[Mesa-dev] [PATCH 00/59] Initial arb_gpu_shader_fp64 support to the i965 scalar backend

Francisco Jerez currojerez at riseup.net
Mon May 16 17:28:28 UTC 2016


Samuel Iglesias Gonsálvez <siglesias at igalia.com> writes:

> On 30/04/16 09:52, Francisco Jerez wrote:
>> Samuel Iglesias Gonsálvez <siglesias at igalia.com> writes:
>> 
>>> Hello,
>>>
>>> This patch series continues adding arb_gpu_shader_fp64 support to the
>>> Intel driver. Specifically, this targets the  i965 scalar backend for
>>> BDW+ hardware (vec4 is still under research and gen7 has its own
>>> issues which we intend tackle after gen8).
>>>
>>> This adds most of the fp64 scalar implementation, it starts by enabling
>>> the various lowering passes in NIR for doubles and then adds all the
>>> infrastructure required in the backend to operate with 64-bit floating
>>> point data.
>>>
>>> For reference, this series fixes 1009 fp64 piglit tests in BDW. Fp64
>>> totals look like this:
>>>
>>>      pass:                  2523
>>>      fail:                    46
>>>      crash:                  447
>>>      skip:                    16
>>>      total:                 3032
>>>
>>> There are a few missing things in this series to achieve a perfect fp64
>>> pass rate:
>>>
>>> 1. Fixes to copy propagation. The fp64 code creates new code patterns
>>>    that copy-propagation isn't really ready to handle yet leading to
>>>    incorrect results in some cases. We have 9 patches to fix copy
>>>    propagation for fp64 that we intend to send separately after the
>>>    main fp64 infrastructure has been reviewed.
>>>
>>> 2. ubo/ssbo/shared-variables. We will also send the patches for this in
>>>    a separate series after this one.
>>>
>>> 3. A fix for the SIMD lowering pass to properly handle execmasking when
>>>    transposing the results of split instructions back together. We have
>>>    a local fix for this, but Curro hit the same problem while working
>>>    on SIMD32 and has a better solution for it so we intend to use his
>>>    solution when it is ready.
>>>
>>> 4. Spilling. We don't support spilling of DF registers yet and some
>>>    piglit tests need this to compile. Jason had plans to work on the
>>>    spilling code and address the needs of fp64 along the way.
>>>
>> I wonder if this branch here helps?  It does with SIMD32 at least but
>> it's only lightly tested on master, YMMV.
>> 
>> https://cgit.freedesktop.org/~currojerez/mesa/log/?h=i965-spilling-fixes
>> 
>
> I have tested your branch. It fixes all the register spilling errors we
> hit in fp64 tests \o/
>
Cool, I'll send a largely cleaned up version of that series today
assuming nothing unexpected comes up.

> Sam
>
>>> The series does not introduce any regressions in piglit on ILK, SNB,
>>> HSW, BDW and SKL.
>>>
>>> A branch with this series is available for testing here:
>>>
>>> $ git clone -b i965-fp64-scalar-backend-part-1 https://github.com/Igalia/mesa.git
>>>
>>> You will have to enable the extension with:
>>>
>>> $ export MESA_EXTENSION_OVERRIDE=GL_ARB_gpu_shader_fp64
>>>
>>> The full scalar fp64 implementation, containing also the fixes to
>>> copy-propagation as well as ubo/ssbo and our local fix for the SIMD
>>> lowering pass is available here:
>>>
>>> git clone -b i965-fp64 https://github.com/Igalia/mesa.git
>>>
>>> And for the adventurous, there is also a work-in-progress branch that
>>> adds scalar support for HSW here:
>>>
>>> git clone -b i965-fp64-gen7 https://github.com/Igalia/mesa.git
>>>
>>> Thanks,
>>>
>>> Sam
>>>
>>>
>>> Connor Abbott (33):
>>>   i965: use double lowering pass
>>>   i965: use pack/unpackDouble lowering
>>>   i965/disasm: fix disasm of 3-src doubles
>>>   i965/eu: allow doubles in math instructions
>>>   i965: add brw_imm_df
>>>   i965: add support for getting/setting DF immediates
>>>   i965: add support for disassembling DF immediates
>>>   i965/eu: add support for DF immediates
>>>   i965: fix brw_negate_immediate() for doubles
>>>   i965: fix is_zero(), is_one() and is_negative_one() for doubles
>>>   i965: fixup uniform setup for doubles
>>>   i965/fs: print writemask_all when it's enabled
>>>   i965/fs: use the NIR bit size when creating registers
>>>   i965/fs: don't propagate 64-bit immediates
>>>   i965/fs: add support for printing double immediates
>>>   i965/fs: always pass the bitsize to brw_type_for_nir_type()
>>>   i965/fs: add a stride helper
>>>   i965/fs: add PACK opcode
>>>   i965/fs: add a pass for lowering PACK opcodes
>>>   i965/fs/nir: translate double pack/unpack
>>>   i965/fs: fix type_size() for doubles
>>>   i965/fs: handle uniforms in byte_offset()
>>>   i965/fs: use byte_offset() in offset() for uniforms
>>>   i965/fs: fix assign_constant_locations() for doubles
>>>   i965/fs: generalize SIMD16 interference workaround
>>>   i965/fs: extend exec_size halving in the generator
>>>   i965/fs: fix compares for doubles
>>>   i965/fs: fix regs_read() for uniforms
>>>   i965/fs: fix is_copy_payload() for doubles
>>>   i965/fs: fix regs_written in LOAD_PAYLOAD for doubles
>>>   i965/fs: fix dst width calculation in CSE
>>>   i965/fs: add a pass for legalizing d2f
>>>   i965/fs: add support for f2d and d2f
>>>
>>> Iago Toral Quiroga (15):
>>>   i965: fix brw_saturate_immediate() for doubles
>>>   i965: fix brw_abs_immediate() for doubles
>>>   i965: two-argument instructions can only use 32-bit immediates
>>>   i965/fs: optimize pack double
>>>   i965/fs: optimize unpack double
>>>   i965/fs: handle fp64 opcodes in brw_do_channel_expressions
>>>   i965/fs: We only support 32-bit integer ALU operations for now
>>>   i965/fs: add null_reg_df
>>>   i965/fs: implement fsign() for doubles
>>>   i965/fs: implement d2b
>>>   i965/fs: implement d2i and d2u
>>>   i965/fs: implement i2d and u2d
>>>   i965/fs: rename our lower_d2f pass to lower_d2x
>>>   i965/fs/lower_simd_width: Fix registers written for split instructions
>>>   i965/fs: recognize writes with a subreg_offset > 0 as partial
>>>
>>> Samuel Iglesias Gonsálvez (7):
>>>   i965: enable lrp lowering for doubles
>>>   vc4: lower lrp when operating with double operands
>>>   freedreno/ir3: lower lrp when operating with double operands
>>>   i965/fs: align access to double-based uniforms in push constant buffer
>>>   i965/fs: demote_pull_constants() did not take into account double
>>>     types
>>>   i965/fs: take into account doubles when calculating read_size for
>>>     MOV_INDIRECT
>>>   i965/fs: fix MOV_INDIRECT exec_size for doubles
>>>
>>> Topi Pohjolainen (4):
>>>   i965: Lower DFRACEXP/DLDEXP
>>>   i965: Determine size of double precision float register
>>>   i965: Tell backend register about double precision type
>>>   i965/eu: Allow 3-src float ops with doubles
>>>
>>>  src/gallium/drivers/freedreno/ir3/ir3_nir.c        |   1 +
>>>  src/gallium/drivers/vc4/vc4_program.c              |   1 +
>>>  src/mesa/drivers/dri/i965/Makefile.sources         |   2 +
>>>  src/mesa/drivers/dri/i965/brw_compiler.c           |   2 +
>>>  src/mesa/drivers/dri/i965/brw_compiler.h           |   8 +
>>>  src/mesa/drivers/dri/i965/brw_defines.h            |   9 +
>>>  src/mesa/drivers/dri/i965/brw_disasm.c             |   3 +-
>>>  src/mesa/drivers/dri/i965/brw_eu_emit.c            |  60 +++--
>>>  src/mesa/drivers/dri/i965/brw_fs.cpp               | 106 ++++++--
>>>  src/mesa/drivers/dri/i965/brw_fs.h                 |   6 +-
>>>  src/mesa/drivers/dri/i965/brw_fs_builder.h         |  15 +-
>>>  .../dri/i965/brw_fs_channel_expressions.cpp        |  23 +-
>>>  .../drivers/dri/i965/brw_fs_copy_propagation.cpp   |   3 +
>>>  src/mesa/drivers/dri/i965/brw_fs_cse.cpp           |   3 +-
>>>  src/mesa/drivers/dri/i965/brw_fs_generator.cpp     |  16 +-
>>>  src/mesa/drivers/dri/i965/brw_fs_lower_d2x.cpp     |  75 ++++++
>>>  src/mesa/drivers/dri/i965/brw_fs_lower_pack.cpp    |  59 +++++
>>>  src/mesa/drivers/dri/i965/brw_fs_nir.cpp           | 287 ++++++++++++++++++---
>>>  src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp  |  67 +++--
>>>  src/mesa/drivers/dri/i965/brw_inst.h               |  25 ++
>>>  src/mesa/drivers/dri/i965/brw_ir_fs.h              |  14 +-
>>>  src/mesa/drivers/dri/i965/brw_link.cpp             |   1 +
>>>  src/mesa/drivers/dri/i965/brw_nir.c                |  10 +
>>>  src/mesa/drivers/dri/i965/brw_nir_uniforms.cpp     |   7 +-
>>>  src/mesa/drivers/dri/i965/brw_program.c            |   1 +
>>>  src/mesa/drivers/dri/i965/brw_reg.h                |  10 +
>>>  src/mesa/drivers/dri/i965/brw_shader.cpp           |  73 ++++--
>>>  src/mesa/drivers/dri/i965/brw_shader.h             |   1 +
>>>  src/mesa/drivers/dri/i965/brw_wm.c                 |   2 +
>>>  src/mesa/drivers/dri/i965/gen6_constant_state.c    |  12 +-
>>>  30 files changed, 773 insertions(+), 129 deletions(-)
>>>  create mode 100644 src/mesa/drivers/dri/i965/brw_fs_lower_d2x.cpp
>>>  create mode 100644 src/mesa/drivers/dri/i965/brw_fs_lower_pack.cpp
>>>
>>> -- 
>>> 2.5.0
>>>
>>> _______________________________________________
>>> mesa-dev mailing list
>>> mesa-dev at lists.freedesktop.org
>>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 212 bytes
Desc: not available
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20160516/6499eb7a/attachment.sig>


More information about the mesa-dev mailing list