[PATCH 1/6 V2] [configure] check if the compiler is aware of the MMX technology

older
Issue with Asus UX534FTC / Realtek...

frederic.recoules＠univ-grenoble-alpes.fr

29 Apr 2020 29 Apr '20

9:10 p.m.

From: Frédéric Recoules frederic.recoules@orange.fr

Signed-off-by: Frédéric Recoules frederic.recoules@orange.fr --- configure.ac | 7 +++++++ 1 file changed, 7 insertions(+)

diff --git a/configure.ac b/configure.ac index 4b5ab662..1838e50b 100644 --- a/configure.ac +++ b/configure.ac @@ -516,6 +516,13 @@ if test -z "$gcc_have_atomics"; then fi AC_MSG_RESULT($gcc_have_atomics)

+dnl check mmx register for pcm_dmix_i386 + +AC_TRY_LINK([], + [__asm__ volatile ("" : : : "mm0");], + [AC_DEFINE([HAVE_MMX], "1", [MMX technology is enabled])], + []) + PCM_PLUGIN_LIST="copy linear route mulaw alaw adpcm rate plug multi shm file null empty share meter hooks lfloat ladspa dmix dshare dsnoop asym iec958 softvol extplug ioplug mmap_emul"

build_pcm_plugin="no" -- 2.17.1

Show replies by date

frederic.recoules＠univ-grenoble-alpes.fr

29 Apr 29 Apr

9:10 p.m.

New subject: [PATCH 2/6 V2] [pcm_dmix assembly] change the token by symbolic names

From: Frédéric Recoules frederic.recoules@orange.fr

Note: produce exactly the same binary output.

Signed-off-by: Frédéric Recoules frederic.recoules@orange.fr --- src/pcm/pcm_dmix_i386.h | 147 ++++++++++++++++++++------------------ src/pcm/pcm_dmix_x86_64.h | 93 ++++++++++++------------ 2 files changed, 124 insertions(+), 116 deletions(-)

diff --git a/src/pcm/pcm_dmix_i386.h b/src/pcm/pcm_dmix_i386.h index 2778cb1d..e816d5e0 100644 --- a/src/pcm/pcm_dmix_i386.h +++ b/src/pcm/pcm_dmix_i386.h @@ -47,14 +47,14 @@ static void MIX_AREAS_16(unsigned int size, __asm__ __volatile__ ( "\n"

- "\tmovl %%ebx, %7\n" /* ebx is GOT pointer (-fPIC) */ + "\tmovl %%ebx, %[old_ebx]\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" + "\tmovl %[dst], %%edi\n" + "\tmovl %[src], %%esi\n" + "\tmovl %[sum], %%ebx\n" + "\tcmpl $0, %[size]\n" "\tjnz 2f\n" "\tjmp 7f\n"

@@ -64,9 +64,9 @@ static void MIX_AREAS_16(unsigned int size, */ "\t.p2align 4,,15\n" "1:" - "\tadd %4, %%edi\n" - "\tadd %5, %%esi\n" - "\tadd %6, %%ebx\n" + "\tadd %[dst_step], %%edi\n" + "\tadd %[src_step], %%esi\n" + "\tadd %[sum_step], %%ebx\n"

/* * sample = *src; @@ -108,7 +108,7 @@ static void MIX_AREAS_16(unsigned int size, /* * while (size-- > 0) */ - "\tdecl %0\n" + "\tdecl %[size]\n" "\tjnz 1b\n" "\tjmp 7f\n"

@@ -122,7 +122,7 @@ static void MIX_AREAS_16(unsigned int size, "\tmovw $0x7fff, (%%edi)\n" "\tcmpl %%ecx,(%%ebx)\n" "\tjnz 4b\n" - "\tdecl %0\n" + "\tdecl %[size]\n" "\tjnz 1b\n" "\tjmp 7f\n"

@@ -136,16 +136,17 @@ static void MIX_AREAS_16(unsigned int size, "\tmovw $-0x8000, (%%edi)\n" "\tcmpl %%ecx, (%%ebx)\n" "\tjnz 4b\n" - "\tdecl %0\n" + "\tdecl %[size]\n" "\tjnz 1b\n" - + "7:" - "\tmovl %7, %%ebx\n" /* ebx is GOT pointer (-fPIC) */ + "\tmovl %[old_ebx], %%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) + : [size] "m" (size), + [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), + [dst_step] "m" (dst_step), [src_step] "m" (src_step), + [sum_step] "m" (sum_step), [old_ebx] "m" (old_ebx) : "esi", "edi", "edx", "ecx", "eax" ); } @@ -171,22 +172,22 @@ static void MIX_AREAS_16_MMX(unsigned int size, __asm__ __volatile__ ( "\n"

"\t.p2align 4,,15\n" "1:" - "\tadd %4, %%edi\n" - "\tadd %5, %%esi\n" - "\tadd %6, %%ebx\n" + "\tadd %[dst_step], %%edi\n" + "\tadd %[src_step], %%esi\n" + "\tadd %[sum_step], %%ebx\n"

"2:" /* @@ -226,16 +227,17 @@ static void MIX_AREAS_16_MMX(unsigned int size, /* * while (size-- > 0) */ - "\tdecl %0\n" + "\tdecl %[size]\n" "\tjnz 1b\n" "\temms\n" "5:" - "\tmovl %7, %%ebx\n" /* ebx is GOT pointer (-fPIC) */ + "\tmovl %[old_ebx], %%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) + : [size] "m" (size), + [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), + [dst_step] "m" (dst_step), [src_step] "m" (src_step), + [sum_step] "m" (sum_step), [old_ebx] "m" (old_ebx) : "esi", "edi", "edx", "ecx", "eax" ); } @@ -261,14 +263,14 @@ static void MIX_AREAS_32(unsigned int size, __asm__ __volatile__ ( "\n"

@@ -335,20 +337,21 @@ static void MIX_AREAS_32(unsigned int size, /* * while (size-- > 0) */ - "\tdecl %0\n" + "\tdecl %[size]\n" "\tjz 6f\n" - "\tadd %4, %%edi\n" - "\tadd %5, %%esi\n" - "\tadd %6, %%ebx\n" + "\tadd %[dst_step], %%edi\n" + "\tadd %[src_step], %%esi\n" + "\tadd %[sum_step], %%ebx\n" "\tjmp 1b\n" - + "6:" - "\tmovl %7, %%ebx\n" /* ebx is GOT pointer (-fPIC) */ + "\tmovl %[old_ebx], %%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) + : [size] "m" (size), + [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), + [dst_step] "m" (dst_step), [src_step] "m" (src_step), + [sum_step] "m" (sum_step), [old_ebx] "m" (old_ebx) : "esi", "edi", "edx", "ecx", "eax" ); } @@ -374,14 +377,14 @@ static void MIX_AREAS_24(unsigned int size, __asm__ __volatile__ ( "\n"

@@ -441,20 +444,21 @@ static void MIX_AREAS_24(unsigned int size, /* * while (size-- > 0) */ - "\tdecl %0\n" + "\tdecl %[size]\n" "\tjz 6f\n" - "\tadd %4, %%edi\n" - "\tadd %5, %%esi\n" - "\tadd %6, %%ebx\n" + "\tadd %[dst_step], %%edi\n" + "\tadd %[src_step], %%esi\n" + "\tadd %[sum_step], %%ebx\n" "\tjmp 1b\n" - + "6:" - "\tmovl %7, %%ebx\n" /* ebx is GOT pointer (-fPIC) */ + "\tmovl %[old_ebx], %%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) + : [size] "m" (size), + [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), + [dst_step] "m" (dst_step), [src_step] "m" (src_step), + [sum_step] "m" (sum_step), [old_ebx] "m" (old_ebx) : "esi", "edi", "edx", "ecx", "eax" ); } @@ -480,14 +484,14 @@ static void MIX_AREAS_24_CMOV(unsigned int size, __asm__ __volatile__ ( "\n"

"\t.p2align 4,,15\n" @@ -541,19 +545,20 @@ static void MIX_AREAS_24_CMOV(unsigned int size, /* * while (size-- > 0) */ - "\tadd %4, %%edi\n" - "\tadd %5, %%esi\n" - "\tadd %6, %%ebx\n" - "\tdecl %0\n" + "\tadd %[dst_step], %%edi\n" + "\tadd %[src_step], %%esi\n" + "\tadd %[sum_step], %%ebx\n" + "\tdecl %[size]\n" "\tjnz 1b\n" - + "6:" - "\tmovl %7, %%ebx\n" /* ebx is GOT pointer (-fPIC) */ + "\tmovl %[old_ebx], %%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) + : [size] "m" (size), + [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), + [dst_step] "m" (dst_step), [src_step] "m" (src_step), + [sum_step] "m" (sum_step), [old_ebx] "m" (old_ebx) : "esi", "edi", "edx", "ecx", "eax" ); } diff --git a/src/pcm/pcm_dmix_x86_64.h b/src/pcm/pcm_dmix_x86_64.h index 1ef608af..dc433b35 100644 --- a/src/pcm/pcm_dmix_x86_64.h +++ b/src/pcm/pcm_dmix_x86_64.h @@ -48,18 +48,18 @@ static void MIX_AREAS_16(unsigned int size, __asm__ __volatile__ ( "\n"

- "\tmovq %%rbx, %7\n" + "\tmovq %%rbx, %[old_rbx]\n" /* * initialization, load RSI, RDI, RBX registers */ - "\tmovq %1, %%rdi\n" - "\tmovq %2, %%rsi\n" - "\tmovq %3, %%rbx\n" + "\tmovq %[dst], %%rdi\n" + "\tmovq %[src], %%rsi\n" + "\tmovq %[sum], %%rbx\n"

/* * while (size-- > 0) { */ - "\tcmpl $0, %0\n" + "\tcmpl $0, %[size]\n" "jz 6f\n"

"\t.p2align 4,,15\n" @@ -103,21 +103,22 @@ static void MIX_AREAS_16(unsigned int size, /* * while (size-- > 0) */ - "\tadd %4, %%rdi\n" - "\tadd %5, %%rsi\n" - "\tadd %6, %%rbx\n" - "\tdecl %0\n" + "\tadd %[dst_step], %%rdi\n" + "\tadd %[src_step], %%rsi\n" + "\tadd %[sum_step], %%rbx\n" + "\tdecl %[size]\n" "\tjnz 1b\n"

"6:" - + "\temms\n" - "\tmovq %7, %%rbx\n" + "\tmovq %[old_rbx], %%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) + : [size] "m" (size), + [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), + [dst_step] "m" (dst_step), [src_step] "m" (src_step), + [sum_step] "m" (sum_step), [old_rbx] "m" (old_rbx) : "rsi", "rdi", "edx", "ecx", "eax" ); } @@ -143,18 +144,18 @@ static void MIX_AREAS_32(unsigned int size, __asm__ __volatile__ ( "\n"

- "\tmovq %%rbx, %7\n" + "\tmovq %%rbx, %[old_rbx]\n" /* - * initialization, load ESI, EDI, EBX registers + * initialization, load RSI, RDI, RBX registers */ - "\tmovq %1, %%rdi\n" - "\tmovq %2, %%rsi\n" - "\tmovq %3, %%rbx\n" + "\tmovq %[dst], %%rdi\n" + "\tmovq %[src], %%rsi\n" + "\tmovq %[sum], %%rbx\n"

/* * while (size-- > 0) { */ - "\tcmpl $0, %0\n" + "\tcmpl $0, %[size]\n" "jz 6f\n"

"\t.p2align 4,,15\n" @@ -220,19 +221,20 @@ static void MIX_AREAS_32(unsigned int size, /* * while (size-- > 0) */ - "\tadd %4, %%rdi\n" - "\tadd %5, %%rsi\n" - "\tadd %6, %%rbx\n" - "\tdecl %0\n" + "\tadd %[dst_step], %%rdi\n" + "\tadd %[src_step], %%rsi\n" + "\tadd %[sum_step], %%rbx\n" + "\tdecl %[size]\n" "\tjnz 1b\n" - + "6:" - "\tmovq %7, %%rbx\n" + "\tmovq %[old_rbx], %%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) + : [size] "m" (size), + [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), + [dst_step] "m" (dst_step), [src_step] "m" (src_step), + [sum_step] "m" (sum_step), [old_rbx] "m" (old_rbx) : "rsi", "rdi", "edx", "ecx", "eax" ); } @@ -258,18 +260,18 @@ static void MIX_AREAS_24(unsigned int size, __asm__ __volatile__ ( "\n"

/* * while (size-- > 0) { */ - "\tcmpl $0, %0\n" + "\tcmpl $0, %[size]\n" "jz 6f\n"

"\t.p2align 4,,15\n" @@ -316,26 +318,27 @@ static void MIX_AREAS_24(unsigned int size, "\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" + "\tadd %[dst_step], %%rdi\n" + "\tadd %[src_step], %%rsi\n" + "\tadd %[sum_step], %%rbx\n" + "\tdecl %[size]\n" "\tjnz 1b\n" - + "6:" - "\tmovq %7, %%rbx\n" + "\tmovq %[old_rbx], %%rbx\n"

frederic.recoules＠univ-grenoble-alpes.fr

9:10 p.m.

New subject: [PATCH 3/6 V2] [pcm_dmix assembly] fix the interface for safety

From: Frédéric Recoules frederic.recoules@orange.fr

- move 'size' and 'old_ebx' in the output section since they are clobbered; - add the "memory" clobber since input pointers are dereferenced; - (minor) add the "cc" clobber since flags are clobbered.

Note: produce exactly the same binary output.

Signed-off-by: Frédéric Recoules frederic.recoules@orange.fr --- src/pcm/pcm_dmix_i386.h | 45 +++++++++++++++++---------------------- src/pcm/pcm_dmix_x86_64.h | 27 +++++++++++------------ 2 files changed, 32 insertions(+), 40 deletions(-)

diff --git a/src/pcm/pcm_dmix_i386.h b/src/pcm/pcm_dmix_i386.h index e816d5e0..3ea0737d 100644 --- a/src/pcm/pcm_dmix_i386.h +++ b/src/pcm/pcm_dmix_i386.h @@ -142,12 +142,11 @@ static void MIX_AREAS_16(unsigned int size, "7:" "\tmovl %[old_ebx], %%ebx\n" /* ebx is GOT pointer (-fPIC) */

@@ -233,12 +232,11 @@ static void MIX_AREAS_16_MMX(unsigned int size, "5:" "\tmovl %[old_ebx], %%ebx\n" /* ebx is GOT pointer (-fPIC) */

@@ -347,12 +345,11 @@ static void MIX_AREAS_32(unsigned int size, "6:" "\tmovl %[old_ebx], %%ebx\n" /* ebx is GOT pointer (-fPIC) */

@@ -454,12 +451,11 @@ static void MIX_AREAS_24(unsigned int size, "6:" "\tmovl %[old_ebx], %%ebx\n" /* ebx is GOT pointer (-fPIC) */

@@ -554,11 +550,10 @@ static void MIX_AREAS_24_CMOV(unsigned int size, "6:" "\tmovl %[old_ebx], %%ebx\n" /* ebx is GOT pointer (-fPIC) */

- : /* no output regs */ - : [size] "m" (size), - [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), + : [size] "+m" (size), [old_ebx] "=m" (old_ebx) + : [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), [dst_step] "m" (dst_step), [src_step] "m" (src_step), - [sum_step] "m" (sum_step), [old_ebx] "m" (old_ebx) - : "esi", "edi", "edx", "ecx", "eax" + [sum_step] "m" (sum_step) + : "esi", "edi", "edx", "ecx", "eax", "memory", "cc" ); } diff --git a/src/pcm/pcm_dmix_x86_64.h b/src/pcm/pcm_dmix_x86_64.h index dc433b35..7f711547 100644 --- a/src/pcm/pcm_dmix_x86_64.h +++ b/src/pcm/pcm_dmix_x86_64.h @@ -114,12 +114,11 @@ static void MIX_AREAS_16(unsigned int size, "\temms\n" "\tmovq %[old_rbx], %%rbx\n"

- : /* no output regs */ - : [size] "m" (size), - [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), + : [size] "+m" (size), [old_rbx] "=m" (old_rbx) + : [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), [dst_step] "m" (dst_step), [src_step] "m" (src_step), - [sum_step] "m" (sum_step), [old_rbx] "m" (old_rbx) - : "rsi", "rdi", "edx", "ecx", "eax" + [sum_step] "m" (sum_step) + : "rsi", "rdi", "edx", "ecx", "eax", "memory", "cc" ); }

@@ -230,12 +229,11 @@ static void MIX_AREAS_32(unsigned int size, "6:" "\tmovq %[old_rbx], %%rbx\n"

@@ -334,11 +332,10 @@ static void MIX_AREAS_24(unsigned int size, "6:" "\tmovq %[old_rbx], %%rbx\n"

frederic.recoules＠univ-grenoble-alpes.fr

9:10 p.m.

New subject: [PATCH 4/6 V2] [pcm_dmix assembly] fix the interface for safety (mmx)

From: Frédéric Recoules frederic.recoules@orange.fr

- initialize mm1 with 0s; - add mm0 and mm1 in the clobber list if the compiler is aware of the mmx technology; - otherwise, add the mmx aliased x87 floating point register in the clobbers.

Note: produce exactly the same binary output except for mm1 initialization implying address shifts.

Signed-off-by: Frédéric Recoules frederic.recoules@orange.fr --- src/pcm/pcm_dmix_i386.h | 9 ++++++++- src/pcm/pcm_dmix_x86_64.h | 9 ++++++++- 2 files changed, 16 insertions(+), 2 deletions(-)

diff --git a/src/pcm/pcm_dmix_i386.h b/src/pcm/pcm_dmix_i386.h index 3ea0737d..d9bf8f50 100644 --- a/src/pcm/pcm_dmix_i386.h +++ b/src/pcm/pcm_dmix_i386.h @@ -173,8 +173,9 @@ static void MIX_AREAS_16_MMX(unsigned int size,

"\tmovl %%ebx, %[old_ebx]\n" /* ebx is GOT pointer (-fPIC) */ /* - * initialization, load ESI, EDI, EBX registers + * initialization, load ESI, EDI, EBX registers and clear MM1 */ + "\tpxor %%mm1, %%mm1\n" "\tmovl %[dst], %%edi\n" "\tmovl %[src], %%esi\n" "\tmovl %[sum], %%ebx\n" @@ -237,6 +238,12 @@ static void MIX_AREAS_16_MMX(unsigned int size, [dst_step] "m" (dst_step), [src_step] "m" (src_step), [sum_step] "m" (sum_step) : "esi", "edi", "edx", "ecx", "eax", "memory", "cc" +# ifdef HAVE_MMX + , "mm0", "mm1" +# else + , "st", "st(1)", "st(2)", "st(3)", + "st(4)", "st(5)", "st(6)", "st(7)" +# endif ); }

diff --git a/src/pcm/pcm_dmix_x86_64.h b/src/pcm/pcm_dmix_x86_64.h index 7f711547..8ad1b2a0 100644 --- a/src/pcm/pcm_dmix_x86_64.h +++ b/src/pcm/pcm_dmix_x86_64.h @@ -50,8 +50,9 @@ static void MIX_AREAS_16(unsigned int size,

"\tmovq %%rbx, %[old_rbx]\n" /* - * initialization, load RSI, RDI, RBX registers + * initialization, load RSI, RDI, RBX registers and clear MM1 */ + "\tpxor %%mm1, %%mm1\n" "\tmovq %[dst], %%rdi\n" "\tmovq %[src], %%rsi\n" "\tmovq %[sum], %%rbx\n" @@ -119,6 +120,12 @@ static void MIX_AREAS_16(unsigned int size, [dst_step] "m" (dst_step), [src_step] "m" (src_step), [sum_step] "m" (sum_step) : "rsi", "rdi", "edx", "ecx", "eax", "memory", "cc" +# ifdef HAVE_MMX + , "mm0", "mm1" +# else + , "st", "st(1)", "st(2)", "st(3)", + "st(4)", "st(5)", "st(6)", "st(7)" +# endif ); }

-- 2.17.1

frederic.recoules＠univ-grenoble-alpes.fr

9:10 p.m.

New subject: [PATCH 5/6 V2] [pcm_dmix assembly] refactor the PIC mode

From: Frédéric Recoules frederic.recoules@orange.fr

Note: newer versions of GCC accept that ebx is declared in the clobber list.

Signed-off-by: Frédéric Recoules frederic.recoules@orange.fr --- src/pcm/pcm_dmix_i386.h | 93 ++++++++++++++++++++++++++++++--------- src/pcm/pcm_dmix_x86_64.h | 59 ++++++++++++++++++++----- 2 files changed, 120 insertions(+), 32 deletions(-)

diff --git a/src/pcm/pcm_dmix_i386.h b/src/pcm/pcm_dmix_i386.h index d9bf8f50..b28ae787 100644 --- a/src/pcm/pcm_dmix_i386.h +++ b/src/pcm/pcm_dmix_i386.h @@ -26,6 +26,10 @@ * */

+#if defined(__GNUC__) && __GNUC__ < 5 && defined(__PIC__) +# define BOUNDED_EBX +#endif + /* * for plain i386 */ @@ -34,8 +38,9 @@ static void MIX_AREAS_16(unsigned int size, volatile signed int *sum, size_t dst_step, size_t src_step, size_t sum_step) { +# ifdef BOUNDED_EBX unsigned int old_ebx; - +# endif /* * ESI - src * EDI - dst @@ -46,8 +51,9 @@ static void MIX_AREAS_16(unsigned int size, */ __asm__ __volatile__ ( "\n" - +# ifdef BOUNDED_EBX "\tmovl %%ebx, %[old_ebx]\n" /* ebx is GOT pointer (-fPIC) */ +# endif /* * initialization, load ESI, EDI, EBX registers */ @@ -140,13 +146,20 @@ static void MIX_AREAS_16(unsigned int size, "\tjnz 1b\n"

"7:" +# ifdef BOUNDED_EBX "\tmovl %[old_ebx], %%ebx\n" /* ebx is GOT pointer (-fPIC) */ - - : [size] "+m" (size), [old_ebx] "=m" (old_ebx) +# endif + : [size] "+m" (size) +# ifdef BOUNDED_EBX + , [old_ebx] "=m" (old_ebx) +# endif : [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), [dst_step] "m" (dst_step), [src_step] "m" (src_step), [sum_step] "m" (sum_step) : "esi", "edi", "edx", "ecx", "eax", "memory", "cc" +# ifndef BOUNDED_EBX + , "ebx" +# endif ); }

@@ -158,8 +171,9 @@ static void MIX_AREAS_16_MMX(unsigned int size, volatile signed int *sum, size_t dst_step, size_t src_step, size_t sum_step) { +# ifdef BOUNDED_EBX unsigned int old_ebx; - +# endif /* * ESI - src * EDI - dst @@ -170,8 +184,9 @@ static void MIX_AREAS_16_MMX(unsigned int size, */ __asm__ __volatile__ ( "\n" - +# ifdef BOUNDED_EBX "\tmovl %%ebx, %[old_ebx]\n" /* ebx is GOT pointer (-fPIC) */ +# endif /* * initialization, load ESI, EDI, EBX registers and clear MM1 */ @@ -231,13 +246,20 @@ static void MIX_AREAS_16_MMX(unsigned int size, "\tjnz 1b\n" "\temms\n" "5:" +# ifdef BOUNDED_EBX "\tmovl %[old_ebx], %%ebx\n" /* ebx is GOT pointer (-fPIC) */ - - : [size] "+m" (size), [old_ebx] "=m" (old_ebx) +# endif + : [size] "+m" (size) +# ifdef BOUNDED_EBX + , [old_ebx] "=m" (old_ebx) +# endif : [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), [dst_step] "m" (dst_step), [src_step] "m" (src_step), [sum_step] "m" (sum_step) : "esi", "edi", "edx", "ecx", "eax", "memory", "cc" +# ifndef BOUNDED_EBX + , "ebx" +# endif # ifdef HAVE_MMX , "mm0", "mm1" # else @@ -255,8 +277,9 @@ static void MIX_AREAS_32(unsigned int size, volatile signed int *sum, size_t dst_step, size_t src_step, size_t sum_step) { +# ifdef BOUNDED_EBX unsigned int old_ebx; - +# endif /* * ESI - src * EDI - dst @@ -267,8 +290,9 @@ static void MIX_AREAS_32(unsigned int size, */ __asm__ __volatile__ ( "\n" - +# ifdef BOUNDED_EBX "\tmovl %%ebx, %[old_ebx]\n" /* ebx is GOT pointer (-fPIC) */ +# endif /* * initialization, load ESI, EDI, EBX registers */ @@ -350,13 +374,20 @@ static void MIX_AREAS_32(unsigned int size, "\tjmp 1b\n"

@@ -368,8 +399,9 @@ static void MIX_AREAS_24(unsigned int size, volatile signed int *sum, size_t dst_step, size_t src_step, size_t sum_step) { +# ifdef BOUNDED_EBX unsigned int old_ebx; - +# endif /* * ESI - src * EDI - dst @@ -380,8 +412,9 @@ static void MIX_AREAS_24(unsigned int size, */ __asm__ __volatile__ ( "\n" - +# ifdef BOUNDED_EBX "\tmovl %%ebx, %[old_ebx]\n" /* ebx is GOT pointer (-fPIC) */ +# endif /* * initialization, load ESI, EDI, EBX registers */ @@ -456,13 +489,20 @@ static void MIX_AREAS_24(unsigned int size, "\tjmp 1b\n"

@@ -474,8 +514,9 @@ static void MIX_AREAS_24_CMOV(unsigned int size, volatile signed int *sum, size_t dst_step, size_t src_step, size_t sum_step) { +# ifdef BOUNDED_EBX unsigned int old_ebx; - +# endif /* * ESI - src * EDI - dst @@ -486,8 +527,9 @@ static void MIX_AREAS_24_CMOV(unsigned int size, */ __asm__ __volatile__ ( "\n" - +# ifdef BOUNDED_EBX "\tmovl %%ebx, %[old_ebx]\n" /* ebx is GOT pointer (-fPIC) */ +# endif /* * initialization, load ESI, EDI, EBX registers */ @@ -555,12 +597,23 @@ static void MIX_AREAS_24_CMOV(unsigned int size, "\tjnz 1b\n"

"6:" +# ifdef BOUNDED_EBX "\tmovl %[old_ebx], %%ebx\n" /* ebx is GOT pointer (-fPIC) */ - - : [size] "+m" (size), [old_ebx] "=m" (old_ebx) +# endif + : [size] "+m" (size) +# ifdef BOUNDED_EBX + , [old_ebx] "=m" (old_ebx) +# endif : [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), [dst_step] "m" (dst_step), [src_step] "m" (src_step), [sum_step] "m" (sum_step) : "esi", "edi", "edx", "ecx", "eax", "memory", "cc" +# ifndef BOUNDED_EBX + , "ebx" +# endif ); } + +#ifdef BOUNDED_EBX +# undef BOUNDED_EBX +#endif diff --git a/src/pcm/pcm_dmix_x86_64.h b/src/pcm/pcm_dmix_x86_64.h index 8ad1b2a0..f543a134 100644 --- a/src/pcm/pcm_dmix_x86_64.h +++ b/src/pcm/pcm_dmix_x86_64.h @@ -27,6 +27,10 @@ * */

+#if defined(__GNUC__) && __GNUC__ < 5 && defined(__PIC__) +# define BOUNDED_RBX +#endif + /* * MMX optimized */ @@ -35,8 +39,9 @@ static void MIX_AREAS_16(unsigned int size, volatile signed int *sum, size_t dst_step, size_t src_step, size_t sum_step) { +# ifdef BOUNDED_RBX unsigned long long old_rbx; - +# endif /* * RSI - src * RDI - dst @@ -47,8 +52,9 @@ static void MIX_AREAS_16(unsigned int size, */ __asm__ __volatile__ ( "\n" - +# ifdef BOUNDED_RBX "\tmovq %%rbx, %[old_rbx]\n" +# endif /* * initialization, load RSI, RDI, RBX registers and clear MM1 */ @@ -113,13 +119,20 @@ static void MIX_AREAS_16(unsigned int size, "6:"

"\temms\n" +# ifdef BOUNDED_RBX "\tmovq %[old_rbx], %%rbx\n" - - : [size] "+m" (size), [old_rbx] "=m" (old_rbx) +# endif + : [size] "+m" (size) +# ifdef BOUNDED_RBX + , [old_rbx] "=m" (old_rbx) +# endif : [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), [dst_step] "m" (dst_step), [src_step] "m" (src_step), [sum_step] "m" (sum_step) : "rsi", "rdi", "edx", "ecx", "eax", "memory", "cc" +# ifndef BOUNDED_RBX + , "rbx" +# endif # ifdef HAVE_MMX , "mm0", "mm1" # else @@ -137,8 +150,9 @@ static void MIX_AREAS_32(unsigned int size, volatile signed int *sum, size_t dst_step, size_t src_step, size_t sum_step) { +# ifdef BOUNDED_RBX unsigned long long old_rbx; - +# endif /* * RSI - src * RDI - dst @@ -149,8 +163,9 @@ static void MIX_AREAS_32(unsigned int size, */ __asm__ __volatile__ ( "\n" - +# ifdef BOUNDED_RBX "\tmovq %%rbx, %[old_rbx]\n" +# endif /* * initialization, load RSI, RDI, RBX registers */ @@ -234,13 +249,20 @@ static void MIX_AREAS_32(unsigned int size, "\tjnz 1b\n"

"6:" +# ifdef BOUNDED_RBX "\tmovq %[old_rbx], %%rbx\n" - - : [size] "+m" (size), [old_rbx] "=m" (old_rbx) +# endif + : [size] "+m" (size) +# ifdef BOUNDED_RBX + , [old_rbx] "=m" (old_rbx) +# endif : [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), [dst_step] "m" (dst_step), [src_step] "m" (src_step), [sum_step] "m" (sum_step) : "rsi", "rdi", "edx", "ecx", "eax", "memory", "cc" +# ifndef BOUNDED_RBX + , "rbx" +# endif ); }

@@ -252,8 +274,9 @@ static void MIX_AREAS_24(unsigned int size, volatile signed int *sum, size_t dst_step, size_t src_step, size_t sum_step) { +# ifdef BOUNDED_RBX unsigned long long old_rbx; - +# endif /* * RSI - src * RDI - dst @@ -264,8 +287,9 @@ static void MIX_AREAS_24(unsigned int size, */ __asm__ __volatile__ ( "\n" - +# ifdef BOUNDED_RBX "\tmovq %%rbx, %[old_rbx]\n" +# endif /* * initialization, load RSI, RDI, RBX registers */ @@ -337,12 +361,23 @@ static void MIX_AREAS_24(unsigned int size, "\tjnz 1b\n"

Note: it has not been tested but I can not figure out it could break something because the compiler is now aware that ebx is clobbered. Curiously, it seems it significantly reduces the size of the code.

frederic.recoules＠univ-grenoble-alpes.fr

9:10 p.m.

New subject: [PATCH 6/6 V2] [pcm_dmix assembly] (minor) give more freedom over the constraints

From: Frédéric Recoules frederic.recoules@orange.fr

Signed-off-by: Frédéric Recoules frederic.recoules@orange.fr --- src/pcm/pcm_dmix_i386.h | 30 +++++++++++++++--------------- src/pcm/pcm_dmix_x86_64.h | 18 +++++++++--------- 2 files changed, 24 insertions(+), 24 deletions(-)

diff --git a/src/pcm/pcm_dmix_i386.h b/src/pcm/pcm_dmix_i386.h index b28ae787..65ed6d43 100644 --- a/src/pcm/pcm_dmix_i386.h +++ b/src/pcm/pcm_dmix_i386.h @@ -149,13 +149,13 @@ static void MIX_AREAS_16(unsigned int size, # ifdef BOUNDED_EBX "\tmovl %[old_ebx], %%ebx\n" /* ebx is GOT pointer (-fPIC) */ # endif - : [size] "+m" (size) + : [size] "+&rm" (size) # ifdef BOUNDED_EBX , [old_ebx] "=m" (old_ebx) # endif : [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), - [dst_step] "m" (dst_step), [src_step] "m" (src_step), - [sum_step] "m" (sum_step) + [dst_step] "im" (dst_step), [src_step] "im" (src_step), + [sum_step] "im" (sum_step) : "esi", "edi", "edx", "ecx", "eax", "memory", "cc" # ifndef BOUNDED_EBX , "ebx" @@ -249,13 +249,13 @@ static void MIX_AREAS_16_MMX(unsigned int size, # ifdef BOUNDED_EBX "\tmovl %[old_ebx], %%ebx\n" /* ebx is GOT pointer (-fPIC) */ # endif - : [size] "+m" (size) + : [size] "+&rm" (size) # ifdef BOUNDED_EBX , [old_ebx] "=m" (old_ebx) # endif : [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), - [dst_step] "m" (dst_step), [src_step] "m" (src_step), - [sum_step] "m" (sum_step) + [dst_step] "im" (dst_step), [src_step] "im" (src_step), + [sum_step] "im" (sum_step) : "esi", "edi", "edx", "ecx", "eax", "memory", "cc" # ifndef BOUNDED_EBX , "ebx" @@ -377,13 +377,13 @@ static void MIX_AREAS_32(unsigned int size, # ifdef BOUNDED_EBX "\tmovl %[old_ebx], %%ebx\n" /* ebx is GOT pointer (-fPIC) */ # endif - : [size] "+m" (size) + : [size] "+&rm" (size) # ifdef BOUNDED_EBX , [old_ebx] "=m" (old_ebx) # endif : [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), - [dst_step] "m" (dst_step), [src_step] "m" (src_step), - [sum_step] "m" (sum_step) + [dst_step] "im" (dst_step), [src_step] "im" (src_step), + [sum_step] "im" (sum_step) : "esi", "edi", "edx", "ecx", "eax", "memory", "cc" # ifndef BOUNDED_EBX , "ebx" @@ -492,13 +492,13 @@ static void MIX_AREAS_24(unsigned int size, # ifdef BOUNDED_EBX "\tmovl %[old_ebx], %%ebx\n" /* ebx is GOT pointer (-fPIC) */ # endif - : [size] "+m" (size) + : [size] "+&rm" (size) # ifdef BOUNDED_EBX , [old_ebx] "=m" (old_ebx) # endif : [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), - [dst_step] "m" (dst_step), [src_step] "m" (src_step), - [sum_step] "m" (sum_step) + [dst_step] "im" (dst_step), [src_step] "im" (src_step), + [sum_step] "im" (sum_step) : "esi", "edi", "edx", "ecx", "eax", "memory", "cc" # ifndef BOUNDED_EBX , "ebx" @@ -600,13 +600,13 @@ static void MIX_AREAS_24_CMOV(unsigned int size, # ifdef BOUNDED_EBX "\tmovl %[old_ebx], %%ebx\n" /* ebx is GOT pointer (-fPIC) */ # endif - : [size] "+m" (size) + : [size] "+&rm" (size) # ifdef BOUNDED_EBX , [old_ebx] "=m" (old_ebx) # endif : [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), - [dst_step] "m" (dst_step), [src_step] "m" (src_step), - [sum_step] "m" (sum_step) + [dst_step] "im" (dst_step), [src_step] "im" (src_step), + [sum_step] "im" (sum_step) : "esi", "edi", "edx", "ecx", "eax", "memory", "cc" # ifndef BOUNDED_EBX , "ebx" diff --git a/src/pcm/pcm_dmix_x86_64.h b/src/pcm/pcm_dmix_x86_64.h index f543a134..4848ff80 100644 --- a/src/pcm/pcm_dmix_x86_64.h +++ b/src/pcm/pcm_dmix_x86_64.h @@ -122,13 +122,13 @@ static void MIX_AREAS_16(unsigned int size, # ifdef BOUNDED_RBX "\tmovq %[old_rbx], %%rbx\n" # endif - : [size] "+m" (size) + : [size] "+&rm" (size) # ifdef BOUNDED_RBX , [old_rbx] "=m" (old_rbx) # endif : [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), - [dst_step] "m" (dst_step), [src_step] "m" (src_step), - [sum_step] "m" (sum_step) + [dst_step] "im" (dst_step), [src_step] "im" (src_step), + [sum_step] "im" (sum_step) : "rsi", "rdi", "edx", "ecx", "eax", "memory", "cc" # ifndef BOUNDED_RBX , "rbx" @@ -252,13 +252,13 @@ static void MIX_AREAS_32(unsigned int size, # ifdef BOUNDED_RBX "\tmovq %[old_rbx], %%rbx\n" # endif - : [size] "+m" (size) + : [size] "+&rm" (size) # ifdef BOUNDED_RBX , [old_rbx] "=m" (old_rbx) # endif : [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), - [dst_step] "m" (dst_step), [src_step] "m" (src_step), - [sum_step] "m" (sum_step) + [dst_step] "im" (dst_step), [src_step] "im" (src_step), + [sum_step] "im" (sum_step) : "rsi", "rdi", "edx", "ecx", "eax", "memory", "cc" # ifndef BOUNDED_RBX , "rbx" @@ -364,13 +364,13 @@ static void MIX_AREAS_24(unsigned int size, # ifdef BOUNDED_RBX "\tmovq %[old_rbx], %%rbx\n" # endif - : [size] "+m" (size) + : [size] "+&rm" (size) # ifdef BOUNDED_RBX , [old_rbx] "=m" (old_rbx) # endif : [dst] "m" (dst), [src] "m" (src), [sum] "m" (sum), - [dst_step] "m" (dst_step), [src_step] "m" (src_step), - [sum_step] "m" (sum_step) + [dst_step] "im" (dst_step), [src_step] "im" (src_step), + [sum_step] "im" (sum_step) : "rsi", "rdi", "edx", "ecx", "eax", "memory", "cc" # ifndef BOUNDED_RBX , "rbx" -- 2.17.1

1715

Age (days ago)

1715

Last active (days ago)

List overview

Download

5 comments

1 participants

tags (0)

participants (1)

frederic.recoules＠univ-grenoble-alpes.fr