Age | Commit message (Collapse) | Author |
|
|
|
The x86 code seems to fail for exponents of 4 or larger.
See glean's vertProg1/EX2 test.
|
|
Fixes codegen for instructions like MUL dst, CONST[0], IMM[0]; the two operands
would up getting aliased in the x86/sse code.
Fixes glean/vertProg1/fogparams test.
|
|
|
|
|
|
|
|
|
|
|
|
substitute"
There was a bug in emit_degenerate_lit() that caused the SLT to produce
unpredictable results in lit.z
Plus, added a bunch of new comments.
|
|
many drivers
|
|
This reverts commit e841b92d9c8bf48085b4996df828ae745977f931.
This fixes two specular lighting conform failures.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
For numeric options.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Also, check the FEATURE flags in many places.
|
|
|
|
|
|
|
|
|
|
flatshade-first"
This reverts commit ad44e68706877ab06929747e7a82c718c1c27e02.
|
|
The driver/pipeline will still be applying flatshade-first state to the
triangles emitted from vcache, so there's no need to rotate the vertices
of most primitives. The only exception is POLYGON, as explained in the
code.
|
|
|
|
|
|
|
|
|
|
|
|
For situations where one has a reserve memory pool, or
a faster/slower pool.
|
|
Reuse cached buffers up to twice as big a requested.
|