[Mesa-dev] [PATCH 8/8] r600/radeonsi: enable glsl/tgsi on-disk cache
Emil Velikov
emil.l.velikov at gmail.com
Wed Feb 22 17:49:19 UTC 2017
On 22 February 2017 at 14:57, Marc Dietrich <marvin24 at gmx.de> wrote:
> Am Mittwoch, 22. Februar 2017, 15:48:18 CET schrieb Emil Velikov:
>> On 22 February 2017 at 13:31, Marc Dietrich <marvin24 at gmx.de> wrote:
>> > Am Mittwoch, 22. Februar 2017, 04:45:46 CET schrieb Timothy Arceri:
>> >> For gpu generations that use LLVM we create a timestamp string
>> >> containing both the LLVM and Mesa build times, otherwise we just
>> >> use the Mesa build time.
>> >>
>> >> Reviewed-by: Marek Olšák <marek.olsak at amd.com>
>> >> Reviewed-by: Edward O'Callaghan <funfunctor at folklore1984.net>
>> >> ---
>> >>
>> >> src/gallium/drivers/radeon/r600_pipe_common.c | 43
>> >>
>> >> +++++++++++++++++++++++++++ src/gallium/drivers/radeon/r600_pipe_common.h
>> >> |
>> >>
>> >> 3 ++
>> >> 2 files changed, 46 insertions(+)
>> >>
>> >> diff --git a/src/gallium/drivers/radeon/r600_pipe_common.c
>> >> b/src/gallium/drivers/radeon/r600_pipe_common.c index 1781584..bae6d6f
>> >> 100644
>> >> --- a/src/gallium/drivers/radeon/r600_pipe_common.c
>> >> +++ b/src/gallium/drivers/radeon/r600_pipe_common.c
>> >> @@ -43,6 +43,10 @@
>> >>
>> >> #define HAVE_LLVM 0
>> >> #endif
>> >>
>> >> +#if HAVE_LLVM
>> >> +#include <llvm-c/TargetMachine.h>
>> >> +#endif
>> >> +
>> >>
>> >> #ifndef MESA_LLVM_VERSION_PATCH
>> >> #define MESA_LLVM_VERSION_PATCH 0
>> >> #endif
>> >>
>> >> @@ -779,6 +783,41 @@ static const char* r600_get_chip_name(struct
>> >> r600_common_screen *rscreen) }
>> >>
>> >> }
>> >>
>> >> +static void r600_disk_cache_create(struct r600_common_screen *rscreen)
>> >> +{
>> >> + uint32_t mesa_timestamp;
>> >> + if (disk_cache_get_function_timestamp(r600_disk_cache_create,
>> >> + &mesa_timestamp)) {
>> >> + char *timestamp_str;
>> >> + int res = -1;
>> >> + if (rscreen->chip_class < SI) {
>> >> + res = asprintf(×tamp_str,
>> >> "%u",mesa_timestamp);
>> >> + }
>> >> +#if HAVE_LLVM
>> >> + else {
>> >> + uint32_t llvm_timestamp;
>> >> + if
>> >> (disk_cache_get_function_timestamp(LLVMInitializeAMDGPUTargetInfo, +
>> >> &llvm_timestamp))
>> >> { + res = asprintf(×tamp_str, "%u_%u",
>> >> + mesa_timestamp,
>> >> llvm_timestamp); + }
>> >> + }
>> >> +#endif
>> >
>> > maybe this fails to link later in omx with r600 only compile.
>> >
>> > make[4]: Verzeichnis „/usr/src/dri-project/mesa/src/gallium/targets/omx“
>> > wird betreten
>> >
>> > CXXLD libomx_mesa.la
>> >
>> > ../../../../src/gallium/drivers/radeon/.libs/libradeon.a(r600_pipe_common.
>> > o): In function `r600_disk_cache_create':
>> > r600_pipe_common.c:(.text+0x2000): undefined reference to
>> > `LLVMInitializeAMDGPUTargetInfo'
>> > collect2: error: ld returned 1 exit status
>> > make[4]: *** [Makefile:791: libomx_mesa.la] Fehler 1
>> > make[4]: Verzeichnis „/usr/src/dri-project/mesa/src/gallium/targets/omx“
>> > wird verlassen
>>
>> That should not happen since
>> - Compile guard HAVE_LLVM is defined when --enable-llvm is set
>> (explicitly or not)
>> - Makefile/link guard HAVE_GALLIUM_LLVM is set when the --enable-llvm is
>>
>> Please make clean/git clean and report the output of make V=1 if the
>> final link still fails.
>
> still same. I added -lLLVMAMDGPUInfo to LLVM_{LDFLAGS|LIBS}. Now it compiles,
> but GL apps crash during context creation. I also tested the initial version
> posted and this worked just fine. I have llvm enabled.
>
Afaict there is no functional change that would trigger this issue -
either things were broken before (but went unnoticed) or something is
going bonkers here.
First things first:
- is this issue with multiple static LLVM libraries or multiple
shared LLVM libraries ?
- can you reproduce with simple shared LLVM library
Note that multiple shared LLVM libraries is not supported on our end :-\
As a follow-up:
- can you analyse the output of libvulkan_radeon.so linking vs
libomx.so - missing -lLLVMfoo, strange order etc.
- check if correct the component is not added to the LLVM_COMPONENT
list (in configure.ac)
- send us a patch that fixes the problem on your end, please ;-)
Thanks
Emil
More information about the mesa-dev
mailing list