[Mesa-dev] [PATCH 1/2] mesa/st: add support for dynamic ubo selection

[Brian Paul]

On 07/15/2014 05:28 AM, Ilia Mirkin wrote:
> On Wed, Jul 9, 2014 at 9:06 AM, Brian Paul <brianp at vmware.com> wrote:
>> On 07/08/2014 08:40 PM, Ilia Mirkin wrote:
>>>
>>> Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
>>> ---
>>>
>>> With ChrisF's patches to add support for this in core mesa, this generates
>>> code like:
>>>
>>> FRAG
>>> DCL OUT[0], COLOR
>>> DCL CONST[0]
>>> DCL CONST[1][0]
>>> DCL CONST[2][0]
>>> DCL CONST[3][0]
>>> DCL CONST[4][0]
>>> DCL TEMP[0], LOCAL
>>> DCL ADDR[0..1]
>>> IMM[0] UINT32 {0, 0, 0, 0}
>>> IMM[1] INT32 {1, 0, 0, 0}
>>>     0: UADD TEMP[0].x, CONST[0].xxxx, IMM[1].xxxx
>>>     1: UARL ADDR[1].x, TEMP[0].xxxx
>>>     2: UARL ADDR[1].x, TEMP[0].xxxx
>>>     3: MOV TEMP[0], CONST[ADDR[1].x][0]
>>>     4: MOV OUT[0], TEMP[0]
>>>     5: END
>>>
>>> Not sure what the deal is with the two UARL's, but nouveau's backend
>>> removes
>>> one of them pretty easily. I assume others handle this too.
>>
>>
>> Yeah, I noticed this too when I was doing some UBO work last week.  I
>> haven't had time to investigate yet.
>
> Does this change look good though? Looks like Chris's non-const UBO
> indexing patches are ~ready. With some tweaks to the nvc0 logic, the
> simple tests pass on nouveau (and there are no 'complex' tests).

I'm not an expert on this code, but look OK AFAICT.

Reviewed-by: Brian Paul <brianp at vmware.com>


>
>>
>> -Brian
>>
>>
>>>
>>> Unfortunately the core patches aren't quite ready yet, but this patch
>>> doesn't
>>> regress anything.
>>>
>>>    src/mesa/state_tracker/st_glsl_to_tgsi.cpp | 20 +++++++++++++++++---
>>>    1 file changed, 17 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
>>> b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
>>> index 9bc7500..3202c56 100644
>>> --- a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
>>> +++ b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
>>> @@ -1947,16 +1947,16 @@ glsl_to_tgsi_visitor::visit(ir_expression *ir)
>>>          break;
>>>
>>>       case ir_binop_ubo_load: {
>>> -      ir_constant *uniform_block = ir->operands[0]->as_constant();
>>> +      ir_constant *const_uniform_block = ir->operands[0]->as_constant();
>>>          ir_constant *const_offset_ir = ir->operands[1]->as_constant();
>>>          unsigned const_offset = const_offset_ir ?
>>> const_offset_ir->value.u[0] : 0;
>>> +      unsigned const_block = const_uniform_block ?
>>> const_uniform_block->value.u[0] + 1 : 0;
>>>          st_src_reg index_reg = get_temp(glsl_type::uint_type);
>>>          st_src_reg cbuf;
>>>
>>>          cbuf.type = glsl_type::vec4_type->base_type;
>>>          cbuf.file = PROGRAM_CONSTANT;
>>>          cbuf.index = 0;
>>> -      cbuf.index2D = uniform_block->value.u[0] + 1;
>>>          cbuf.reladdr = NULL;
>>>          cbuf.negate = 0;
>>>
>>> @@ -1966,7 +1966,6 @@ glsl_to_tgsi_visitor::visit(ir_expression *ir)
>>>             /* Constant index into constant buffer */
>>>             cbuf.reladdr = NULL;
>>>             cbuf.index = const_offset / 16;
>>> -         cbuf.has_index2 = true;
>>>          }
>>>          else {
>>>             /* Relative/variable index into constant buffer */
>>> @@ -1976,6 +1975,21 @@ glsl_to_tgsi_visitor::visit(ir_expression *ir)
>>>             memcpy(cbuf.reladdr, &index_reg, sizeof(index_reg));
>>>          }
>>>
>>> +      if (const_uniform_block) {
>>> +         /* Constant constant buffer */
>>> +         cbuf.reladdr2 = NULL;
>>> +         cbuf.index2D = const_block;
>>> +         cbuf.has_index2 = true;
>>> +      }
>>> +      else {
>>> +         /* Relative/variable constant buffer */
>>> +         emit(ir, TGSI_OPCODE_UADD, st_dst_reg(index_reg), op[0],
>>> +              st_src_reg_for_int(1));
>>> +         cbuf.reladdr2 = ralloc(mem_ctx, st_src_reg);
>>> +         memcpy(cbuf.reladdr2, &index_reg, sizeof(index_reg));
>>> +         cbuf.has_index2 = true;
>>> +      }
>>> +
>>>          cbuf.swizzle = swizzle_for_size(ir->type->vector_elements);
>>>          cbuf.swizzle += MAKE_SWIZZLE4(const_offset % 16 / 4,
>>>                                        const_offset % 16 / 4,
>>>
>>
>> _______________________________________________
>> mesa-dev mailing list
>> mesa-dev at lists.freedesktop.org
>> https://urldefense.proofpoint.com/v1/url?u=http://lists.freedesktop.org/mailman/listinfo/mesa-dev&k=oIvRg1%2BdGAgOoM1BIlLLqw%3D%3D%0A&r=lGQMzzTgII0I7jefp2FHq7WtZ%2BTLs8wadB%2BiIj9xpBY%3D%0A&m=2PxC4euC4MUhdxrARICGkoYmsaKgMNJkIz%2BtzTOqBt0%3D%0A&s=d21974c40a550eabfa7fea826f13ec3c3f1e9a1f0fd99ad856975bac472793b5