Jason Ekstrand [Wed, 14 Jan 2015 19:19:41 +0000 (11:19 -0800)]
nir/from_ssa: Don't lower constant SSA values to registers
Backends want to be able to do special things with constant values such as
put them into immediates or make decisions based on whether or not a value
is constant. Before, constants always got lowered to a load_const into a
register and then a register use. Now we leave constants as SSA values so
backends can special-case them if they want. Since handling constant SSA
values is trivial, this shouldn't be a problem for backends.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Thu, 13 Nov 2014 00:24:21 +0000 (16:24 -0800)]
i965/fs_nir: Handle SSA constants
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 12 Nov 2014 19:05:51 +0000 (11:05 -0800)]
i965/fs_nir: Use an array rather than a hash table for register lookup
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 12 Nov 2014 00:12:32 +0000 (16:12 -0800)]
i965/fs_nir: Add the CSE pass and actually run in a loop
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 12 Nov 2014 00:11:34 +0000 (16:11 -0800)]
nir: Add a basic CSE pass
This pass is still fairly basic. It only handles ALU operations, constant
loads, and phi nodes. No texture ops or intrinsics yet.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Tue, 11 Nov 2014 20:16:55 +0000 (12:16 -0800)]
nir: Add a fused multiply-add peephole
Jason Ekstrand [Tue, 11 Nov 2014 00:00:03 +0000 (16:00 -0800)]
nir: Validate that the SSA def and register indices are unique
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Sat, 8 Nov 2014 00:07:22 +0000 (16:07 -0800)]
i965/fs_nir: Turn on the peephole select optimization
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Tue, 4 Nov 2014 18:12:14 +0000 (10:12 -0800)]
nir: Add a peephole select optimization
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Sat, 8 Nov 2014 03:35:23 +0000 (19:35 -0800)]
nir/nir: Patch up phi predecessors in move_successors
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Sat, 8 Nov 2014 02:27:36 +0000 (18:27 -0800)]
nir/nir: Use safe iterators when iterating over the CFG
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Sat, 8 Nov 2014 02:26:50 +0000 (18:26 -0800)]
glsl/list: Add a foreach_list_typed_safe_reverse macro
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Sat, 8 Nov 2014 02:25:08 +0000 (18:25 -0800)]
nir/nir: Fix a bug in move_successors
The unlink_blocks function moves successors around to make sure that, if
there is a remaining successor, it is in the first successors slot and not
the second. To fix this, we simply get both successors up front.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Fri, 7 Nov 2014 19:03:12 +0000 (11:03 -0800)]
i965/fs_nir: Validate optimization passes
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Fri, 7 Nov 2014 18:59:16 +0000 (10:59 -0800)]
nir: Differentiate between signed and unsigned versions of find_msb
We also make the return types match GLSL. The GLSL spec specifies that
findMSB and findLSB return a signed integer. Previously, nir had them
return unsigned. This updates nir's behavior to match what GLSL expects.
We also update the nir-to-fs generator to take the new instructions. While
we're at it, we fix the case where the input to findMSB is zero.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Thu, 6 Nov 2014 19:18:42 +0000 (11:18 -0800)]
nir/print: Don't reindex things
These indices should now be reasonably stable/consistent. Redoing the
indices in the print functions makes it harder to debug problems.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 5 Nov 2014 21:58:42 +0000 (13:58 -0800)]
nir: Validate all lists in the validator
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Tue, 11 Nov 2014 18:12:24 +0000 (10:12 -0800)]
glsl/list: Fix the exec_list_validate function
Some time while refactoring things to make it look nicer before pushing to
master, I completely broke the function. This fixes it to be correct.
Just goes to show you why you souldn't push code that has no users yet...
Signed-off-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Jason Ekstrand [Fri, 12 Dec 2014 21:05:25 +0000 (13:05 -0800)]
i965/fs_nir: Do retyping for ALU srouces in get_nir_alu_src
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Fri, 31 Oct 2014 18:17:09 +0000 (11:17 -0700)]
nir: Add a better out-of-SSA pass
This commit rewrites the out-of-SSA pass to not be nearly as naieve. It's
based on "Revisiting Out-of-SSA Translation for Correctness, Code Quality,
and Efficiency" by Boissinot et. al. It should be fairly close to
state-of-the art.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Fri, 12 Dec 2014 20:52:11 +0000 (12:52 -0800)]
nir: Add a function for comparing two sources
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Fri, 31 Oct 2014 04:04:15 +0000 (21:04 -0700)]
nir: Add a parallel copy instruction type
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 5 Nov 2014 01:18:48 +0000 (17:18 -0800)]
nir: Add a function for rewriting all the uses of a SSA def
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Tue, 4 Nov 2014 19:02:09 +0000 (11:02 -0800)]
nir: Automatically handle SSA uses when an instruction is inserted
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Tue, 4 Nov 2014 18:40:48 +0000 (10:40 -0800)]
nir: Add an initialization function for SSA definitions
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 29 Oct 2014 21:17:17 +0000 (14:17 -0700)]
nir: Add an SSA-based liveness analysis pass.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Fri, 31 Oct 2014 04:18:22 +0000 (21:18 -0700)]
nir: set reg_alloc and ssa_alloc when indexing registers and SSA values
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 29 Oct 2014 23:25:51 +0000 (16:25 -0700)]
nir: Add a function to detect if a block is immediately followed by an if
Since we don't actually have an "if" instruction, this is a very common
pattern when iterating over instructions. This adds a helper function for
it to make things a little less painful.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 29 Oct 2014 21:16:54 +0000 (14:16 -0700)]
nir: Add a foreach_block_reverse function
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 29 Oct 2014 21:16:39 +0000 (14:16 -0700)]
nir/foreach_block: Return false if the callback on the last block fails
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 29 Oct 2014 19:42:54 +0000 (12:42 -0700)]
nir: Add a basic metadata management system
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 29 Oct 2014 19:42:33 +0000 (12:42 -0700)]
nir/lower_variables_scalar: Silence a compiler warning
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 22 Oct 2014 18:24:33 +0000 (11:24 -0700)]
i965/fs_nir: Convert the shader to/from SSA
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 22 Oct 2014 19:57:28 +0000 (12:57 -0700)]
nir: Add a lower_vec_to_movs pass
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 22 Oct 2014 18:22:53 +0000 (11:22 -0700)]
nir: Add a naieve from-SSA pass
This pass is kind of stupidly implemented but it should be enough to get us
up and going. We probably want something better that doesn't generate all
of the redundant moves eventually. However, the i965 backend should be
able to handle the movs, so I'm not too worried about it in the short term.
Jason Ekstrand [Tue, 21 Oct 2014 01:07:28 +0000 (18:07 -0700)]
i965/fs_nir: Don't duplicate emit_general_interpolation
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Tue, 21 Oct 2014 01:05:36 +0000 (18:05 -0700)]
i965/fs: Don't take an ir_variable for emit_general_interpolation
Previously, emit_general_interpolation took an ir_variable and pulled the
information it needed from that. This meant that in fs_fp, we were
constructing a dummy ir_variable just to pass into it. This commit makes
emit_general_interpolation take only the information it needs and gets rid
of the fs_fp cruft.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Sat, 18 Oct 2014 00:11:34 +0000 (17:11 -0700)]
nir: Add intrinsics to do alternate interpolation on inputs
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Thu, 16 Oct 2014 23:53:03 +0000 (16:53 -0700)]
nir: Add NIR_TRUE and NIR_FALSE constants and use them for boolean immediates
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Thu, 16 Oct 2014 04:52:58 +0000 (21:52 -0700)]
i965/fs_nir: Add atomic counters support
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Thu, 16 Oct 2014 16:56:14 +0000 (09:56 -0700)]
nir/lower_atomics: Multiply array offsets by ATOMIC_COUNTER_SIZE
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 15 Oct 2014 21:44:00 +0000 (14:44 -0700)]
i965/fs_nir: Handle coarse/fine derivatives
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 15 Oct 2014 23:57:10 +0000 (16:57 -0700)]
nir/glsl: Add support for coarse and fine derivatives
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 15 Oct 2014 23:56:43 +0000 (16:56 -0700)]
nir: Add fine and coarse derivative opcodes
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 15 Oct 2014 23:19:26 +0000 (16:19 -0700)]
nir/glsl: Add support for saturate
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 15 Oct 2014 23:01:04 +0000 (16:01 -0700)]
i965/fs_nir: Add support for sample_pos and sample_id
Jason Ekstrand [Wed, 15 Oct 2014 22:36:43 +0000 (15:36 -0700)]
Fix up varying pull constants
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 15 Oct 2014 20:56:48 +0000 (13:56 -0700)]
Fix what I think are a few NIR typos
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 15 Oct 2014 22:25:10 +0000 (15:25 -0700)]
i965/fs_nir: Use the correct texture offset immediate
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 15 Oct 2014 19:18:25 +0000 (12:18 -0700)]
i965/fs_nir: Use the correct types for texture inputs
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Wed, 15 Oct 2014 17:41:04 +0000 (10:41 -0700)]
i965/fs_nir: Make the sampler register always unsigned
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Jason Ekstrand [Tue, 14 Oct 2014 23:40:04 +0000 (16:40 -0700)]
i965/fs: Only use nir for 8-wide non-fast-clear shaders.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Connor Abbott [Fri, 15 Aug 2014 17:32:07 +0000 (10:32 -0700)]
i965/fs: add a NIR frontend
This is similar to the GLSL IR frontend, except consuming NIR. This lets
us test NIR as part of an actual compiler.
v2: Jason Ekstrand <jason.ekstrand@intel.com>:
Make brw_fs_nir build again
Only use NIR of INTEL_USE_NIR is set
whitespace fixes
Connor Abbott [Fri, 15 Aug 2014 17:17:26 +0000 (10:17 -0700)]
i965/fs: Don't pass through the coordinate type
All we really need is the number of components.
Connor Abbott [Tue, 5 Aug 2014 18:02:02 +0000 (11:02 -0700)]
i965/fs: make emit_fragcoord_interpolation() not take an ir_variable
Connor Abbott [Thu, 24 Jul 2014 22:51:58 +0000 (15:51 -0700)]
nir: add an SSA-based dead code elimination pass
v2: Jason Ekstrand <jason.ekstrand@intel.com>:
whitespace fixes
Connor Abbott [Wed, 23 Jul 2014 18:19:50 +0000 (11:19 -0700)]
nir: add an SSA-based copy propagation pass
Connor Abbott [Tue, 22 Jul 2014 21:05:06 +0000 (14:05 -0700)]
nir: add a pass to convert to SSA
v2: Jason Ekstrand <jason.ekstrand@intel.com>:
whitespace fixes
Connor Abbott [Fri, 18 Jul 2014 23:13:11 +0000 (16:13 -0700)]
nir: calculate dominance information
Connor Abbott [Wed, 30 Jul 2014 19:08:13 +0000 (12:08 -0700)]
nir: add an optimization to turn global registers into local registers
After linking and inlining, this allows us to convert these registers
into SSA values and optimise more code.
Connor Abbott [Wed, 30 Jul 2014 21:43:26 +0000 (14:43 -0700)]
nir: add a pass to lower atomics
v2: Jason Ekstrand <jason.ekstrand@intel.com>
whitespace fixes
Connor Abbott [Wed, 30 Jul 2014 19:07:45 +0000 (12:07 -0700)]
nir: add a pass to lower system value reads
v2: Jason Ekstrand <jason.ekstrand@intel.com>:
whitespace fixes
Connor Abbott [Wed, 30 Jul 2014 19:04:49 +0000 (12:04 -0700)]
nir: add a pass to lower sampler instructions
Connor Abbott [Wed, 30 Jul 2014 18:56:52 +0000 (11:56 -0700)]
nir: add a pass to remove unused variables
After we lower variables, we want to delete them in order to free up
some memory.
v2: Jason Ekstrand <jason.ekstrand@intel.com>:
whitespace fixes
Connor Abbott [Tue, 5 Aug 2014 17:54:27 +0000 (10:54 -0700)]
nir: keep track of the number of input, output, and uniform slots
Connor Abbott [Thu, 17 Jul 2014 16:12:52 +0000 (09:12 -0700)]
nir: add a pass to lower variables for scalar backends
Connor Abbott [Fri, 11 Jul 2014 01:18:17 +0000 (18:18 -0700)]
nir: add a glsl-to-nir pass
v2: Jason Ekstrand <jason.ekstrand@intel.com>:
Make glsl_to_nir build again
fix whitespace
Connor Abbott [Wed, 30 Jul 2014 22:20:53 +0000 (15:20 -0700)]
nir: add a validation pass
This is similar to ir_validate.cpp.
v2: Jason Ekstrand <jason.ekstrand@intel.com>:
whitespace fixes
Connor Abbott [Wed, 30 Jul 2014 22:29:27 +0000 (15:29 -0700)]
nir: add a printer
This is similar to ir_print_visitor.cpp.
v2: Jason Ekstrand <jason.ekstrand@intel.com>:
whitespace fixes
Jason Ekstrand [Thu, 18 Dec 2014 01:30:27 +0000 (17:30 -0800)]
SQUASH: Fix comments from eric
Reviewed-by: Eric Anholt <eric@anholt.net>
Jason Ekstrand [Wed, 29 Oct 2014 21:15:13 +0000 (14:15 -0700)]
SQUASH: Add an assert
Connor Abbott [Thu, 31 Jul 2014 23:16:23 +0000 (16:16 -0700)]
nir: add core helper functions
These include functions for adding and removing various bits of IR and
helpers for iterating over all the sources and destinations of an
instruction. This is similar to ir.cpp.
v2: Jason Ekstrand <jason.ekstrand@intel.com>:
whitespace and automake fixes
Jason Ekstrand [Wed, 26 Nov 2014 23:08:19 +0000 (15:08 -0800)]
SQUASH: Use the enum for the variable mode
Connor Abbott [Thu, 31 Jul 2014 23:14:51 +0000 (16:14 -0700)]
nir: add the core datastructures
This includes all the instructions, ifs, loops, functions, etc. This is
similar to the information in ir.h.
v2: Jason Ekstrand <jason.ekstrand@intel.com>:
Include ralloc and hash_table from the util directory
whitespace fixes
Signed-off-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-By glenn.kennard <glenn.kennard@gmail.com>
Connor Abbott [Wed, 30 Jul 2014 22:33:32 +0000 (15:33 -0700)]
nir: add a simple C wrapper around glsl_types.h
v2: Jason Ekstrand <jason.ekstrand@intel.com>:
whitespace and automake fixes
Reviewed-by: Eric Anholt <eric@anholt.net>
Connor Abbott [Wed, 30 Jul 2014 22:32:21 +0000 (15:32 -0700)]
nir: add initial README
Reviewed-by: Eric Anholt <eric@anholt.net>
Connor Abbott [Tue, 22 Jul 2014 00:11:53 +0000 (17:11 -0700)]
exec_list: add a list_foreach_typed_reverse() macro
Reviewed-by: Eric Anholt <eric@anholt.net>
Eric Anholt [Tue, 13 Jan 2015 22:23:43 +0000 (11:23 +1300)]
vc4: Add some dumping for STORE_TILE_BUFFER_GENERAL.
Eric Anholt [Tue, 13 Jan 2015 21:53:20 +0000 (10:53 +1300)]
vc4: Add dumping for the TILE_RENDERING_MODE_CONFIG packet.
I wanted to read it, so I wrote parsing.
Eric Anholt [Tue, 13 Jan 2015 21:06:02 +0000 (10:06 +1300)]
vc4: Fix CL dumping trying to dump too far.
Execution will end at the cl->next, because that's what ct0ea/ct1ea get
programmed to.
Eric Anholt [Tue, 13 Jan 2015 03:43:16 +0000 (16:43 +1300)]
vc4: Fix texture type masking.
Everything from ETC1 to RGBA64 was getting its top bit dropped, but we
didn't use any of those formats.
Eric Anholt [Mon, 12 Jan 2015 01:53:48 +0000 (14:53 +1300)]
vc4: Colormask should apply after all other fragment ops (like logic op).
Theoretically it should apply after dithering as well, but ditehring for
565 happens in fixed function in the TLB store.
Eric Anholt [Sun, 11 Jan 2015 20:14:41 +0000 (09:14 +1300)]
vc4: No turning unpack arguments into small immediates.
Since unpack only happens on things read from the A register file, we have
to leave them as something that can be allocated to A (temp or uniform).
Eric Anholt [Sun, 11 Jan 2015 20:10:35 +0000 (09:10 +1300)]
vc4: Move the tests for src needing to be an A register to vc4_qir.c.
I want it from another location.
Eric Anholt [Sun, 11 Jan 2015 20:16:26 +0000 (09:16 +1300)]
vc4: Don't swap the raddr on instructions doing unpacks.
It would mean different unpacking behavior, since only the A file does
unpack (with PM==0).
Eric Anholt [Sun, 11 Jan 2015 06:31:59 +0000 (19:31 +1300)]
vc4: Don't let pairing happen with badly mismatched unpack flags.
No difference on shader-db, but prevents definite regressions in the
blending changes.
Eric Anholt [Sun, 11 Jan 2015 05:27:07 +0000 (18:27 +1300)]
vc4: Don't let pairing happen with badly mismatched pack flags.
No difference on shader-db, but will become more important as I introduce
more use of pack flags with the blending changes.
Eric Anholt [Wed, 14 Jan 2015 04:11:59 +0000 (17:11 +1300)]
vc4: Fix early Z behavior on hardware.
It turns out the simulator was not treating this bit the same as the RPi,
and I'd forgotten to remove it when turning on early Z. The result was
that you'd get big chunks of your rendering missing.
Michel Dänzer [Tue, 13 Jan 2015 07:38:52 +0000 (16:38 +0900)]
Revert "radeonsi: only set BC_OPTIMIZE_DISABLE when necessary"
This reverts commit
0543630d0b0d9d9f6eefbc14fbd3385d4de37ba0.
It caused flickering artifacts in Steam games such as Team Fortress 2 or
Left 4 Dead 2.
We could probably only enable this optimization by also making sure the
shader code only uses either SI_PARAM_LINEAR_CENTROID or
SI_PARAM_LINEAR_CENTER, not both. This would probably require a shader
variant.
Sorry I didn't remember this when reviewing the reverted change.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Michel Dänzer [Thu, 15 Jan 2015 03:57:05 +0000 (12:57 +0900)]
st/clover: Adapt to TargetLibraryInfo.h move in LLVM SVN r226078
Trivial.
Ian Romanick [Fri, 7 Nov 2014 06:51:45 +0000 (22:51 -0800)]
mesa: Micro-optimize _mesa_is_valid_prim_mode
You would not believe the mess GCC 4.8.3 generated for the old
switch-statement.
On Bay Trail-D using Fedora 20 compile flags (-m64 -O2 -mtune=generic
for 64-bit and -m32 -march=i686 -mtune=atom for 32-bit), affects
Gl32Batch7:
32-bit: Difference at 95.0% confidence -0.37374% +/- 0.184057% (n=40)
64-bit: Difference at 95.0% confidence 0.966722% +/- 0.338442% (n=40)
The regression on 32-bit is odd. Callgrind says the caller,
_mesa_is_valid_prim_mode is faster. Before it says 2,293,760
cycles, and after it says 917,504.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Ian Romanick [Tue, 11 Nov 2014 10:29:34 +0000 (10:29 +0000)]
mesa: Check for vertex program the same way in desktop GL and ES
On Bay Trail-D using Fedora 20 compile flags (-m64 -O2 -mtune=generic
for 64-bit and -m32 -march=i686 -mtune=atom for 32-bit), affects
Gl32Multithread:
32-bit: Difference at 95.0% confidence 0.416027% +/- 0.163529% (n=40)
64-bit: Difference at 95.0% confidence 0.494771% +/- 0.259985% (n=40)
Gl32Batch7 had no difference proven at 95.0% confidence (n=120) on
32-bit or 64-bit.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Ian Romanick [Tue, 11 Nov 2014 09:21:40 +0000 (09:21 +0000)]
mesa: Drop index buffer bounds check
The previous check was insufficient (as it did not take 'indices' into
consideration), and DX10 hardware does not need this check anyway.
Since index_bytes is no longer used, remove it.
On Bay Trail-D using Fedora 20 compile flags (-m64 -O2 -mtune=generic
for 64-bit and -m32 -march=i686 -mtune=atom for 32-bit), affects
Gl32Batch7:
32-bit: Difference at 95.0% confidence 1.66929% +/- 0.230107% (n=40)
64-bit: Difference at 95.0% confidence -1.40848% +/- 0.288038% (n=40)
The regression on 64-bit is odd. Callgrind says the caller,
validate_DrawElements_common is faster. Before it says 10,321,920
cycles, and after it says 8,945,664.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Ian Romanick [Tue, 11 Nov 2014 11:28:28 +0000 (11:28 +0000)]
mesa: Only check for a current vertex shader in core profile
This doesn't affect performance, but it feels more correct.
On Bay Trail-D using Fedora 20 compile flags (-m64 -O2 -mtune=generic
for 64-bit and -m32 -march=i686 -mtune=atom for 32-bit), affects
Gl32Batch7:
32-bit: No difference proven at 95.0% confidence (n=120)
64-bit: No difference proven at 95.0% confidence (n=120)
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Ian Romanick [Tue, 11 Nov 2014 12:31:22 +0000 (12:31 +0000)]
mesa: Only validate shaders that can exist in the context
On Bay Trail-D using Fedora 20 compile flags (-m64 -O2 -mtune=generic
for 64-bit and -m32 -march=i686 -mtune=atom for 32-bit), affects
Gl32Batch7:
32-bit: Difference at 95.0% confidence 0.495267% +/- 0.202063% (n=40)
64-bit: Difference at 95.0% confidence 3.57576% +/- 0.288175% (n=40)
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Ian Romanick [Tue, 11 Nov 2014 14:51:29 +0000 (14:51 +0000)]
i965: Store the atoms directly in the context
Instead of having an extra pointer indirection in one of the hottest
loops in the driver.
On Bay Trail-D using Fedora 20 compile flags (-m64 -O2 -mtune=generic
for 64-bit and -m32 -march=i686 -mtune=atom for 32-bit), affects
Gl32Batch7:
32-bit: Difference at 95.0% confidence 1.98515% +/- 0.20814% (n=40)
64-bit: Difference at 95.0% confidence 1.5163% +/- 0.811016% (n=60)
v2 (Ken): Cut size of array from 64 to 57 to save memory.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Ian Romanick [Mon, 10 Nov 2014 14:06:47 +0000 (06:06 -0800)]
i965: Micro-optimize brw_get_index_type
With the switch-statement, GCC 4.8.3 produces a small pile of code with
a branch.
00000000 <brw_get_index_type>:
000000: 8b 54 24 04 mov 0x4(%esp),%edx
000004: b8 01 00 00 00 mov $0x1,%eax
000009: 81 fa 03 14 00 00 cmp $0x1403,%edx
00000f: 74 0d je 00001e <brw_get_index_type+0x1e>
000011: 31 c0 xor %eax,%eax
000013: 81 fa 05 14 00 00 cmp $0x1405,%edx
000019: 0f 94 c0 sete %al
00001c: 01 c0 add %eax,%eax
00001e: c3 ret
However, this could be two instructions.
00000000 <brw_get_index_type>:
000000: 2d 01 14 00 00 sub $0x1401,%eax
000005: d1 e8 shr %eax
000007: 90 nop
000008: 90 nop
000009: 90 nop
00000a: 90 nop
00000b: c3 ret
The function was also moved to the header so that it could be inlined at
the two call sites. Without this, 32-bit also needs to pull the
parameter from the stack. This means there is a push, a call, a move,
and a ret added to a two instruction function. The above code shows the
function with __attribute__((regparm=1)), but even this adds several
extra instructions. There is also an extra instruction on 64-bit to
move the parameter to %eax for the subtract.
On Bay Trail-D using Fedora 20 compile flags (-m64 -O2 -mtune=generic
for 64-bit and -m32 -march=i686 -mtune=atom for 32-bit), affects
Gl32Batch7:
32-bit: Difference at 95.0% confidence 0.818589% +/- 0.234661% (n=40)
64-bit: Difference at 95.0% confidence 0.54554% +/- 0.354092% (n=40)
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Ian Romanick [Tue, 11 Nov 2014 14:14:14 +0000 (14:14 +0000)]
meta: Put _mesa_meta_in_progress in the header file
...so that it can be inlined in the two places that call it.
On Bay Trail-D using Fedora 20 compile flags (-m64 -O2 -mtune=generic
for 64-bit and -m32 -march=i686 -mtune=atom for 32-bit), affects
Gl32Batch7:
32-bit: No difference proven at 95.0% confidence (n=120)
64-bit: Difference at 95.0% confidence 1.24042% +/- 0.382277% (n=40)
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Kenneth Graunke [Tue, 13 Jan 2015 22:56:54 +0000 (14:56 -0800)]
i965: Fix "vertex" vs. "geometry" and "VS" vs. "GS" in debug output.
We were happily printing "Native code for unnamed vertex shader" and
"VS vec4" program for geometry shaders in our INTEL_DEBUG=gs output,
as well as the KHR_debug output used by shader-db.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Kenneth Graunke [Tue, 13 Jan 2015 22:28:13 +0000 (14:28 -0800)]
i965: Pass a shader stage abbreviation to fs_generator().
A lot of messages hardcoded the string "FS", which is confusing on
Broadwell, where we use this code for VS support as well.
shader-db particularly got confused, as it reported two "FS SIMD8"
shaders, and no vertex shaders at all. Craziness ensued.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>