Module: Mesa Branch: master Commit: 97c078488f4cfbf80af154463b67866e8fbe185c URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=97c078488f4cfbf80af154463b67866e8fbe185c
Author: Iago Toral Quiroga <[email protected]> Date: Thu Feb 11 12:09:28 2021 +0100 broadcom/compiler: disallow unifa overlap with thread switch/end Reviewed-by: Alejandro Piñeiro <[email protected]> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8980> --- src/broadcom/compiler/qpu_schedule.c | 17 +++++++++++++++++ 1 file changed, 17 insertions(+) diff --git a/src/broadcom/compiler/qpu_schedule.c b/src/broadcom/compiler/qpu_schedule.c index 2d563d0c5d8..af864b86e64 100644 --- a/src/broadcom/compiler/qpu_schedule.c +++ b/src/broadcom/compiler/qpu_schedule.c @@ -1217,6 +1217,23 @@ valid_thrsw_sequence(struct v3d_compile *c, struct choose_scoreboard *scoreboard return false; } + /* unifa and the following 3 instructions can't overlap a + * thread switch/end. The docs further clarify that this means + * the cycle at which the actual thread switch/end happens + * and not when the thrsw instruction is processed, which would + * be after the 2 delay slots following the thrsw instruction. + * This means that we can move up a thrsw up to the instruction + * right after unifa: + * + * unifa, r5 + * thrsw + * delay slot 1 + * delay slot 2 + * Thread switch happens here, 4 instructions away from unifa + */ + if (v3d_qpu_writes_unifa(c->devinfo, &qinst->qpu)) + return false; + /* Note that the list is circular, so we can only do this up * to instructions_in_sequence. */ _______________________________________________ mesa-commit mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/mesa-commit
