libdrm: Move intel_atomic.h to libdrm core for sharing. (V3)
PATCH 1/3 libdrm: Move intel_atomic.h to libdrm core for sharing. (V3)
intel_atomic.h includes very usefull atomic operations for lock free parrallel access of variables. Moving these to core libdrm for code sharing with radeon.
V2: Fix remaining references to intel_atomic.h and libdrm-intel.
V3: Remove useless wrapper header intel_atomic.h
--- Makefile.am | 3 + configure.ac | 2 +- intel/Makefile.am | 1 - intel/intel_atomic.h | 93 ---------------------------------------------- intel/intel_bufmgr_gem.c | 2 +- xf86atomic.h | 93 ++++++++++++++++++++++++++++++++++++++++++++++ 6 files changed, 98 insertions(+), 96 deletions(-) delete mode 100644 intel/intel_atomic.h create mode 100644 xf86atomic.h
Makefile.am +3 -0
@@ -61,6 +61,9 @@ libdrm_la_SOURCES = \ libdrmincludedir = ${includedir} libdrminclude_HEADERS = xf86drm.h xf86drmMode.h +noinst_HEADERS = xf86atomic.h + + EXTRA_DIST = libdrm.pc.in include/drm/* copy-headers :
configure.ac +1 -1
@@ -198,7 +198,7 @@ if test "x$INTEL" != "xno"; then ]) if test "x$drm_cv_atomic_primitives" = xIntel; then - AC_DEFINE(HAVE_INTEL_ATOMIC_PRIMITIVES, 1, + AC_DEFINE(HAVE_LIBDRM_ATOMIC_PRIMITIVES, 1, [Enable if your compiler supports the Intel __sync_* atomic primitives]) fi if test "x$drm_cv_atomic_primitives" = "xlibatomic-ops"; then
intel/Makefile.am +0 -1
@@ -35,7 +35,6 @@ libdrm_intel_la_LDFLAGS = -version-number 1:0:0 -no-undefined libdrm_intel_la_LIBADD = ../libdrm.la @PTHREADSTUBS_LIBS@ @CLOCK_LIB@ libdrm_intel_la_SOURCES = \ - intel_atomic.h \ intel_bufmgr.c \ intel_bufmgr_priv.h \ intel_bufmgr_fake.c \
intel/intel_atomic.h deleted +0 -93
93 lines not on this page · raw message
intel/intel_bufmgr_gem.c +1 -1
@@ -39,6 +39,7 @@#endif #include <xf86drm.h> +#include <xf86atomic.h> #include <fcntl.h> #include <stdio.h> #include <stdlib.h>
@@ -53,7 +54,6 @@#include "errno.h" #include "libdrm_lists.h" -#include "intel_atomic.h" #include "intel_bufmgr.h" #include "intel_bufmgr_priv.h" #include "intel_chipset.h"
xf86atomic.h new file +93 -0
93 lines not on this page · raw message
Signature · 1 line
1.6.3.3
------------------------------------------------------------------------------
Download Intel® Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs proactively, and fine-tune applications for parallel performance. See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--
_______________________________________________
Dri-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/dri-devel
PATCH 2/3 libdrm_radeon: Optimize cs_gem_reloc to do less looping. (V2)
bo->referenced_in_cs is checked if bo is already in cs. Adding and removing reference in bo is done with atomic operations to allow parallel access to a bo from multiple contexts.
cs->id generation code quarentees there is not duplicated ids which limits number of cs->ids to 32. If there is more cs objects rest will get id 0.
V2:
- Fix configure to check for atomics operations if libdrm_radeon is selected and libdrm_intel is not.
- Make atomic operations private to libdrm_radeon.
- Add warning messages if disabling automaticaly selected libdrm_(radeon|intel)
This optimization decreases cs_write_reloc share of torcs profiling from 4.3% to 2.6%.
--- configure.ac | 27 ++++++++--- radeon/radeon_bo_gem.c | 9 ++++ radeon/radeon_bo_gem.h | 1 + radeon/radeon_cs.c | 6 ++ radeon/radeon_cs.h | 2 +- radeon/radeon_cs_gem.c | 123 ++++++++++++++++++++++++++++++++++++------------ radeon/radeon_cs_int.h | 1 + xf86atomic.h | 6 ++ 8 files changed, 136 insertions(+), 39 deletions(-)
configure.ac +20 -7
@@ -56,8 +56,8 @@ AC_ARG_ENABLE(intel, AC_ARG_ENABLE(radeon, AS_HELP_STRING([--disable-radeon], - [Enable support for radeon's KMS API (default: enabled)]), - [RADEON=$enableval], [RADEON=yes]) + [Enable support for radeon's KMS API (default: auto)]), + [RADEON=$enableval], [RADEON=auto]) AC_ARG_ENABLE(vmwgfx-experimental-api, AS_HELP_STRING([--enable-vmwgfx-experimental-api],
@@ -173,7 +173,7 @@ if test "x$HAVE_LIBUDEV" = xyes; then fi AM_CONDITIONAL(HAVE_LIBUDEV, [test "x$HAVE_LIBUDEV" = xyes]) -if test "x$INTEL" != "xno"; then +if test "x$INTEL" != "xno" -o "x$RADEON" != "xno"; then # Check for atomic intrinsics AC_CACHE_CHECK([for native atomic primitives], drm_cv_atomic_primitives, [
@@ -206,13 +206,26 @@ if test "x$INTEL" != "xno"; then fi if test "x$drm_cv_atomic_primitives" = "xnone"; then - if test "x$INTEL" != "xauto"; then - AC_MSG_ERROR([libdrm_intel depends upon atomic operations, which were not found for your compiler/cpu. Try compiling with -march=native, or install the libatomics-op-dev package, or, failing both of those, disable support for Intel GPUs by passing --disable-intel to ./configure]) - else + if test "x$INTEL" != "xauto"; then + AC_MSG_ERROR([libdrm_intel depends upon atomic operations, which were not found for your compiler/cpu. Try compiling with -march=native, or install the libatomics-op-dev package, or, failing both of those, disable support for Intel GPUs by passing --disable-intel to ./configure]) + else + AC_MSG_WARN([Disabling libdrm_intel. It depends on atomic operations, which were not found for your compiler/cpu. Try compiling with -march=native, or install the libatomics-op-dev package.]) INTEL=no fi + if test "x$RADEON" != "xauto"; then + AC_MSG_ERROR([libdrm_radeon depends upon atomic operations, which were not found for your compiler/cpu. Try compiling with -march=native, or install the libatomics-op-dev package, or, failing both of those, disable support for Radeon support by passing --disable-radeon to ./configure]) + else + AC_MSG_WARN([Disabling libdrm_radeon. It depends on atomic operations, which were not found for your compiler/cpu. Try compiling with -march=native, or install the libatomics-op-dev package.]) + RADEON=no + fi + else - INTEL=yes + if test "x$INTEL" != "xno"; then + INTEL=yes + fi + if test "x$RADEON" != "xno"; then + RADEON=yes + fi fi fi
radeon/radeon_bo_gem.c +9 -0
@@ -39,6 +39,7 @@#include <sys/mman.h> #include <errno.h> #include "xf86drm.h" +#include "xf86atomic.h" #include "drm.h" #include "radeon_drm.h" #include "radeon_bo.h"
@@ -49,6 +50,7 @@ struct radeon_bo_gem {struct radeon_bo_int base;uint32_t name;int map_count; +atomic_t reloc_in_cs;void *priv_ptr; };
@@ -80,6 +82,7 @@ static struct radeon_bo *bo_open(struct radeon_bo_manager *bom, bo->base.domains = domains; bo->base.flags = flags; bo->base.ptr =NULL ; +atomic_set (&bo->reloc_in_cs,0 ); bo->map_count =0 ;if (handle) {struct drm_gem_open open_arg;
@@ -309,6 +312,12 @@ uint32_t radeon_gem_name_bo(struct radeon_bo *bo)return bo_gem->name; } +void *radeon_gem_get_reloc_in_cs (struct radeon_bo *bo) +{ +struct radeon_bo_gem *bo_gem = (struct radeon_bo_gem *)bo; +return &bo_gem->reloc_in_cs; +} +int radeon_gem_get_kernel_name (struct radeon_bo *bo,uint32_t *name) {struct radeon_bo_int *boi = (struct radeon_bo_int *)bo;
radeon/radeon_bo_gem.h +1 -0
@@ -38,6 +38,7 @@ struct radeon_bo_manager *radeon_bo_manager_gem_ctor(int fd);void radeon_bo_manager_gem_dtor (struct radeon_bo_manager *bom);uint32_t radeon_gem_name_bo (struct radeon_bo *bo); +void *radeon_gem_get_reloc_in_cs (struct radeon_bo *bo);int radeon_gem_set_domain (struct radeon_bo *bo,uint32_t read_domains,uint32_t write_domain);int radeon_gem_get_kernel_name (struct radeon_bo *bo,uint32_t *name);#endif
radeon/radeon_cs.c +6 -0
@@ -88,3 +88,9 @@ void radeon_cs_space_set_flush(struct radeon_cs *cs, void (*fn)(void *), void *d csi->space_flush_fn = fn; csi->space_flush_data = data; } + +uint32_t radeon_cs_get_id (struct radeon_cs *cs) +{ +struct radeon_cs_int *csi = (struct radeon_cs_int *)cs; +return csi->id; +}
radeon/radeon_cs.h +1 -1
@@ -85,7 +85,7 @@ extern int radeon_cs_write_reloc(struct radeon_cs *cs,uint32_t read_domain, uint32_t write_domain, uint32_t flags); - +extern uint32_t radeon_cs_get_id (struct radeon_cs *cs);/* * add a persistent BO to the list * a persistent BO is one that will be referenced across flushes,
radeon/radeon_cs_gem.c +92 -31
178 lines not on this page · raw message
radeon/radeon_cs_int.h +1 -0
@@ -28,6 +28,7 @@ struct radeon_cs_int {int bo_count;void (*space_flush_fn)(void *);void *space_flush_data; +uint32_t id; };/* cs functions */
xf86atomic.h +6 -0
@@ -50,6 +50,8 @@ typedef struct {# define atomic_set (x, val) ((x)->atomic = (val))# define atomic_inc (x) ((void) __sync_fetch_and_add (&(x)->atomic, 1))# define atomic_dec_and_test (x) (__sync_fetch_and_add (&(x)->atomic, -1) == 1) +# define atomic_add (x, v) ((void) __sync_add_and_fetch(&(x)->atomic, (v))) +# define atomic_dec (x, v) ((void) __sync_sub_and_fetch(&(x)->atomic, (v)))# define atomic_cmpxchg (x, oldv, newv) __sync_val_compare_and_swap (&(x)->atomic, oldv, newv)#endif
@@ -66,6 +68,8 @@ typedef struct {# define atomic_read (x) AO_load_full(&(x)->atomic)# define atomic_set (x, val) AO_store_full(&(x)->atomic, (val))# define atomic_inc (x) ((void) AO_fetch_and_add1_full(&(x)->atomic)) +# define atomic_add (x, v) ((void) AO_fetch_and_add_full(&(x)->atomic, (v))) +# define atomic_dec (x, v) ((void) AO_fetch_and_add_full(&(x)->atomic, -(v)))# define atomic_dec_and_test (x) (AO_fetch_and_sub1_full(&(x)->atomic) == 1)# define atomic_cmpxchg (x, oldv, newv) AO_compare_and_swap_full(&(x)->atomic, oldv, newv)
@@ -82,6 +86,8 @@ typedef struct { uint_t atomic; } atomic_t;# define atomic_set (x, val) ((x)->atomic = (uint_t)(val))# define atomic_inc (x) (atomic_inc_uint (&(x)->atomic))# define atomic_dec_and_test (x) (atomic_dec_uint_nv(&(x)->atomic) == 1) +# define atomic_add (x, v) (atomic_add_uint(&(x)->atomic, (v))) +# define atomic_dec (x, v) (atomic_dec_uint(&(x)->atomic, (v)))# define atomic_cmpxchg (x, oldv, newv) atomic_cas_uint (&(x)->atomic, oldv, newv)#endif
Signature · 10 lines
1.6.3.3 ------------------------------------------------------------------------------ Download Intel® Parallel Studio Eval Try the new software tools for yourself. Speed compiling, find bugs proactively, and fine-tune applications for parallel performance. See why Intel Parallel Studio got high marks during beta. http://p.sf.net/sfu/intel-sw-dev --
PATCH 3/3 libdrm: Move all noninstalled headers to noinst_HEADERS.
Cleanup make system so that all noninstalled headers are put to noinst_HEADERS. This quarentees that header will be present in tar ball but not installed with make install.
--- Makefile.am | 6 +++--- intel/Makefile.am | 10 ++++++---- nouveau/Makefile.am | 4 +++- 3 files changed, 12 insertions(+), 8 deletions(-)
Makefile.am +3 -3
@@ -55,13 +55,13 @@ libdrm_la_SOURCES = \ xf86drmHash.c \ xf86drmRandom.c \ xf86drmSL.c \ - xf86drmMode.c \ - libdrm_lists.h + xf86drmMode.c libdrmincludedir = ${includedir} libdrminclude_HEADERS = xf86drm.h xf86drmMode.h -noinst_HEADERS = xf86atomic.h +noinst_HEADERS = xf86atomic.h \ + libdrm_lists.h EXTRA_DIST = libdrm.pc.in include/drm/*
intel/Makefile.am +6 -4
@@ -36,14 +36,16 @@ libdrm_intel_la_LIBADD = ../libdrm.la @PTHREADSTUBS_LIBS@ @CLOCK_LIB@ libdrm_intel_la_SOURCES = \ intel_bufmgr.c \ - intel_bufmgr_priv.h \ intel_bufmgr_fake.c \ intel_bufmgr_gem.c \ - intel_chipset.h \ - mm.c \ - mm.h + mm.c libdrm_intelincludedir = ${includedir} libdrm_intelinclude_HEADERS = intel_bufmgr.h +noinst_HEADERS = \ + intel_bufmgr_priv.h \ + intel_chipset.h \ + mm.h + pkgconfig_DATA = libdrm_intel.pc
nouveau/Makefile.am +3 -1
@@ -18,7 +18,6 @@ libdrm_nouveau_la_SOURCES = \ nouveau_notifier.c \ nouveau_bo.c \ nouveau_resource.c \ - nouveau_private.h \ nouveau_reloc.c libdrm_nouveaucommonincludedir = ${includedir}/nouveau
@@ -38,5 +37,8 @@ libdrm_nouveauincludedir = ${includedir}/drm libdrm_nouveauinclude_HEADERS = \ nouveau_drmif.h +noinst_HEADERS = \ + nouveau_private.h + pkgconfigdir = @pkgconfigdir@ pkgconfig_DATA = libdrm_nouveau.pc
Signature · 10 lines
1.6.3.3 ------------------------------------------------------------------------------ Download Intel® Parallel Studio Eval Try the new software tools for yourself. Speed compiling, find bugs proactively, and fine-tune applications for parallel performance. See why Intel Parallel Studio got high marks during beta. http://p.sf.net/sfu/intel-sw-dev --
A message that is not in the archive (sent off-list or not yet received): 1268398235-21806-3-git-send-email-suokkos-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org
Re: [Intel-gfx] [PATCH 3/3] libdrm: Move all noninstalled headers to noinst_HEADERS.
Quoted 77 lines · Pauli Nieminen
Cleanup make system so that all noninstalled headers are put to
noinst_HEADERS. This quarentees that header will be present in tar ball but not installed with make install.Signed-off-by Pauli Nieminen <[email protected]>---
Makefile.am | 6 +++---
intel/Makefile.am | 10 ++++++----
nouveau/Makefile.am | 4 +++-3 files changed, 12 insertions(+), 8 deletions(-)Makefile.am +1
-2@@ -55,13 +55,13 @@ libdrm_la_SOURCES = \ xf86drmHash.c \ xf86drmRandom.c \ xf86drmSL.c \ - xf86drmMode.c \ - libdrm_lists.h + xf86drmMode.c libdrmincludedir = ${includedir}libdrminclude_HEADERS =
xf86drm.hxf86drmMode.hMakefile.am +2
-1-noinst_HEADERS = xf86atomic.h +noinst_HEADERS = xf86atomic.h \ + libdrm_lists.hEXTRA_DIST = libdrm.pc.in
include/drm/*intel/Makefile.am +1
-4@@ -36,14 +36,16 @@ libdrm_intel_la_LIBADD = ../libdrm.la @PTHREADSTUBS_LIBS@ @CLOCK_LIB@ libdrm_intel_la_SOURCES = \ intel_bufmgr.c \ - intel_bufmgr_priv.h \ intel_bufmgr_fake.c \ intel_bufmgr_gem.c \ - intel_chipset.h \ - mm.c \ - mm.h + mm.c libdrm_intelincludedir = ${includedir}libdrm_intelinclude_HEADERS =
intel_bufmgr.hintel/Makefile.am +5
-0+noinst_HEADERS = \ + intel_bufmgr_priv.h \ + intel_chipset.h \ + mm.h +pkgconfig_DATA = libdrm_intel.pc
nouveau/Makefile.am +0
-1@@ -18,7 +18,6 @@ libdrm_nouveau_la_SOURCES = \ nouveau_notifier.c \ nouveau_bo.c \ nouveau_resource.c \ - nouveau_private.h \nouveau_reloc.c
libdrm_nouveaucommonincludedir = ${includedir}/nouveau
nouveau/Makefile.am +0
-0@@ -38,5 +37,8 @@ libdrm_nouveauincludedir = ${includedir}/drm libdrm_nouveauinclude_HEADERS = \nouveau_drmif.h
nouveau/Makefile.am +3
-0+noinst_HEADERS = \ + nouveau_private.h +
Quoted 2 lines
pkgconfigdir = @pkgconfigdir@
pkgconfig_DATA = libdrm_nouveau.pc
I don't think this is the best strategy. Having the headers under _SOURCES already means they'll get distributed. The only reason for the separate _HEADERS declarations is if you need to install them.
http://www.gnu.org/software/automake/manual/automake.html#Headers
--
Dan
_______________________________________________
Nouveau mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/nouveau
Re: [PATCH 2/3] libdrm_radeon: Optimize cs_gem_reloc to do less looping. (V2)
In reply to Pauli NieminenQuoted 12 lines · Pauli Nieminen
bo->referenced_in_csis checked if bo is already in cs. Adding and removing reference in bo is done with atomic operations to allow parallel access to a bo from multiple contexts.
cs->idgeneration code quarentees there is not duplicated ids which limits number ofcs->idsto 32. If there is more cs objects rest will get id 0.V2:
- Fix configure to check for atomics operations iflibdrm_radeonis selected andlibdrm_intelis not.
- Make atomic operations private tolibdrm_radeon.
- Add warning messages if disabling automaticaly selectedlibdrm_(radeon|intel)
Quoted 4 lines
This optimization decreases
cs_write_relocshare of torcs profiling from 4.3% to 2.6%.Signed-off-by Pauli Nieminen <[email protected]>
Signature · 14 lines
Earthling Michel Dänzer | http://www.vmware.com Libre software enthusiast | Debian, X and DRI developer ------------------------------------------------------------------------------ Download Intel® Parallel Studio Eval Try the new software tools for yourself. Speed compiling, find bugs proactively, and fine-tune applications for parallel performance. See why Intel Parallel Studio got high marks during beta. http://p.sf.net/sfu/intel-sw-dev -- _______________________________________________ Dri-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/dri-devel