Skip to contentSkip to conversation

libdrm: Move intel_atomic.h to libdrm core for sharing.

PATCH In review Pauli Nieminen 6 participants 9 messages Mar 10, 2010 – Mar 12, 2010 Kbuild

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 libdrm: Move intel_atomic.h to libdrm core for sharing.

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.

---
 Makefile.am          |    2 +-
 configure.ac         |    2 +-
 intel/intel_atomic.h |   56 +-----------------------------
 xf86atomic.h         |   93 ++++++++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 96 insertions(+), 57 deletions(-)
 create mode 100644 xf86atomic.h
Makefile.am +1 -1
index ee3ccc7..295121f 100644
@@ -59,7 +59,7 @@ libdrm_la_SOURCES =				\
 	libdrm_lists.h
 
 libdrmincludedir = ${includedir}
-libdrminclude_HEADERS = xf86drm.h xf86drmMode.h
+libdrminclude_HEADERS = xf86drm.h xf86drmMode.h xf86atomic.h
 
 EXTRA_DIST = libdrm.pc.in include/drm/*
 
configure.ac +1 -1
index aaa8efa..953a758 100644
@@ -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/intel_atomic.h +1 -55

61 lines not on this page · raw message

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

#1 lore raw

PATCH 2/2 libdrm_radeon: Optimize reloc writing to do less looping.

Bit has table will be first checked from BO if we can quarentee this BO is not in this cs already.

To quarentee that there is no other cs with same id number of CS that can have id is limited to 32. Adding and remocing reference in bo is done with atomic operations to allow parallel access to a bo from multiple contexts.

This optimization decreases cs_write_reloc share of torcs profiling from 4.3% to 2.6%.

---
 radeon/radeon_bo_gem.c |    1 +
 radeon/radeon_cs.c     |    6 ++
 radeon/radeon_cs.h     |    2 +-
 radeon/radeon_cs_gem.c |  133 ++++++++++++++++++++++++++++++++++++-----------
 radeon/radeon_cs_int.h |    1 +
 5 files changed, 111 insertions(+), 32 deletions(-)
radeon/radeon_bo_gem.c +1 -0
index bc8058d..1b33bdb 100644
@@ -80,6 +80,7 @@ static struct radeon_bo *bo_open(struct radeon_bo_manager *bom,
     bo->base.domains = domains;
     bo->base.flags = flags;
     bo->base.ptr = NULL;
+    bo->base.referenced_in_cs = 0;
     bo->map_count = 0;
     if (handle) {
         struct drm_gem_open open_arg;
radeon/radeon_cs.c +6 -0
index cc9be39..d0e922b 100644
@@ -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
index 49d5d9a..7f6ee68 100644
@@ -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 +102 -31

182 lines not on this page · raw message

radeon/radeon_cs_int.h +1 -0
index 8ba76bf..6cee574 100644
@@ -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 */
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
--
#2 lore raw

Re: [PATCH 2/2] libdrm_radeon: Optimize reloc writing to do less looping.

Quoted 11 lines · Pauli Nieminen

Bit has table will be first checked from BO if we can quarentee this BO is not in this cs already.

To quarentee that there is no other cs with same id number of CS that can have id is limited to 32. Adding and remocing reference in bo is done with atomic operations to allow parallel access to a bo from multiple contexts.

This optimization decreases cs_write_reloc share of torcs profiling from 4.3% to 2.6%.

[...]

Quoted 13 lines
radeon/radeon_cs_gem.c +5 -0
index 45a219c..83aabea 100644
@@ -68,6 +69,66 @@ struct cs_gem {
     struct radeon_bo_int        **relocs_bo;
 };
 
+
+#if !defined(__GNUC__) || __GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ < 2)
+/* no built in sync support in compiler define place holders */
+uint32_t __sync_add_and_fetch(uint32_t *a, uint32_t val)
+{

radeon/radeon_cs_gem.c struct cs_gem { · line 86

Quoted 10 lines
radeon/radeon_cs_gem.c +10 -0
+	*a += val;
+	return val;
+}
+
+uint32_t __sync_add_and_fetch(uint32_t *a, uint32_t val)
+{
+	*a -= val;
+	return val;
+}
+#endif

This doesn't look like it could build... presumably the latter should be called __sync_sub_and_fetch()?

Do these stand any chance of working properly in circumstances where atomicity is actually important though?

Signature · 14 lines
Earthling Michel Dänzer           |                http://www.vmware.com
Libre software enthusiast         |          Debian, X and DRI developer

------------------------------------------------------------------------------
Download Intel&#174; 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
#3 lore raw

Re: [PATCH 2/2] libdrm_radeon: Optimize reloc writing to do less looping.

2010/3/10 Michel Dänzer <[email protected]>:

Quoted 33 lines · Pauli Nieminen
Quoted 11 lines

Bit has table will be first checked from BO if we can quarentee this BO is not in this cs already.

To quarentee that there is no other cs with same id number of CS that can have id is limited to 32. Adding and remocing reference in bo is done with atomic operations to allow parallel access to a bo from multiple contexts.

This optimization decreases cs_write_reloc share of torcs profiling from 4.3% to 2.6%.

[...]

Quoted 18 lines
radeon/radeon_cs_gem.c +10 -0
index 45a219c..83aabea 100644
@@ -68,6 +69,66 @@ struct cs_gem {
      struct radeon_bo_int        **relocs_bo;
  };
 
+
+#if !defined(__GNUC__) || __GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ < 2)
+/* no built in sync support in compiler define place holders */
+uint32_t __sync_add_and_fetch(uint32_t *a, uint32_t val)
+{
+     *a += val;
+     return val;
+}
+
+uint32_t __sync_add_and_fetch(uint32_t *a, uint32_t val)
Quoted 9 lines

radeon/radeon_cs_gem.c struct cs_gem { · line 86

Quoted 5 lines
radeon/radeon_cs_gem.c +5 -0
+{
+     *a -= val;
+     return val;
+}
+#endif

This doesn't look like it could build... presumably the latter should be called __sync_sub_and_fetch()?

sorry .wrong patch coming from somewhere :/

Quoted 8 lines

Do these stand any chance of working properly in circumstances where atomicity is actually important though?

--
Earthling Michel Dänzer           |                http://www.vmware.com
Libre software enthusiast         |          Debian, X and DRI developer

------------------------------------------------------------------------------
Download Intel&#174; 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
--

#4 lore raw
In reply to Pauli Nieminen
Quoted 130 lines · Pauli Nieminen

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.

---
 Makefile.am          |    2 +-
 configure.ac         |    2 +-
 intel/intel_atomic.h |   56 +----------------------------- xf86atomic.h         |   93 ++++++++++++++++++++++++++++++++++++++++++++++++++

 4 files changed, 96 insertions(+), 57 deletions(-)

 create mode 100644 xf86atomic.h

Makefile.am +0 -0
index ee3ccc7..295121f 100644
@@ -59,7 +59,7 @@ libdrm_la_SOURCES =                           \

       libdrm_lists.h

 libdrmincludedir = ${includedir}

Makefile.am +1 -1
-libdrminclude_HEADERS = xf86drm.h xf86drmMode.h
+libdrminclude_HEADERS = xf86drm.h xf86drmMode.h xf86atomic.h

 EXTRA_DIST = libdrm.pc.in include/drm/*

configure.ac +0 -0
index aaa8efa..953a758 100644
@@ -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/intel_atomic.h +1 -55
index 12bb96b..dcb4ec8 100644
@@ -34,60 +34,6 @@
  #ifndef INTEL_ATOMICS_H
  #define INTEL_ATOMICS_H
 
-#ifdef HAVE_CONFIG_H
-#include "config.h"
-#endif
-
-#if HAVE_INTEL_ATOMIC_PRIMITIVES
-
-#define HAS_ATOMIC_OPS 1
-
-typedef struct {
-       int atomic;
-} atomic_t;
-
-# define atomic_read(x) ((x)->atomic)
-# 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_cmpxchg(x, oldv, newv) __sync_val_compare_and_swap (&(x)->atomic, oldv, newv)
-
-#endif
-
-#if HAVE_LIB_ATOMIC_OPS
-#include <atomic_ops.h>
-
-#define HAS_ATOMIC_OPS 1
-
-typedef struct {
-       AO_t atomic;
-} atomic_t;
-
-# 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_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)
-
-#endif
-
-#if defined(__sun) && !defined(HAS_ATOMIC_OPS)  /* Solaris & OpenSolaris */
-
-#include <sys/atomic.h>
-#define HAS_ATOMIC_OPS 1
-
-typedef struct { uint_t atomic; } atomic_t;
-
-# define atomic_read(x) (int) ((x)->atomic)
-# 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_cmpxchg(x, oldv, newv) atomic_cas_uint (&(x)->atomic, oldv, newv)
-
-#endif
-
-#if ! HAS_ATOMIC_OPS
-#error libdrm-intel requires atomic operations, please define them for your CPU/compiler.
-#endif
+#include <xf86atomic.h>
 
  #endif
xf86atomic.h new file +19 -0
new file mode 100644
index 0000000..627dcf2
@@ -0,0 +1,93 @@
+/*
+ * Copyright © 2009 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING

xf86atomic.h line 29

Quoted 10 lines
xf86atomic.h new file +10 -0
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Chris Wilson <[email protected]>
+ *
+ */
+
+/**
+ * @file intel_atomics.h

This needs to be updated.

Quoted 52 lines
xf86atomic.h new file +52 -0
+ *
+ * Private definitions for atomic operations
+ */
+
+#ifndef LIBDRM_ATOMICS_H
+#define LIBDRM_ATOMICS_H
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#if HAVE_LIBDRM_ATOMIC_PRIMITIVES
+
+#define HAS_ATOMIC_OPS 1
+
+typedef struct {
+       int atomic;
+} atomic_t;
+
+# define atomic_read(x) ((x)->atomic)
+# 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_cmpxchg(x, oldv, newv) __sync_val_compare_and_swap (&(x)->atomic, oldv, newv)
+
+#endif
+
+#if HAVE_LIB_ATOMIC_OPS
+#include <atomic_ops.h>
+
+#define HAS_ATOMIC_OPS 1
+
+typedef struct {
+       AO_t atomic;
+} atomic_t;
+
+# 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_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)
+
+#endif
+
+#if defined(__sun) && !defined(HAS_ATOMIC_OPS)  /* Solaris & OpenSolaris */
+
+#include <sys/atomic.h>
+#define HAS_ATOMIC_OPS 1
+
+typedef struct { uint_t atomic; } atomic_t;
+
+# define atomic_read(x) (int) ((x)->atomic)

xf86atomic.h line 91

Quoted 10 lines
xf86atomic.h new file +10 -0
+# 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_cmpxchg(x, oldv, newv) atomic_cas_uint (&(x)->atomic, oldv, newv)
+
+#endif
+
+#if ! HAS_ATOMIC_OPS
+#error libdrm-intel requires atomic operations, please define them for your CPU/compiler.
+#endif

This needs to be updated/fixed.

Quoted 4 lines
xf86atomic.h new file +2 -1
+
+#endif
--

1.6.3.3

Matt

------------------------------------------------------------------------------
Download Intel&#174; 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
--

#5 lore raw
Quoted 21 lines · Pauli Nieminen
Quoted 20 lines

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.

---
 Makefile.am          |    2 +-
 configure.ac         |    2 +-
 intel/intel_atomic.h |   56 +----------------------------- xf86atomic.h         |   93 ++++++++++++++++++++++++++++++++++++++++++++++++++

 4 files changed, 96 insertions(+), 57 deletions(-)

 create mode 100644 xf86atomic.h

Makefile.am +0 -0
index ee3ccc7..295121f 100644
@@ -59,7 +59,7 @@ libdrm_la_SOURCES =                           \

       libdrm_lists.h

Quoted 4 lines

Makefile.am libdrm_la_SOURCES = \ · line 62

Quoted 4 lines

 libdrmincludedir = ${includedir}

Makefile.am +1 -1
-libdrminclude_HEADERS = xf86drm.h xf86drmMode.h
+libdrminclude_HEADERS = xf86drm.h xf86drmMode.h xf86atomic.h

Should this really get installed? I'm not sure this should be public libdrm interface.

Cheers,
Julien

------------------------------------------------------------------------------
Download Intel&#174; 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
--

#6 lore raw
Quoted 29 lines · Julien Cristau
Quoted 21 lines · Pauli Nieminen
Quoted 20 lines

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.

---
 Makefile.am          |    2 +-
 configure.ac         |    2 +-
 intel/intel_atomic.h |   56 +----------------------------- xf86atomic.h         |   93 ++++++++++++++++++++++++++++++++++++++++++++++++++

 4 files changed, 96 insertions(+), 57 deletions(-)

 create mode 100644 xf86atomic.h

Makefile.am +0 -0
index ee3ccc7..295121f 100644
@@ -59,7 +59,7 @@ libdrm_la_SOURCES =                           \

       libdrm_lists.h

Quoted 4 lines

Makefile.am libdrm_la_SOURCES = \ · line 62

Quoted 4 lines

 libdrmincludedir = ${includedir}

Makefile.am +1 -1
-libdrminclude_HEADERS = xf86drm.h xf86drmMode.h
+libdrminclude_HEADERS = xf86drm.h xf86drmMode.h xf86atomic.h

Should this really get installed?  I'm not sure this should be public libdrm interface.

Quoted 3 lines

Cheers,
Julien

Inlte had it installed. And my patch adds it to radeon so that mesa will reference it indirectly tough radeon_bo_int.h. Maybe dependency in mesa to radeon_bo_int.h is bad idea in first place.

------------------------------------------------------------------------------
Download Intel&#174; 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
--

#7 lore raw
Quoted 9 lines · Pauli Nieminen
Quoted 6 lines · Julien Cristau

Should this really get installed? �I'm not sure this should be public libdrm interface.

Cheers,
Julien

Intel had it installed.

My bad, change it to noinst_HEADERS please.

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre

------------------------------------------------------------------------------
Download Intel&#174; 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

#8 lore raw
In reply to Pauli Nieminen
Quoted 3 lines · Pauli Nieminen

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.

s/xf86/libdrm/ but other than that, cool.

------------------------------------------------------------------------------
Download Intel&#174; 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