We now express arrays in terms of indirect addressing. For example:
dst = a[i];
becomes:
MOV dst, TEMP[1 + TEMP[2].y];
At instruction-emit time indirect addressing is converted into ARL/
ADDR-relative form:
ARL ADDR.x, TEMP[2].y;
MOV dst, TEMP[1 + ADDR.x];
This fixes a number of array-related issues. Arrays of arrays and complex
array/struct nesting works now.
There may be some regressions, but more work is coming.
(cherry picked from commit ae0ff8097b)
This is a step toward better array handling code. In particular, when more
than one operand of an instruction uses indirect addressing, we'll need some
temporary instructions and registers. By converting IR storage to instruction
operands all in one place (emit_instruction()) we can be smarter about this.
Also, somewhat better handling of dst register swizzle/writemask handling.
This results in tighter writemasks on some instructions which is good for
SOA execution.
And, cleaner instruction commenting with inst_comment().
Next: remove some more dead code and additional clean-ups...
(cherry picked from commit 3a7ed9779b)
This lets one specify initial values for uniforms in the code, avoiding
the need to call glUniform() in some cases.
(cherry picked from commit 379ff8c956)
MSVC may not support full C99, but supports more than plain C90. And
-pedantic without -std=c99 generates too many spurious warnings
(specially C++ style comments) to be of any use.
Note that using certain C99 features in the cross-platform parts of Gallium
is still not possible; namely mid-of-scope variable declarations and named
structure initializers will break MSVC builds.
The following construction in util_surface_copy() in
gallium/auxiliary/util/u_rect.c, introduced in commit
d177c9ddda, incorrectly inverts
the Y coordinate in the last parameter to pipe_copy_rect().
/* If do_flip, invert src_y position and pass negative src stride
*/
pipe_copy_rect(dst_map,
&dst->block,
dst->stride,
dst_x, dst_y,
w, h,
src_map,
do_flip ? -(int) src->stride : src->stride,
src_x,
do_flip ? w - src_y : src_y);
The intention is to start at the last Y coordinate line and move
backwards, in the case of a flip; in that case, the correct
calculation is "src_y + h - 1", not "w - src_y".
This fixes a Gallium assertion failure in the conformance tests:
u_rect.c:65:pipe_copy_rect: Assertion `src_y >= 0' failed.
debug_get_bool_option: GALLIUM_ABORT_ON_ASSERT = TRUE
Trace/breakpoint trap
This is a work-around the for the fact that we do fragment shader state
validation before vertex shader validation (see comments in state.c) so in
get_fp_input_mask() we can't rely on ctx->VertexProgram._Current being up to
date yet.
This fixes a glean glsl1 test failure.
Remove the old/initial vbuf allocation in util_create_gen_mipmap().
We were allocating a small vbuf at this point so get_next_slot() didn't have
as large of buffer as it expected. So all but the first set_vertex_data()
was writing out of bounds.
Also added some comments.
The problem we're solving only occured when there was a user-defined
vertex shader but no fragment shader. Check for that case now.
Fixes glean api2 vertex array failure.
