These events were silently truncated in get_counters_for_event.
The integer types in this pass are a bit all over the place, maybe we should
consider using typedefs for clarity or a different solution with type safety.
Fixes: 9e9cabd2fa ("aco/waitcnt: support GFX12 in waitcnt pass")
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30163>
Both null def and op result in the same correct encoding, but these
instructions optionally read a sgpr, so it makes more sense to use an operand.
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26163>
Next part don't know whether p_end_with_regs args are loaded from
memory ops or not, need to wait it's done here.
Other memory load needs to be waited too like:
a = load_mem()
b = ...
if (...) {
wait_mem(a)
store_mem(a)
}
p_end_with_regs(b)
"a" still needs to be waited, otherwise next shader part regs may
be overwritten by unfinished memory loads.
Memory stores are waited too. When >=gfx10 and last VGT has no
parameter export, we need to wait all memeory stores done before
pos export (see ac_nir_export_position). So when merged shader
(ES+GS or VS+GS) is partially built, first stage needs to wait
all memory stores done, otherwise second stage don't know if
any memory stores pending before.
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Signe-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24973>
Previously, a loop could be revisited until a counter reaches it's
maximum:
loop {
store()
}
Each visit of that loop would increase vs_cnt until it reaches max.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25373>
For tcs epilog to add p_barrier at the beginning to sync
main shader part tess factor LDS write.
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24442>
When letting the overlapping waves enter their ordered sections, there must
be no memory accesses to resources which need primitive-ordered access that
are still pending, or there would be a race between the current wave and
the overlapping waves.
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Vitaliy Triang3l Kuzmin <triang3l@yandex.ru>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22250>
Waits are needed for early exits from inside a Primitive Ordered Pixel
Shading ordered section, but that code doesn't insert them reliably anyway
because it doesn't obtain the counters for the exact locations of the
jumps, which may be anywhere inside the predecessor blocks.
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Vitaliy Triang3l Kuzmin <triang3l@yandex.ru>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22250>
For:
v_mov_b32_e32 v0, 1.0
exp mrtz v0, off, off, off
we should completely remove the ALU entry before creating the EXP's WaR entry for v0.
Otherwise, the two will be combined into an entry which will wait for
expcnt(0) for later uses of v0.
gen_alu() should also be before gen(), since gen_alu() performs the clear
while gen() creates the WaR entry.
fossil-db (gfx1100):
Totals from 3589 (2.69% of 133428) affected shaders:
Instrs: 5591041 -> 5589047 (-0.04%); split: -0.04%, +0.00%
CodeSize: 28580840 -> 28572864 (-0.03%); split: -0.03%, +0.00%
Latency: 65427923 -> 65427543 (-0.00%); split: -0.00%, +0.00%
InvThroughput: 11109079 -> 11109065 (-0.00%); split: -0.00%, +0.00%
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23213>
Previously, we could safely read/write unused lanes of VMEM/DS
destination VGPRs without waiting for the load to finish. That doesn't
seem to be the case on GFX11.
fossil-db (gfx1100):
Totals from 6698 (4.94% of 135636) affected shaders:
Instrs: 11184274 -> 11199420 (+0.14%); split: -0.00%, +0.14%
CodeSize: 57578344 -> 57638928 (+0.11%); split: -0.00%, +0.11%
Latency: 198348808 -> 198382472 (+0.02%); split: -0.00%, +0.02%
InvThroughput: 24376324 -> 24378439 (+0.01%); split: -0.00%, +0.01%
VClause: 192420 -> 192559 (+0.07%)
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8722
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8239
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22965>
Handle arguments that need a waitcnt without relying on
RADV specific VS input information.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21696>
It's just v_fma with fixed DPP8 and builtin s_waitcnt_expcnt, so it can mostly
be handled as a pure VALU instruction.
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21023>
Use gpr_map to determine how many cycles each dependency of the
s_delay_alu needs. This information helps the pass avoid further
s_delay_alu instructions.
fossil-db (gfx1100):
Totals from 13097 (9.73% of 134574) affected shaders:
Instrs: 30711894 -> 30702692 (-0.03%)
CodeSize: 153462500 -> 153425692 (-0.02%)
Latency: 372758612 -> 372741922 (-0.00%)
InvThroughput: 50164111 -> 50160717 (-0.01%); split: -0.01%, +0.00%
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20512>
The LLVM backend does this when lowering ordered_xfb_counter_add_amd. I
guess there is some missing dependency checking or something.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19345>
The first operand of this new pseudo-instruction is a 64-bit SGPR for
the continue PC, followed by a variable list of fixed VGPRS for the
color exports which are the PS epilog inputs.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17485>
According to LLVM, image_bvh64_intersect_ray does not write results in
order with other VMEM instructions.
fossil-db (navi21):
Totals from 7 (0.00% of 162293) affected shaders:
Instrs: 39978 -> 39985 (+0.02%)
CodeSize: 219356 -> 219384 (+0.01%)
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17079>
If we don't reset ctx.vm_cnt/gpr_map/etc, this will spam a lot of
s_waitcnt instructions.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17207>
By default, std::stack uses std::deque to allocate its elements, which has
poor cache efficiency. std::vector makes appending elements more expensive
(due to potential reallocations), but in the changed contexts the element
count should always be low anyway.
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11925>
