Daniel Schürmann
2ae27b96ef
aco: change live_out variables to std::unordered_set
...
Improves performance of live_var_analysis for larger shaders
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4130 >
2020-04-09 15:08:57 +00:00
Daniel Schürmann
b66f474121
aco: improve speed of live_var_analysis
...
by merging live_sgprs and live_vgprs sets.
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4130 >
2020-04-09 15:08:57 +00:00
Samuel Pitoiset
2f424c83e0
aco: only break SMEM clauses if XNACK is enabled (mostly APUs)
...
According to LLVM, it seems only required for APUs like RAVEN, but
we still ensure that SMEM stores are in their own clause.
pipeline-db (VEGA10):
Totals from affected shaders:
SGPRS: 1775364 -> 1775364 (0.00 %)
VGPRS: 1287176 -> 1287176 (0.00 %)
Spilled SGPRs: 725 -> 725 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Code Size: 65386620 -> 65107460 (-0.43 %) bytes
Max Waves: 287099 -> 287099 (0.00 %)
pipeline-db (POLARIS10):
Totals from affected shaders:
SGPRS: 1797743 -> 1797743 (0.00 %)
VGPRS: 1271108 -> 1271108 (0.00 %)
Spilled SGPRs: 730 -> 730 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Code Size: 64046244 -> 63782324 (-0.41 %) bytes
Max Waves: 254875 -> 254875 (0.00 %)
This only affects GFX6-GFX9 chips because the compiler uses a
different pass for GFX10.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4349 >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4349 >
2020-04-01 17:50:31 +00:00
Timur Kristóf
0f35b3795d
aco: Fix workgroup size calculation.
...
Clear the workgroup size for all supported shader stages.
Also, unify the workgroup size calculation accross various places.
As a result, insert_waitcnt can use the proper workgroup size
which means that some waits can be dropped from tessellation
shaders. Also, in cases where the previous calculation was wrong,
we now insert s_barrier instructions.
Totals from affected shaders (GFX10):
Code Size: 340116 -> 338484 (-0.48 %) bytes
Fixes: a8d15ab6dahttps://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4165 >
2020-03-30 13:09:08 +00:00
Rhys Perry
1872759f55
aco: add a late kill flag
...
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/3914 >
2020-03-16 16:09:02 +00:00
Rhys Perry
c51348bd9b
aco: move some register demand helpers into aco_live_var_analysis.cpp
...
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/3914 >
2020-03-16 16:09:02 +00:00
Rhys Perry
b088a4b113
aco: only reserve sgprs for vcc if it's used
...
pipeline-db (Vega):
Totals:
SGPRS: 5186302 -> 5075616 (-2.13 %)
VGPRS: 3704580 -> 3704580 (0.00 %)
Spilled SGPRs: 144859 -> 144859 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Scratch size: 4124 -> 4124 (0.00 %) dwords per thread
Code Size: 247315944 -> 247315944 (0.00 %) bytes
LDS: 1311 -> 1311 (0.00 %) blocks
Max Waves: 674560 -> 674562 (0.00 %)
Totals from affected shaders:
SGPRS: 536992 -> 426306 (-20.61 %)
VGPRS: 356404 -> 356404 (0.00 %)
Spilled SGPRs: 0 -> 0 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 8498748 -> 8498748 (0.00 %) bytes
LDS: 8 -> 8 (0.00 %) blocks
Max Waves: 113832 -> 113834 (0.00 %)
There are some small code size changes in a few RotTR shaders and a small
increase in max_waves in two Detroit: Become Human shaders.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3906 >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3906 >
2020-03-05 20:18:34 +00:00
Rhys Perry
4e83e05e62
aco: error when block has no logical preds but VGPRs are live at the start
...
This would have caught the liveness error fixed in the previous commit.
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/3257 >
2020-01-29 18:02:27 +00:00
Rhys Perry
3f96a1ed86
aco: fix operand kill flags when a temporary is used more than once
...
Helps create v_mac_f32 from v_mad_f32(b, a, b)
Totals from affected shaders:
SGPRS: 35824 -> 35824 (0.00 %)
VGPRS: 33460 -> 33456 (-0.01 %)
Spilled SGPRs: 0 -> 0 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 2187264 -> 2180976 (-0.29 %) bytes
LDS: 127 -> 127 (0.00 %) blocks
Max Waves: 3802 -> 3802 (0.00 %)
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3486 >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3486 >
2020-01-22 15:55:00 +00:00
Rhys Perry
b5c9688516
aco: limit register usage for large work groups
...
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
2020-01-10 12:10:37 +00:00
Rhys Perry
afe1a8ff5b
aco: fix vgpr alloc granule with wave32
...
We still need to increase the number of physical vgprs
Totals from affected shaders:
SGPRS: 671976 -> 675288 (0.49 %)
VGPRS: 550112 -> 562596 (2.27 %)
Spilled SGPRs: 0 -> 0 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Code Size: 27621660 -> 27606532 (-0.05 %) bytes
Max Waves: 81083 -> 87833 (8.32 %)
Instructions: 5391560 -> 5389031 (-0.05 %)
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
2019-12-21 12:38:42 +01:00
Timur Kristóf
e0bcefc3a0
aco/wave32: Use lane mask regclass for exec/vcc.
...
Currently all usages of exec and vcc are hardcoded to use s2 regclass.
This commit makes it possible to use s1 in wave32 mode and
s2 in wave64 mode.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
2019-12-04 10:36:01 +00:00
Daniel Schürmann
78bca0d0ce
aco: improve live variable analysis
...
This patch makes the live variable analysis more precise
w.r.t. killed phi operands and the block's register pressure.
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
2019-10-30 19:48:32 +00:00
Timur Kristóf
c52ebbcea4
aco: Introduce vgpr_limit to keep track of available VGPRs.
...
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
2019-10-28 23:52:50 +00:00
Rhys Perry
fc04a2fc31
aco: take LDS into account when calculating num_waves
...
pipeline-db (Vega):
SGPRS: 344 -> 344 (0.00 %)
VGPRS: 424 -> 524 (23.58 %)
Spilled SGPRs: 84 -> 80 (-4.76 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 52812 -> 52484 (-0.62 %) bytes
LDS: 135 -> 135 (0.00 %) blocks
Max Waves: 56 -> 53 (-5.36 %)
v2: consider WGP, rework to be clearer and apply the
"maximum 16 workgroups per CU" limit properly
v2: use "SIMD" instead of "EU"
v2: fix spiller by introducing "Program::max_waves"
v2: rename "lds_size" to "lds_limit"
v3: make max_waves actually independant of register usage
v3: fix issue where max_waves was way too high
v3: use DIV_ROUND_UP(a, b) instead of max(a / b, 1)
v3: rename "workgroups_per_cu" to "workgroups_per_cu_wgp"
v4: fix typo from "workgroups_per_cu" rename
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev> (v3)
2019-10-23 19:11:21 +01:00
Rhys Perry
08d510010b
aco: increase accuracy of SGPR limits
...
SGPRs are allocated in groups of 16 on GFX8/GFX9. GFX10 allocates a fixed
number of SGPRs and has 106 addressable SGPRs.
pipeline-db (Vega):
SGPRS: 5912 -> 6232 (5.41 %)
VGPRS: 1772 -> 1780 (0.45 %)
Spilled SGPRs: 0 -> 0 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 88228 -> 87904 (-0.37 %) bytes
LDS: 0 -> 0 (0.00 %) blocks
Max Waves: 559 -> 571 (2.15 %)
piepline-db (Navi):
SGPRS: 341256 -> 363384 (6.48 %)
VGPRS: 171536 -> 170960 (-0.34 %)
Spilled SGPRs: 832 -> 581 (-30.17 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 14207332 -> 14190872 (-0.12 %) bytes
LDS: 33 -> 33 (0.00 %) blocks
Max Waves: 18072 -> 18251 (0.99 %)
v2: unconditionally count vcc as an extra sgpr on GFX10+
v3: pass SGPRs rounded to 8
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
2019-10-23 19:11:21 +01:00
Daniel Schürmann
93c8ebfa78
aco: Initial commit of independent AMD compiler
...
ACO (short for AMD Compiler) is a new compiler backend with the goal to replace
LLVM for Radeon hardware for the RADV driver.
ACO currently supports only VS, PS and CS on VI and Vega.
There are some optimizations missing because of unmerged NIR changes
which may decrease performance.
Full commit history can be found at
https://github.com/daniel-schuermann/mesa/commits/backend
Co-authored-by: Daniel Schürmann <daniel@schuermann.dev>
Co-authored-by: Rhys Perry <pendingchaos02@gmail.com>
Co-authored-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Co-authored-by: Connor Abbott <cwabbott0@gmail.com>
Co-authored-by: Michael Schellenberger Costa <mschellenbergercosta@googlemail.com>
Co-authored-by: Timur Kristóf <timur.kristof@gmail.com>
Acked-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Acked-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
2019-09-19 12:10:00 +02:00
