Re: [Mesa-dev] [PATCH] ir_to_mesa: Fix reallocating registers for shaders with loops

2011-03-27 Thread Ian Romanick
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 On 03/27/2011 01:17 AM, Tom Stellard wrote: > Registers that are used inside of loops need to be considered live > starting with the very first instruction in the loop. > > This fixes the following piglit tests on r300: > - glsl-fs-loop-redundant-cond

Re: [Mesa-dev] [PATCH] ir_to_mesa: Fix reallocating registers for shaders with loops

2011-03-27 Thread Marek Olšák
On Sun, Mar 27, 2011 at 7:03 PM, Kenneth Graunke wrote: > Strangely, glsl-vs-vec4-indexing-temp-dst-inloop already passes on i965. > I'm not sure why it would work on i965 but not r300 if the bug is in the > common code. But it's quite possible. Could the reason be that, unlike i965 (guess), i

Re: [Mesa-dev] [PATCH] ir_to_mesa: Fix reallocating registers for shaders with loops

2011-03-27 Thread Kenneth Graunke
On 03/27/2011 01:17 AM, Tom Stellard wrote: Registers that are used inside of loops need to be considered live starting with the very first instruction in the loop. This fixes the following piglit tests on r300: - glsl-fs-loop-redundant-condition - glsl-vs-loop-redundant-condition - glsl-vs-vec4

[Mesa-dev] [PATCH] ir_to_mesa: Fix reallocating registers for shaders with loops

2011-03-27 Thread Tom Stellard
Registers that are used inside of loops need to be considered live starting with the very first instruction in the loop. This fixes the following piglit tests on r300: - glsl-fs-loop-redundant-condition - glsl-vs-loop-redundant-condition - glsl-vs-vec4-indexing-temp-dst-in-loop NOTE: This is a ca