mesa.git
9 years agoi965/nir: Move the other lowering passes to before out-of-SSA
Jason Ekstrand [Tue, 16 Dec 2014 00:12:04 +0000 (16:12 -0800)]
i965/nir: Move the other lowering passes to before out-of-SSA

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir/lower_system_values: Handle SSA destinations
Jason Ekstrand [Tue, 16 Dec 2014 01:37:07 +0000 (17:37 -0800)]
nir/lower_system_values: Handle SSA destinations

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir/lower_atomics: Use/support SSA
Jason Ekstrand [Tue, 16 Dec 2014 00:23:21 +0000 (16:23 -0800)]
nir/lower_atomics: Use/support SSA

Previously, lower_atomics was non-SSA only.  We assert-failed if the
destination of an atomic operation intrinsic was an SSA def and we used
temporary registers for computing offsets.  This commit changes both of
these behaviors.  We now use SSA values for computing offsets (so we can
optimize them) and we handle SSA destinations.  We also move the pass to
run before we go out of SSA on i965 as it now generates SSA values.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir/live_variables: Use the new ssa_def iterator
Jason Ekstrand [Mon, 15 Dec 2014 23:15:01 +0000 (15:15 -0800)]
nir/live_variables: Use the new ssa_def iterator

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Use nir_foreach_ssa_def for setting up ssa destinations
Jason Ekstrand [Tue, 16 Dec 2014 03:38:14 +0000 (19:38 -0800)]
nir: Use nir_foreach_ssa_def for setting up ssa destinations

Before, we were using foreach_dest and switching on whether the destination
was an SSA value.  This works, except not all destinations are SSA values
so we have to special-case ssa_undef instructions.  Now that we have a
foreach_ssa_def function, we can iterate over all of the register
destinations in one pass and iterate over the SSA destinations in a second.
This way, if we add other ssa-only instructions, we won't have to worry
about adding them to the special case we have for ssa_undef.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Add a foreach_ssa_def function
Jason Ekstrand [Mon, 15 Dec 2014 23:12:59 +0000 (15:12 -0800)]
nir: Add a foreach_ssa_def function

There are some functions whose destinations are SSA-only and so aren't a
nir_dest.  This provides a function that is capable of iterating over the
SSA definitions defined by those functions.  If you want registers, you
should use the old iterator.

v2: Kenneth Graunke <kenneth@whitecape.org>:
 - Fix nir_foreach_ssa_def's return value.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir/lower_variables: Use a real dominance DFS for variable renaming
Jason Ekstrand [Mon, 15 Dec 2014 22:06:58 +0000 (14:06 -0800)]
nir/lower_variables: Use a real dominance DFS for variable renaming

Previously, we were just iterating over the program "in order" which
kind-of approximates a DFS, but not really.  In particular, we got the
following case wrong:

loop {
   a = 3;
   if (foo) {
      a = 5;
   } else {
      break;
   }
   use(a);
}

where use(a) would get 3 instead of 5 because of premature popping of the
SSA def stack.  Now, since we do an actaul DFS, we should evaluate use(a)
immediately after a = 5 and we should be ok.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Remove predication
Jason Ekstrand [Sat, 13 Dec 2014 06:38:41 +0000 (22:38 -0800)]
nir: Remove predication

We stopped generating predicates in glsl_to_nir some time ago.  Right now,
it's all dead untested code that I'm not convinced always worked in the
first place.  If we decide we want them back, we can revert this patch.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Make bcsel a fully vector operation
Jason Ekstrand [Sat, 13 Dec 2014 04:37:04 +0000 (20:37 -0800)]
nir: Make bcsel a fully vector operation

Previously, the condition was a scalar that applied to all components
simultaneously.  As of this commit, the condition is a vector and each
component is switched seperately.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Call nir_metadata_preserve more places
Jason Ekstrand [Sat, 13 Dec 2014 00:25:38 +0000 (16:25 -0800)]
nir: Call nir_metadata_preserve more places

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir/metadata: Rename metadata_dirty to metadata_preserve
Jason Ekstrand [Sat, 13 Dec 2014 00:22:46 +0000 (16:22 -0800)]
nir/metadata: Rename metadata_dirty to metadata_preserve

nir_metadata_dirty was a terrible name because the parameter it takes is
the metadata to be preserved.  This is really confusing because it looks
like it's doing the opposite of what it is actually doing.  Now it's named
sensibly.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agoi965/fs_nir: Add support for indirect texture arrays
Jason Ekstrand [Sat, 6 Dec 2014 00:43:56 +0000 (16:43 -0800)]
i965/fs_nir: Add support for indirect texture arrays

v2 Jason Ekstrand <jason.ekstrand@intel.com>:
 - Use the nir_tex_src_sampler_offset source type instead of the
   sampler_indirect thing that I cooked up before.

Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
9 years agonir: Rework the way samplers are lowered
Jason Ekstrand [Sat, 6 Dec 2014 00:09:53 +0000 (16:09 -0800)]
nir: Rework the way samplers are lowered

v2 Jason Ekstrand <jason.ekstrand@intel.com>:
 - Use the nir_tex_src_sampler_offset source type instead of the
   sampler_indirect thing that I cooked up before.

Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
9 years agonir/tex_instr_create: Initialize all 4 sources
Jason Ekstrand [Thu, 8 Jan 2015 01:52:37 +0000 (17:52 -0800)]
nir/tex_instr_create: Initialize all 4 sources

This helps a lot with things like lowering passes that may need to add
sources.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir/tex_instr: Rename the indirect source type and add an array size
Jason Ekstrand [Fri, 5 Dec 2014 22:46:24 +0000 (14:46 -0800)]
nir/tex_instr: Rename the indirect source type and add an array size

In particular, we rename nir_tex_src_sampler_index to _sampler_offset and
add a sampler_array_size field to nir_tex_instr.  This way we can pass the
size of sampler arrays through to backends even after removing the variable
information and, with it, the type.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Use a source for uniform buffer indices instead of an index
Jason Ekstrand [Tue, 9 Dec 2014 01:34:52 +0000 (17:34 -0800)]
nir: Use a source for uniform buffer indices instead of an index

In GLSL-to-NIR we were just setting the base index to 0 whenever there was
an indirect so having it expressed as a sum makes no sense.  Also, while a
base offset may make sense for the memory location (first element in the
array, etc.) it makes less sense for the actual uniform buffer index.  This
may change later, but it seems to make more sense for now.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Constant fold array indirects
Jason Ekstrand [Fri, 5 Dec 2014 20:05:55 +0000 (12:05 -0800)]
nir: Constant fold array indirects

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Make texture instruction names more consistent
Jason Ekstrand [Fri, 5 Dec 2014 19:03:06 +0000 (11:03 -0800)]
nir: Make texture instruction names more consistent

This commit renames nir_instr_as_texture to nir_instr_as_tex and renames
nir_instr_type_texture to nir_instr_type_tex to be consistent with
nir_tex_instr.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Remove the ffma peephole
Jason Ekstrand [Sat, 15 Nov 2014 06:09:27 +0000 (22:09 -0800)]
nir: Remove the ffma peephole

This is no longer needed because it's now part of the algebraic
optimization pass

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Add a basic constant folding pass
Jason Ekstrand [Sat, 15 Nov 2014 05:35:25 +0000 (21:35 -0800)]
nir: Add a basic constant folding pass

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Add an algebraic optimization pass
Jason Ekstrand [Fri, 12 Dec 2014 19:13:10 +0000 (11:13 -0800)]
nir: Add an algebraic optimization pass

This pass uses the previously built algebraic transformations framework and
should act as an example for anyone else wanting to make an algebraic
transformation pass for NIR.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Add infastructure for generating algebraic transformation passes
Jason Ekstrand [Sat, 15 Nov 2014 01:47:56 +0000 (17:47 -0800)]
nir: Add infastructure for generating algebraic transformation passes

This commit builds on the nir_search.h infastructure by adding a bit of
python code that makes it stupid easy to write an algebraic transformation
pass.  The nir_algebraic.py file contains four python classes that
correspond directly to the datastructures in nir_search.c and allow you to
easily generate the C code to represent them.  Given a list of
search-and-replace operations, it can then generate a function that applies
those transformations to a shader.

The transformations can be specified manually, or they can be specified
using nested tuples.  The nested tuples make a neat little language for
specifying expression trees and search-and-replace operations in a very
readable and easy-to-edit fasion.

The generated code is also fairly efficient.  Insteady of blindly calling
nir_replace_instr with every single transformation and on every single
instruction, it uses a switch statement on the instruction opcode to do a
first-order culling and only calls nir_replace_instr if the opcode is known
to match the first opcode in the search expression.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Add an expression matching framework
Jason Ekstrand [Fri, 14 Nov 2014 05:19:28 +0000 (21:19 -0800)]
nir: Add an expression matching framework

This framework provides a simple way to do simple search-and-replace
operations on NIR code.  The nir_search.h header provides four simple data
structures for representing expressions:  nir_value and four subtypes:
nir_variable, nir_constant, and nir_expression.  An expression tree can
then be represented by nesting these data structures as needed.  The
nir_replace_instr function takes an instruction, an expression, and a
value; if the instruction matches the expression, it is replaced with a new
chain of instructions to generate the given replacement value.  The
framework keeps track of swizzles on sources and automatically generates
the currect swizzles for the replacement value.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir/glsl: Emit abs, neg, and sat operations instead of source modifiers
Jason Ekstrand [Fri, 14 Nov 2014 01:23:58 +0000 (17:23 -0800)]
nir/glsl: Emit abs, neg, and sat operations instead of source modifiers

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Make the type casting operations static inline functions
Jason Ekstrand [Fri, 5 Dec 2014 19:00:05 +0000 (11:00 -0800)]
nir: Make the type casting operations static inline functions

Previously, the casting operations were macros.  While this is usually
fine, the casting macro used the input parameter twice leading to strange
behavior when you passed the result of another function into it.  Since we
know the source and destination types explicitly, we don't loose anything
by making it a function.

Also, this gives us a nice little macro for creating cast function that
will hopefully prevent mistyping.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Add a lowering pass for adding source modifiers where possible
Jason Ekstrand [Thu, 13 Nov 2014 03:18:05 +0000 (19:18 -0800)]
nir: Add a lowering pass for adding source modifiers where possible

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Add neg, abs, and sat opcodes
Jason Ekstrand [Fri, 14 Nov 2014 01:19:20 +0000 (17:19 -0800)]
nir: Add neg, abs, and sat opcodes

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agoi965/fs_nir: Implement the ARB_gpu_shader5 interpolation intrinsics
Jason Ekstrand [Thu, 4 Dec 2014 23:24:13 +0000 (15:24 -0800)]
i965/fs_nir: Implement the ARB_gpu_shader5 interpolation intrinsics

Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
9 years agoi965/fs_nir: Add a has_indirect flag and clean up some of the input/output code
Jason Ekstrand [Thu, 4 Dec 2014 20:27:29 +0000 (12:27 -0800)]
i965/fs_nir: Add a has_indirect flag and clean up some of the input/output code

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Add a helper for getting a constant value from an SSA source
Jason Ekstrand [Tue, 9 Dec 2014 01:34:23 +0000 (17:34 -0800)]
nir: Add a helper for getting a constant value from an SSA source

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir/glsl: Add support for gpu_shader5 interpolation instrinsics
Jason Ekstrand [Thu, 4 Dec 2014 20:16:33 +0000 (12:16 -0800)]
nir/glsl: Add support for gpu_shader5 interpolation instrinsics

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Add gpu_shader5 interpolation intrinsics
Jason Ekstrand [Thu, 4 Dec 2014 20:16:06 +0000 (12:16 -0800)]
nir: Add gpu_shader5 interpolation intrinsics

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir/validate: Validate intrinsic source/destination sizes
Jason Ekstrand [Thu, 4 Dec 2014 18:41:17 +0000 (10:41 -0800)]
nir/validate: Validate intrinsic source/destination sizes

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Vectorize intrinsics
Jason Ekstrand [Thu, 4 Dec 2014 01:03:19 +0000 (17:03 -0800)]
nir: Vectorize intrinsics

We used to have the number of components built into the intrinsic.  This
meant that all of our load/store intrinsics had vec1, vec2, vec3, and vec4
variants.  This lead to piles of switch statements to generate the correct
intrinsic names, and introspection to figure out the number of components.
We can make things much nicer by allowing "vectorized" intrinsics.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Remove the old variable lowering code
Jason Ekstrand [Thu, 4 Dec 2014 00:25:35 +0000 (16:25 -0800)]
nir: Remove the old variable lowering code

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir/validate: Ensure that outputs are write-only and inputs are read-only
Jason Ekstrand [Wed, 3 Dec 2014 22:47:17 +0000 (14:47 -0800)]
nir/validate: Ensure that outputs are write-only and inputs are read-only

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agoi965/fs_nir: Use the new variable lowering code
Jason Ekstrand [Wed, 26 Nov 2014 23:07:27 +0000 (15:07 -0800)]
i965/fs_nir: Use the new variable lowering code

This commit switches us over to the new variable lowering code which is
capable of properly handling lowering indirects as we go.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir/glsl: Generate SSA NIR
Jason Ekstrand [Mon, 1 Dec 2014 22:11:04 +0000 (14:11 -0800)]
nir/glsl: Generate SSA NIR

With this commit, the GLSL IR -> NIR pass generates NIR in more-or-less SSA
form.  It's SSA in the sense that it doesn't have any registers, but it
isn't really useful SSA because it still has a pile of load/store
intrinsics that we will need to get rid of.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Add a pass to lower global variables to local variables
Jason Ekstrand [Tue, 2 Dec 2014 20:48:38 +0000 (12:48 -0800)]
nir: Add a pass to lower global variables to local variables

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Add a pass for lowering input/output loads/stores
Jason Ekstrand [Tue, 2 Dec 2014 06:01:05 +0000 (22:01 -0800)]
nir: Add a pass for lowering input/output loads/stores

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Add a pass to lower local variables to registers
Jason Ekstrand [Tue, 2 Dec 2014 04:29:35 +0000 (20:29 -0800)]
nir: Add a pass to lower local variables to registers

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Add a pass to lower local variable accesses to SSA values
Jason Ekstrand [Fri, 14 Nov 2014 01:16:31 +0000 (17:16 -0800)]
nir: Add a pass to lower local variable accesses to SSA values

This pass analizes all of the load/store operations and, when a variable is
never aliased (potentially used by an indirect operation), it is lowered
directly to an SSA value.  This pass translates to SSA directly and does
not require any fixup by the original to-SSA pass.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Add a copy splitting pass
Jason Ekstrand [Wed, 19 Nov 2014 22:52:30 +0000 (14:52 -0800)]
nir: Add a copy splitting pass

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Automatically update SSA if uses
Jason Ekstrand [Mon, 1 Dec 2014 22:39:10 +0000 (14:39 -0800)]
nir: Automatically update SSA if uses

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agoi965/fs_nir: Don't dump the shader.
Jason Ekstrand [Mon, 1 Dec 2014 23:09:17 +0000 (15:09 -0800)]
i965/fs_nir: Don't dump the shader.

This is killing piglit.  I'll leave the logging local

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir/glsl: Don't allocate a state_slots array for 0 state slots
Jason Ekstrand [Wed, 3 Dec 2014 02:07:13 +0000 (18:07 -0800)]
nir/glsl: Don't allocate a state_slots array for 0 state slots

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Validate that the sources of a phi have the same size as the destination
Jason Ekstrand [Tue, 25 Nov 2014 05:26:41 +0000 (21:26 -0800)]
nir: Validate that the sources of a phi have the same size as the destination

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir/copy_propagate: Don't cause size mismatches on phi node sources
Jason Ekstrand [Tue, 25 Nov 2014 06:42:16 +0000 (22:42 -0800)]
nir/copy_propagate: Don't cause size mismatches on phi node sources

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Don't require a function in ssa_def_init
Jason Ekstrand [Thu, 20 Nov 2014 00:06:32 +0000 (16:06 -0800)]
nir: Don't require a function in ssa_def_init

Instead, we give SSA definitions a temporary index of 0xFFFFFFFF if the
instruction does not have a block and a proper index when it actually gets
added to the list.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Use an integer index for specifying structure fields
Jason Ekstrand [Wed, 26 Nov 2014 05:36:25 +0000 (21:36 -0800)]
nir: Use an integer index for specifying structure fields

Previously, we used a string name.  It was nice for translating out of GLSL
IR (which also does that) but cumbersome the rest of the time.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Add a concept of a wildcard array dereference
Jason Ekstrand [Wed, 19 Nov 2014 20:59:57 +0000 (12:59 -0800)]
nir: Add a concept of a wildcard array dereference

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Make array deref direct vs. indirect an enum
Jason Ekstrand [Wed, 19 Nov 2014 20:53:08 +0000 (12:53 -0800)]
nir: Make array deref direct vs. indirect an enum

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Clean up nir_deref helper functions
Jason Ekstrand [Thu, 13 Nov 2014 22:53:01 +0000 (14:53 -0800)]
nir: Clean up nir_deref helper functions

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir/lower_samplers: Use the nir_instr_rewrite_src function
Jason Ekstrand [Fri, 14 Nov 2014 17:46:48 +0000 (09:46 -0800)]
nir/lower_samplers: Use the nir_instr_rewrite_src function

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir: Add a helper for rewriting an instruction source
Jason Ekstrand [Fri, 14 Nov 2014 03:07:22 +0000 (19:07 -0800)]
nir: Add a helper for rewriting an instruction source

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agoi965/fs_nir: Properly saturate multiplies
Jason Ekstrand [Fri, 14 Nov 2014 05:34:41 +0000 (21:34 -0800)]
i965/fs_nir: Properly saturate multiplies

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
9 years agonir/from_ssa: Don't lower constant SSA values to registers
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>
9 years agoi965/fs_nir: Handle SSA constants
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>
9 years agoi965/fs_nir: Use an array rather than a hash table for register lookup
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>
9 years agoi965/fs_nir: Add the CSE pass and actually run in a loop
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>
9 years agonir: Add a basic CSE pass
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>
9 years agonir: Add a fused multiply-add peephole
Jason Ekstrand [Tue, 11 Nov 2014 20:16:55 +0000 (12:16 -0800)]
nir: Add a fused multiply-add peephole

9 years agonir: Validate that the SSA def and register indices are unique
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>
9 years agoi965/fs_nir: Turn on the peephole select optimization
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>
9 years agonir: Add a peephole select optimization
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>
9 years agonir/nir: Patch up phi predecessors in move_successors
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>
9 years agonir/nir: Use safe iterators when iterating over the CFG
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>
9 years agoglsl/list: Add a foreach_list_typed_safe_reverse macro
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>
9 years agonir/nir: Fix a bug in move_successors
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>
9 years agoi965/fs_nir: Validate optimization passes
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>
9 years agonir: Differentiate between signed and unsigned versions of find_msb
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>
9 years agonir/print: Don't reindex things
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>
9 years agonir: Validate all lists in the validator
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>
9 years agoglsl/list: Fix the exec_list_validate function
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>
9 years agoi965/fs_nir: Do retyping for ALU srouces in get_nir_alu_src
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>
9 years agonir: Add a better out-of-SSA pass
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>
9 years agonir: Add a function for comparing two sources
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>
9 years agonir: Add a parallel copy instruction type
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>
9 years agonir: Add a function for rewriting all the uses of a SSA def
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>
9 years agonir: Automatically handle SSA uses when an instruction is inserted
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>
9 years agonir: Add an initialization function for SSA definitions
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>
9 years agonir: Add an SSA-based liveness analysis pass.
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>
9 years agonir: set reg_alloc and ssa_alloc when indexing registers and SSA values
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>
9 years agonir: Add a function to detect if a block is immediately followed by an if
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>
9 years agonir: Add a foreach_block_reverse function
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>
9 years agonir/foreach_block: Return false if the callback on the last block fails
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>
9 years agonir: Add a basic metadata management system
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>
9 years agonir/lower_variables_scalar: Silence a compiler warning
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>
9 years agoi965/fs_nir: Convert the shader to/from SSA
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>
9 years agonir: Add a lower_vec_to_movs pass
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>
9 years agonir: Add a naieve from-SSA pass
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.

9 years agoi965/fs_nir: Don't duplicate emit_general_interpolation
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>
9 years agoi965/fs: Don't take an ir_variable for emit_general_interpolation
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>
9 years agonir: Add intrinsics to do alternate interpolation on inputs
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>
9 years agonir: Add NIR_TRUE and NIR_FALSE constants and use them for boolean immediates
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>
9 years agoi965/fs_nir: Add atomic counters support
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>
9 years agonir/lower_atomics: Multiply array offsets by ATOMIC_COUNTER_SIZE
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>
9 years agoi965/fs_nir: Handle coarse/fine derivatives
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>
9 years agonir/glsl: Add support for coarse and fine derivatives
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>
9 years agonir: Add fine and coarse derivative opcodes
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>