[Mesa-dev] [PATCH v2 01/10] tgsi: add MEMBAR opcode to handle memoryBarrier* GLSL intrinsics

[Ilia Mirkin]

This is designed to map the GLSL intrinsics. Should one have a desire
to also support d3d11, one could figure out what the overlap is and
rejigger the arguments so that both sets of desires are expressible. I
glanced at SM5 sync before doing this, and TBH I couldn't really make
sense of it:

https://msdn.microsoft.com/en-us/library/windows/desktop/hh447241(v=vs.85).aspx

If you can interpret it and make concrete recommendations for change,
I'm happy to accommodate.

On Tue, Jan 19, 2016 at 12:04 PM, Roland Scheidegger <sroland at vmware.com> wrote:
> I am actually wondering how well that would work for d3d11.
> d3d11 just has AllMemoryBarrier, DeviceMemoryBarrier plus
> GroupMemoryBarrier - and for each of them also a "WithGroupSync"
> version. Hmm.
>
> Roland
>
> Am 19.01.2016 um 03:30 schrieb Ilia Mirkin:
>> Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
>> Reviewed-by: Marek Olšák <marek.olsak at amd.com> (v1)
>>
>> v1 -> v2: add defines for the various bits
>> ---
>>  src/gallium/auxiliary/tgsi/tgsi_info.c     |  2 +-
>>  src/gallium/docs/source/tgsi.rst           | 17 +++++++++++++++++
>>  src/gallium/include/pipe/p_shader_tokens.h |  7 ++++++-
>>  3 files changed, 24 insertions(+), 2 deletions(-)
>>
>> diff --git a/src/gallium/auxiliary/tgsi/tgsi_info.c b/src/gallium/auxiliary/tgsi/tgsi_info.c
>> index b270dd7..46b296f 100644
>> --- a/src/gallium/auxiliary/tgsi/tgsi_info.c
>> +++ b/src/gallium/auxiliary/tgsi/tgsi_info.c
>> @@ -149,7 +149,7 @@ static const struct tgsi_opcode_info opcode_info[TGSI_OPCODE_LAST] =
>>     { 1, 2, 0, 0, 0, 0, 0, COMP, "FSGE", TGSI_OPCODE_FSGE },
>>     { 1, 2, 0, 0, 0, 0, 0, COMP, "FSLT", TGSI_OPCODE_FSLT },
>>     { 1, 2, 0, 0, 0, 0, 0, COMP, "FSNE", TGSI_OPCODE_FSNE },
>> -   { 0, 1, 0, 0, 0, 0, 1, NONE, "", 112 },      /* removed */
>> +   { 0, 1, 0, 0, 0, 0, 0, OTHR, "MEMBAR", TGSI_OPCODE_MEMBAR },
>>     { 0, 1, 0, 0, 0, 0, 0, NONE, "CALLNZ", TGSI_OPCODE_CALLNZ },
>>     { 0, 1, 0, 0, 0, 0, 0, NONE, "", 114 },     /* removed */
>>     { 0, 1, 0, 0, 0, 0, 0, NONE, "BREAKC", TGSI_OPCODE_BREAKC },
>> diff --git a/src/gallium/docs/source/tgsi.rst b/src/gallium/docs/source/tgsi.rst
>> index 7810a3e..489cbb0 100644
>> --- a/src/gallium/docs/source/tgsi.rst
>> +++ b/src/gallium/docs/source/tgsi.rst
>> @@ -2372,6 +2372,23 @@ programs.
>>    the program.  Results are unspecified if any of the remaining
>>    threads terminates or never reaches an executed BARRIER instruction.
>>
>> +.. opcode:: MEMBAR - Memory barrier
>> +
>> +  ``MEMBAR type``
>> +
>> +  This opcode waits for the completion of all memory accesses based on
>> +  the type passed in. The type is an immediate bitfield with the following
>> +  meaning:
>> +
>> +  Bit 0: Shader storage buffers
>> +  Bit 1: Atomic buffers
>> +  Bit 2: Images
>> +  Bit 3: Shared memory
>> +  Bit 4: Thread group
>> +
>> +  These may be passed in in any combination. An implementation is free to not
>> +  distinguish between these as it sees fit. However these map to all the
>> +  possibilities made available by GLSL.
>>
>>  .. _atomopcodes:
>>
>> diff --git a/src/gallium/include/pipe/p_shader_tokens.h b/src/gallium/include/pipe/p_shader_tokens.h
>> index f300207..6539017 100644
>> --- a/src/gallium/include/pipe/p_shader_tokens.h
>> +++ b/src/gallium/include/pipe/p_shader_tokens.h
>> @@ -420,7 +420,7 @@ struct tgsi_property_data {
>>  #define TGSI_OPCODE_FSLT                110
>>  #define TGSI_OPCODE_FSNE                111
>>
>> -                                /* gap */
>> +#define TGSI_OPCODE_MEMBAR              112
>>  #define TGSI_OPCODE_CALLNZ              113
>>                                  /* gap */
>>  #define TGSI_OPCODE_BREAKC              115
>> @@ -744,6 +744,11 @@ struct tgsi_instruction_memory
>>     unsigned Padding   : 29;
>>  };
>>
>> +#define TGSI_MEMBAR_SHADER_BUFFER (1 << 0)
>> +#define TGSI_MEMBAR_ATOMIC_BUFFER (1 << 1)
>> +#define TGSI_MEMBAR_SHADER_IMAGE  (1 << 2)
>> +#define TGSI_MEMBAR_SHARED        (1 << 3)
>> +#define TGSI_MEMBAR_THREAD_GROUP  (1 << 4)
>>
>>  #ifdef __cplusplus
>>  }
>>
>