On 04/14/2015 04:02 PM, Ben Widawsky wrote:
> On Tue, Apr 14, 2015 at 07:08:14PM +0100, Neil Roberts wrote:
>> +/**
>> + * Emits the instructions needed to perform a pull constant load.
>> before_block
>> + * and before_inst can be NULL in which case the instruction will be
>> appended
>> + * to the end of the instruction list.
>> + */
>> +void
>> +vec4_visitor::emit_pull_constant_load_reg(dst_reg dst,
>> + src_reg surf_index,
>> + src_reg offset_reg,
>> + bblock_t *before_block,
>> + vec4_instruction *before_inst)
>> +{
>> + assert((before_inst == NULL && before_block == NULL) ||
>> + (before_inst && before_block));
>> +
>> + vec4_instruction *pull;
>> +
>> + if (brw->gen >= 9) {
>> + /* Gen9+ needs a message header in order to use SIMD4x2 mode */
>> + src_reg header(this, glsl_type::uvec4_type, 2);
>> +
>
> Just curious why you didn't go for a uvec8_type. It seems more natural for
> what
> we want to do, and how the other code handles things.
Because there is no uvec8 type. The GLSL types used here only expose
types that exist in GLSL.
_______________________________________________
mesa-dev mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
