[Mesa-dev] [PATCH 19/23] gallium/tgsi: add support for 64-bit integer immediates.

[Ilia Mirkin]

On Thu, Jun 9, 2016 at 2:41 PM, Marek Olšák <maraeo at gmail.com> wrote:
> On Thu, Jun 9, 2016 at 4:43 PM, Brian Paul <brianp at vmware.com> wrote:
>> On 06/08/2016 06:48 PM, Dave Airlie wrote:
>>>
>>> From: Dave Airlie <airlied at redhat.com>
>>>
>>> This adds support to TGSI for 64-bit integer immediates.
>>>
>>> Signed-off-by: Dave Airlie <airlied at redhat.com>
>>> ---
>>>   src/gallium/auxiliary/tgsi/tgsi_dump.c     |  14 ++
>>>   src/gallium/auxiliary/tgsi/tgsi_exec.c     | 244
>>> ++++++++++++++++++++++++++++-
>>>   src/gallium/auxiliary/tgsi/tgsi_parse.c    |   2 +
>>>   src/gallium/auxiliary/tgsi/tgsi_text.c     |  44 ++++++
>>>   src/gallium/auxiliary/tgsi/tgsi_ureg.c     |  45 +++++-
>>>   src/gallium/auxiliary/tgsi/tgsi_ureg.h     |  10 ++
>>>   src/gallium/include/pipe/p_shader_tokens.h |   2 +
>>>   7 files changed, 358 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump.c
>>> b/src/gallium/auxiliary/tgsi/tgsi_dump.c
>>> index d59b7ff..614bcb2 100644
>>> --- a/src/gallium/auxiliary/tgsi/tgsi_dump.c
>>> +++ b/src/gallium/auxiliary/tgsi/tgsi_dump.c
>>> @@ -254,6 +254,20 @@ dump_imm_data(struct tgsi_iterate_context *iter,
>>>            i++;
>>>            break;
>>>         }
>>> +      case TGSI_IMM_INT64: {
>>> +         union di d;
>>> +         d.i = data[i].Uint | (uint64_t)data[i+1].Uint << 32;
>>> +         UID( d.i );
>>> +         i++;
>>> +         break;
>>> +      }
>>> +      case TGSI_IMM_UINT64: {
>>> +         union di d;
>>> +         d.ui = data[i].Uint | (uint64_t)data[i+1].Uint << 32;
>>> +         UID( d.ui );
>>> +         i++;
>>> +         break;
>>> +      }
>>>         case TGSI_IMM_FLOAT32:
>>>            if (ctx->dump_float_as_hex)
>>>               HFLT( data[i].Float );
>>> diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c
>>> b/src/gallium/auxiliary/tgsi/tgsi_exec.c
>>> index 1457c06..c929475 100644
>>> --- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
>>> +++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
>>> @@ -77,6 +77,8 @@
>>>   union tgsi_double_channel {
>>>      double d[TGSI_QUAD_SIZE];
>>>      unsigned u[TGSI_QUAD_SIZE][2];
>>> +   uint64_t u64[TGSI_QUAD_SIZE];
>>> +   int64_t i64[TGSI_QUAD_SIZE];
>>>   };
>>>
>>>   struct tgsi_double_vector {
>>> @@ -692,11 +694,251 @@ micro_u2d(union tgsi_double_channel *dst,
>>>      dst->d[3] = (double)src->u[3];
>>>   }
>>>
>>> +static void
>>> +micro_i64abs(union tgsi_double_channel *dst,
>>> +             const union tgsi_double_channel *src)
>>> +{
>>> +   dst->i64[0] = src->i64[0] >= 0.0 ? src->i64[0] : -src->i64[0];
>>> +   dst->i64[1] = src->i64[1] >= 0.0 ? src->i64[1] : -src->i64[1];
>>> +   dst->i64[2] = src->i64[2] >= 0.0 ? src->i64[2] : -src->i64[2];
>>> +   dst->i64[3] = src->i64[3] >= 0.0 ? src->i64[3] : -src->i64[3];
>>> +}
>>> +
>>> +static void
>>> +micro_i64sgn(union tgsi_double_channel *dst,
>>> +             const union tgsi_double_channel *src)
>>> +{
>>> +   dst->i64[0] = src->i64[0] < 0 ? -1 : src->i64[0] > 0 ? 1 : 0;
>>> +   dst->i64[1] = src->i64[1] < 0 ? -1 : src->i64[1] > 0 ? 1 : 0;
>>> +   dst->i64[2] = src->i64[2] < 0 ? -1 : src->i64[2] > 0 ? 1 : 0;
>>> +   dst->i64[3] = src->i64[3] < 0 ? -1 : src->i64[3] > 0 ? 1 : 0;
>>> +}
>>> +
>>> +static void
>>> +micro_i64neg(union tgsi_double_channel *dst,
>>> +             const union tgsi_double_channel *src)
>>> +{
>>> +   dst->i64[0] = -src->i64[0];
>>> +   dst->i64[1] = -src->i64[1];
>>> +   dst->i64[2] = -src->i64[2];
>>> +   dst->i64[3] = -src->i64[3];
>>> +}
>>> +
>>> +static void
>>> +micro_u64seq(union tgsi_double_channel *dst,
>>> +           const union tgsi_double_channel *src)
>>> +{
>>> +   dst->u[0][0] = src[0].u64[0] == src[1].u64[0] ? ~0U : 0U;
>>> +   dst->u[1][0] = src[0].u64[1] == src[1].u64[1] ? ~0U : 0U;
>>> +   dst->u[2][0] = src[0].u64[2] == src[1].u64[2] ? ~0U : 0U;
>>> +   dst->u[3][0] = src[0].u64[3] == src[1].u64[3] ? ~0U : 0U;
>>> +}
>>
>>
>> I haven't been following the 'double' work so dumb questions/comments:
>>
>> First, could you document the new opcodes in gallium/docs/source/tgsi.rst?
>>
>> In the case of micro_u64seq(), etc. why doesn't it do
>>
>> dst->u64[0] = src[0].u64[0] == src[1].u64[0] ? ~0UL : 0U;
>
> Did you mean ULL? UL is 32 bits on a 32-bit arch.

I think that all the comparisons have to return a boolean, which in
TGSI is a 32-bit 0/~0. Otherwise I suspect there's going to be a ton
of bugs from doing like "int(64bitint == 5)".

  -ilia