @@ static void mix_select_callbacks(snd_pcm_direct_t *dmix)
 		dmix->u.dmix.mix_areas1 = mix_areas1_swap;
 		dmix->u.dmix.mix_areas2 = mix_areas2_swap;
 	}
+	dmix->u.dmix.mix_areas3 = mix_areas3;
 }
 
 #endif

src/pcm/pcm_dmix_i386.c

@@ -5,33 +5,43 @@
 #define MIX_AREAS1 mix_areas1
 #define MIX_AREAS1_MMX mix_areas1_mmx
 #define MIX_AREAS2 mix_areas2
+#define MIX_AREAS3 mix_areas3
+#define MIX_AREAS3_CMOV mix_areas3_cmov
 #define LOCK_PREFIX ""
 #include "pcm_dmix_i386.h"
 #undef MIX_AREAS1
 #undef MIX_AREAS1_MMX
 #undef MIX_AREAS2
+#undef MIX_AREAS3
+#undef MIX_AREAS3_CMOV
 #undef LOCK_PREFIX
 
 #define MIX_AREAS1 mix_areas1_smp
 #define MIX_AREAS1_MMX mix_areas1_smp_mmx
 #define MIX_AREAS2 mix_areas2_smp
+#define MIX_AREAS3 mix_areas3_smp
+#define MIX_AREAS3_CMOV mix_areas3_smp_cmov
 #define LOCK_PREFIX "lock ; "
 #include "pcm_dmix_i386.h"
 #undef MIX_AREAS1
 #undef MIX_AREAS1_MMX
 #undef MIX_AREAS2
+#undef MIX_AREAS3
+#undef MIX_AREAS3_CMOV
 #undef LOCK_PREFIX
  
 static unsigned long long dmix_supported_format =
-	(1ULL << SND_PCM_FORMAT_S16_LE) | (1ULL << SND_PCM_FORMAT_S32_LE);
+	(1ULL << SND_PCM_FORMAT_S16_LE) |
+	(1ULL << SND_PCM_FORMAT_S32_LE) |
+	(1ULL << SND_PCM_FORMAT_S24_3LE);
 
 static void mix_select_callbacks(snd_pcm_direct_t *dmix)
 {
 	FILE *in;
 	char line[255];
-	int smp = 0, mmx = 0;
+	int smp = 0, mmx = 0, cmov = 0;
 	
-	/* try to determine, if we have a MMX capable CPU */
+	/* try to determine the capabilities of the CPU */
 	in = fopen("/proc/cpuinfo", "r");
 	if (in) {
 		while (!feof(in)) {
@@ -41,15 +51,21 @@ static void mix_select_callbacks(snd_pcm_direct_t *dmix)
 			else if (!strncmp(line, "flags", 5)) {
 				if (strstr(line, " mmx"))
 					mmx = 1;
+				if (strstr(line, " cmov"))
+					cmov = 1;
 			}
 		}
 		fclose(in);
 	}
-	// printf("MMX: %i, SMP: %i\n", mmx, smp);
 	if (mmx) {
 		dmix->u.dmix.mix_areas1 = smp > 1 ? mix_areas1_smp_mmx : mix_areas1_mmx;
 	} else {
 		dmix->u.dmix.mix_areas1 = smp > 1 ? mix_areas1_smp : mix_areas1;
 	}
 	dmix->u.dmix.mix_areas2 = smp > 1 ? mix_areas2_smp : mix_areas2;
+	if (cmov) {
+		dmix->u.dmix.mix_areas3 = smp > 1 ? mix_areas3_smp_cmov : mix_areas3_cmov;
+	} else {
+		dmix->u.dmix.mix_areas3 = smp > 1 ? mix_areas3_smp: mix_areas3;
+	}
 }

src/pcm/pcm_dmix_i386.h

@@ -352,3 +352,208 @@ static void MIX_AREAS2(unsigned int size,
 		: "esi", "edi", "edx", "ecx", "eax"
 	);
 }
+
+/*
+ * 24-bit version for plain i386
+ */
+static void MIX_AREAS3(unsigned int size,
+		       volatile unsigned char *dst, unsigned char *src,
+		       volatile signed int *sum, size_t dst_step,
+		       size_t src_step, size_t sum_step)
+{
+	unsigned int old_ebx;
+
+	/*
+	 *  ESI - src
+	 *  EDI - dst
+	 *  EBX - sum
+	 *  ECX - old sample
+	 *  EAX - sample / temporary
+	 *  EDX - temporary
+	 */
+	__asm__ __volatile__ (
+		"\n"
+
+		"\tmovl %%ebx, %7\n"	/* ebx is GOT pointer (-fPIC) */
+		/*
+		 *  initialization, load ESI, EDI, EBX registers
+		 */
+		"\tmovl %1, %%edi\n"
+		"\tmovl %2, %%esi\n"
+		"\tmovl %3, %%ebx\n"
+		"\tcmpl $0, %0\n"
+		"\tjnz 1f\n"
+		"\tjmp 6f\n"
+
+		"\t.p2align 4,,15\n"
+
+		"1:"
+
+		/*
+		 *   sample = *src;
+		 *   sum_sample = *sum;
+		 *   if (test_and_set_bit(0, dst) == 0)
+		 *     sample -= sum_sample;
+		 *   *sum += sample;
+		 */
+		"\tmovsbl 2(%%esi), %%eax\n"
+		"\tmovzwl (%%esi), %%ecx\n"
+		"\tmovl (%%ebx), %%edx\n"
+		"\tsall $16, %%eax\n"
+		"\t" LOCK_PREFIX "btsl $0, (%%edi)\n"
+		"\tleal (%%ecx,%%eax,1), %%ecx\n"
+		"\tjc 2f\n"
+		"\tsubl %%edx, %%ecx\n"
+		"2:"
+		"\t" LOCK_PREFIX "addl %%ecx, (%%ebx)\n"
+
+		/*
+		 *   do {
+		 *     sample = old_sample = *sum;
+		 *     saturate(sample);
+		 *     *dst = sample | 1;
+		 *   } while (old_sample != *sum);
+		 */
+
+		"3:"
+		"\tmovl (%%ebx), %%ecx\n"
+		/*
+		 *  if (sample > 0x7fffff)
+		 */
+		"\tmovl $0x7fffff, %%eax\n"
+		"\tcmpl %%eax, %%ecx\n"
+		"\tjg 4f\n"
+		/*
+		 *  if (sample < -0x7fffff)
+		 */
+		"\tmovl $-0x7fffff, %%eax\n"
+		"\tcmpl %%eax, %%ecx\n"
+		"\tjl 4f\n"
+		"\tmovl %%ecx, %%eax\n"
+		"\torl $1, %%eax\n"
+		"4:"
+		"\tmovw %%ax, (%%edi)\n"
+		"\tshrl $16, %%eax\n"
+		"\tmovb %%al, 2(%%edi)\n"
+		"\tcmpl %%ecx, (%%ebx)\n"
+		"\tjnz 3b\n"
+
+		/*
+		 * while (size-- > 0)
+		 */
+		"\tdecl %0\n"
+		"\tjz 6f\n"
+		"\tadd %4, %%edi\n"
+		"\tadd %5, %%esi\n"
+		"\tadd %6, %%ebx\n"
+		"\tjmp 1b\n"
+		
+		"6:"
+		"\tmovl %7, %%ebx\n"	/* ebx is GOT pointer (-fPIC) */
+
+		: /* no output regs */
+		: "m" (size), "m" (dst), "m" (src),
+		  "m" (sum), "m" (dst_step), "m" (src_step),
+		  "m" (sum_step), "m" (old_ebx)
+		: "esi", "edi", "edx", "ecx", "eax"
+	);
+}
+
+/*
+ * 24-bit version for Pentium Pro/II
+ */
+static void MIX_AREAS3_CMOV(unsigned int size,
+			    volatile unsigned char *dst, unsigned char *src,
+			    volatile signed int *sum, size_t dst_step,
+			    size_t src_step, size_t sum_step)
+{
+	unsigned int old_ebx;
+
+	/*
+	 *  ESI - src
+	 *  EDI - dst
+	 *  EBX - sum
+	 *  ECX - old sample
+	 *  EAX - sample / temporary
+	 *  EDX - temporary
+	 */
+	__asm__ __volatile__ (
+		"\n"
+
+		"\tmovl %%ebx, %7\n"	/* ebx is GOT pointer (-fPIC) */
+		/*
+		 *  initialization, load ESI, EDI, EBX registers
+		 */
+		"\tmovl %1, %%edi\n"
+		"\tmovl %2, %%esi\n"
+		"\tmovl %3, %%ebx\n"
+		"\tcmpl $0, %0\n"
+		"\tjz 6f\n"
+
+		"\t.p2align 4,,15\n"
+
+		"1:"
+
+		/*
+		 *   sample = *src;
+		 *   sum_sample = *sum;
+		 *   if (test_and_set_bit(0, dst) == 0)
+		 *     sample -= sum_sample;
+		 *   *sum += sample;
+		 */
+		"\tmovsbl 2(%%esi), %%eax\n"
+		"\tmovzwl (%%esi), %%ecx\n"
+		"\tmovl (%%ebx), %%edx\n"
+		"\tsall $16, %%eax\n"
+		"\t" LOCK_PREFIX "btsl $0, (%%edi)\n"
+		"\tleal (%%ecx,%%eax,1), %%ecx\n"
+		"\tjc 2f\n"
+		"\tsubl %%edx, %%ecx\n"
+		"2:"
+		"\t" LOCK_PREFIX "addl %%ecx, (%%ebx)\n"
+
+		/*
+		 *   do {
+		 *     sample = old_sample = *sum;
+		 *     saturate(sample);
+		 *     *dst = sample | 1;
+		 *   } while (old_sample != *sum);
+		 */
+
+		"3:"
+		"\tmovl (%%ebx), %%ecx\n"
+
+		"\tmovl $0x7fffff, %%eax\n"
+		"\tmovl $-0x7fffff, %%edx\n"
+		"\tcmpl %%eax, %%ecx\n"
+		"\tcmovng %%ecx, %%eax\n"
+		"\tcmpl %%edx, %%ecx\n"
+		"\tcmovl %%edx, %%eax\n"
+
+		"\torl $1, %%eax\n"
+		"\tmovw %%ax, (%%edi)\n"
+		"\tshrl $16, %%eax\n"
+		"\tmovb %%al, 2(%%edi)\n"
+
+		"\tcmpl %%ecx, (%%ebx)\n"
+		"\tjnz 3b\n"
+
+		/*
+		 * while (size-- > 0)
+		 */
+		"\tadd %4, %%edi\n"
+		"\tadd %5, %%esi\n"
+		"\tadd %6, %%ebx\n"
+		"\tdecl %0\n"
+		"\tjnz 1b\n"
+		
+		"6:"
+		"\tmovl %7, %%ebx\n"	/* ebx is GOT pointer (-fPIC) */
+
+		: /* no output regs */
+		: "m" (size), "m" (dst), "m" (src),
+		  "m" (sum), "m" (dst_step), "m" (src_step),
+		  "m" (sum_step), "m" (old_ebx)
+		: "esi", "edi", "edx", "ecx", "eax"
+	);
+}

src/pcm/pcm_dmix_x86_64.c

@@ -4,22 +4,28 @@
 
 #define MIX_AREAS1 mix_areas1
 #define MIX_AREAS2 mix_areas2
+#define MIX_AREAS3 mix_areas3
 #define LOCK_PREFIX ""
 #include "pcm_dmix_x86_64.h"
 #undef MIX_AREAS1
 #undef MIX_AREAS2
+#undef MIX_AREAS3
 #undef LOCK_PREFIX
 
 #define MIX_AREAS1 mix_areas1_smp
 #define MIX_AREAS2 mix_areas2_smp
+#define MIX_AREAS3 mix_areas3_smp
 #define LOCK_PREFIX "lock ; "
 #include "pcm_dmix_x86_64.h"
 #undef MIX_AREAS1
 #undef MIX_AREAS2
+#undef MIX_AREAS3
 #undef LOCK_PREFIX
  
 static unsigned long long dmix_supported_format =
-	(1ULL << SND_PCM_FORMAT_S16_LE) | (1ULL << SND_PCM_FORMAT_S32_LE);
+	(1ULL << SND_PCM_FORMAT_S16_LE) |
+	(1ULL << SND_PCM_FORMAT_S32_LE) |
+	(1ULL << SND_PCM_FORMAT_S24_3LE);
 
 static void mix_select_callbacks(snd_pcm_direct_t *dmix)
 {
@@ -40,4 +46,5 @@ static void mix_select_callbacks(snd_pcm_direct_t *dmix)
 	// printf("SMP: %i\n", smp);
 	dmix->u.dmix.mix_areas1 = smp > 1 ? mix_areas1_smp : mix_areas1;
 	dmix->u.dmix.mix_areas2 = smp > 1 ? mix_areas2_smp : mix_areas2;
+	dmix->u.dmix.mix_areas3 = smp > 1 ? mix_areas3_smp : mix_areas3;
 }

src/pcm/pcm_dmix_x86_64.h

@@ -237,3 +237,105 @@ static void MIX_AREAS2(unsigned int size,
 	);
 }
 
+/*
+ *  24-bit version
+ */
+static void MIX_AREAS3(unsigned int size,
+		       volatile unsigned char *dst, unsigned char *src,
+		       volatile signed int *sum, size_t dst_step,
+		       size_t src_step, size_t sum_step)
+{
+	unsigned long long old_rbx;
+
+	/*
+	 *  RSI - src
+	 *  RDI - dst
+	 *  RBX - sum
+	 *  ECX - old sample
+	 *  EAX - sample / temporary
+	 *  EDX - temporary
+	 */
+	__asm__ __volatile__ (
+		"\n"
+
+		"\tmovq %%rbx, %7\n"
+		/*
+		 *  initialization, load ESI, EDI, EBX registers
+		 */
+		"\tmovq %1, %%rdi\n"
+		"\tmovq %2, %%rsi\n"
+		"\tmovq %3, %%rbx\n"
+
+		/*
+		 * while (size-- > 0) {
+		 */
+		"\tcmpl $0, %0\n"
+		"jz 6f\n"
+
+		"\t.p2align 4,,15\n"
+
+		"1:"
+
+		/*
+		 *   sample = *src;
+		 *   sum_sample = *sum;
+		 *   if (test_and_set_bit(0, dst) == 0)
+		 *     sample -= sum_sample;
+		 *   *sum += sample;
+		 */
+		"\tmovsbl 2(%%rsi), %%eax\n"
+		"\tmovswl (%%rsi), %%ecx\n"
+		"\tmovl (%%rbx), %%edx\n"
+		"\tsall $16, %%eax\n"
+		"\t" LOCK_PREFIX "btsl $0, (%%rdi)\n"
+		"\tleal (%%ecx,%%eax,1), %%ecx\n"
+		"\tjc 2f\n"
+		"\tsubl %%edx, %%ecx\n"
+		"2:"
+		"\t" LOCK_PREFIX "addl %%ecx, (%%rbx)\n"
+
+		/*
+		 *   do {
+		 *     sample = old_sample = *sum;
+		 *     saturate(sample);
+		 *     *dst = sample | 1;
+		 *   } while (old_sample != *sum);
+		 */
+
+		"3:"
+		"\tmovl (%%rbx), %%ecx\n"
+
+		"\tmovl $0x7fffff, %%eax\n"
+		"\tmovl $-0x7fffff, %%edx\n"
+		"\tcmpl %%eax, %%ecx\n"
+		"\tcmovng %%ecx, %%eax\n"
+		"\tcmpl %%edx, %%ecx\n"
+		"\tcmovl %%edx, %%eax\n"
+
+		"\torl $1, %%eax\n"
+		"\tmovw %%ax, (%%rdi)\n"
+		"\tshrl $16, %%eax\n"
+		"\tmovb %%al, 2(%%rdi)\n"
+	
+		"\tcmpl %%ecx, (%%rbx)\n"
+		"\tjnz 3b\n"
+
+		/*
+		 * while (size-- > 0)
+		 */
+		"\tadd %4, %%rdi\n"
+		"\tadd %5, %%rsi\n"
+		"\tadd %6, %%rbx\n"
+		"\tdecl %0\n"
+		"\tjnz 1b\n"
+		
+		"6:"
+		"\tmovq %7, %%rbx\n"
+
+		: /* no output regs */
+		: "m" (size), "m" (dst), "m" (src),
+		  "m" (sum), "m" (dst_step), "m" (src_step),
+		  "m" (sum_step), "m" (old_rbx)
+		: "rsi", "rdi", "edx", "ecx", "eax"
+	);
+}