diff options
| -rw-r--r-- | src/broadcom/compiler/qpu_schedule.c | 27 |
1 files changed, 23 insertions, 4 deletions
diff --git a/src/broadcom/compiler/qpu_schedule.c b/src/broadcom/compiler/qpu_schedule.c index 370881b00ad..c15218e267e 100644 --- a/src/broadcom/compiler/qpu_schedule.c +++ b/src/broadcom/compiler/qpu_schedule.c @@ -560,6 +560,11 @@ mux_read_stalls(struct choose_scoreboard *scoreboard, scoreboard->last_stallable_sfu_reg); } +/* We define a max schedule priority to allow negative priorities as result of + * substracting this max when an instruction stalls. So instructions that + * stall have lower priority than regular instructions. */ +#define MAX_SCHEDULE_PRIORITY 16 + static int get_instruction_priority(const struct v3d_qpu_instr *inst) { @@ -578,10 +583,6 @@ get_instruction_priority(const struct v3d_qpu_instr *inst) return next_score; next_score++; - /* XXX perf: We should schedule SFU ALU ops so that the reader is 2 - * instructions after the producer if possible, not just 1. - */ - /* Default score for things that aren't otherwise special. */ baseline_score = next_score; next_score++; @@ -591,6 +592,9 @@ get_instruction_priority(const struct v3d_qpu_instr *inst) return next_score; next_score++; + /* We should increase the maximum if we assert here */ + assert(next_score < MAX_SCHEDULE_PRIORITY); + return baseline_score; } @@ -845,6 +849,18 @@ choose_instruction_to_schedule(const struct v3d_device_info *devinfo, int prio = get_instruction_priority(inst); + if (mux_read_stalls(scoreboard, inst)) { + /* Don't merge an instruction that stalls */ + if (prev_inst) + continue; + else { + /* Any instruction that don't stall will have + * higher scheduling priority */ + prio -= MAX_SCHEDULE_PRIORITY; + assert(prio < 0); + } + } + /* Found a valid instruction. If nothing better comes along, * this one works. */ @@ -1005,6 +1021,9 @@ instruction_latency(struct schedule_node *before, struct schedule_node *after) after_inst)); } + if (v3d_qpu_instr_is_sfu(before_inst)) + return 2; + return latency; } |