I did found this when comparing piglit results between 7.11 and 8.0.
With 7.11 this test pass, however merge base was failing so i did a bisect to see what commit fixed this, it was the commit 7d2ff4ae773f2299239d7e1356cb313c11e442a8 so I started another bisect between b1b4ea0b3679db0b8fddaa9663a10d4712bba3b7 (corresponding commit on master) and current head, which ended in
f379d8f73063a4c4d6cf379318c6b37118d46bfa is the first bad commit
Author: Bryan Cain <firstname.lastname@example.org>
Date: Mon Apr 25 23:37:47 2011 -050
st/mesa: Add a GLSL IR to TGSI translator.
It is still a work in progress at this point, but it produces working and
reasonably well-optimized code.
Originally based on ir_to_mesa and st_mesa_to_tgsi, but does not directly use
Mesa IR instructions in TGSI generation, instead generating TGSI from the
intermediate class glsl_to_tgsi_instruction. It also has new optimization
passes to replace _mesa_optimize_program.
glsl-vs-vec4-indexing-temp-dst-in-loop is another test regressed by aforementioned commit. BTW both tests pass with llvmpipe and softpipe, so this is probably just some uncovered r300 compiler bug, not a real regression.
Fixed in git commit b2df031a959f36743527b9abc89913ce4f895de3