Again, this matches what the builder will have to do.
Signed-off-by: Connor Abbott
---
src/glsl/nir/glsl_to_nir.cpp | 36 ++--
1 file changed, 18 insertions(+), 18 deletions(-)
diff --git a/src/glsl/nir/glsl_to_nir.cpp b/src/glsl/nir/glsl_to_nir.cpp
index
p.org/~cwabbott0/mesa glsl-to-nir-builder
Connor Abbott (4):
nir/glsl: make evaluate_rvalue() return a nir_ssa_def *
nir/glsl: convert nir_visitor::result to a nir_ssa_def *
nir/glsl: make emit() take nir_ssa_def * sources
nir/glsl: switch to using the builder
src/glsl/nir/glsl_to_nir.cpp
There was already some work done for this (with i965) a while back:
http://lists.freedesktop.org/archives/mesa-dev/2014-July/063008.html
IIRC the reason it didn't get merged was that Chia never actually made
sure that the compiler was threadsafe (despite some patches fixing
some things). I can't
Reviewed-by: Connor Abbott
On Mon, Nov 2, 2015 at 4:30 AM, Vinson Lee wrote:
> nir/nir_control_flow.c: In function ‘split_block_cursor.isra.11’:
> nir/nir_control_flow.c:460:15: warning: ‘after’ may be used uninitialized in
> this function [-Wmaybe-uninitialized]
>*_a
On Mon, Nov 2, 2015 at 8:35 AM, Emil Velikov wrote:
> Hi all,
>
> From a quick look, it seems that NIR copies (almost ?) all the state
> from GLSL IR even if it doesn't use it.
>
> The particular piece that I'm thinking about is nir_variable::data.
> Afaict this is a remnant from the early days, w
On Tue, Nov 3, 2015 at 5:08 AM, Emil Velikov wrote:
> On 2 November 2015 at 19:41, Jason Ekstrand wrote:
>> On Mon, Nov 2, 2015 at 9:33 AM, Connor Abbott wrote:
>>> On Mon, Nov 2, 2015 at 8:35 AM, Emil Velikov
>>> wrote:
>>>> Hi all,
>>>&g
Hi all,
While working on FP64 for i965, there's an issue that I thought of
with the vec4 backend that I'm not sure how to resolve. From what I
understand, the execmask works the same way in Align16 mode as Align1
mode, except that you only use the first 8 channels in practice for
SIMD4x2, and the
Reviewed-by: Connor Abbott
On Fri, Nov 6, 2015 at 1:59 PM, Jason Ekstrand wrote:
> Previously, we were assuming that everything read/wrote exactly 1 logical
> GRF (1 in SIMD8 and 2 in SIMD16). This isn't actually true. In
> particular, the PLN instruction reads 2 logical regis
Both are r-b me, go ahead and push them.
On Sat, Nov 7, 2015 at 3:03 PM, Jason Ekstrand wrote:
> These are pretty trivial fixes. I pushed a patch that I'd pulled of
> another branch and forgot I had these in it.
>
> Jason Ekstrand (2):
> nir/types: Add an is_vector_or_scalar helper
> nir: Ad
On Tue, Nov 3, 2015 at 8:04 PM, Francisco Jerez wrote:
> Francisco Jerez writes:
>
>> Connor Abbott writes:
>>
>>> Hi all,
>>>
>>> While working on FP64 for i965, there's an issue that I thought of
>>> with the vec4 backend that I
On Mon, Nov 9, 2015 at 6:55 AM, Iago Toral wrote:
> Hi,
>
> Currently, NIR defines vecN operations as unsigned (integer). The fp64
> patches from Connor change this to float (I guess because we need to
> know the case where we are packing vectors of 64-bit floats). However,
> this makes it so that
On Mon, Nov 9, 2015 at 10:41 AM, Jason Ekstrand wrote:
>
> On Nov 9, 2015 7:24 AM, "Connor Abbott" wrote:
>>
>> On Mon, Nov 9, 2015 at 6:55 AM, Iago Toral wrote:
>> > Hi,
>> >
>> > Currently, NIR defines vecN operations as unsigned (int
On Sun, Nov 8, 2015 at 3:12 PM, Rob Clark wrote:
> ---
> src/mesa/state_tracker/st_glsl_to_tgsi.cpp | 410
> -
> src/mesa/state_tracker/st_glsl_to_tgsi.h | 5 +
> src/mesa/state_tracker/st_program.c| 118 +++--
> src/mesa/state_tracker/st_program.h
Since subroutine types are really just integers, should we just make
glsl_type_is_vector_or_scalar() return true for them?
On Tue, Nov 10, 2015 at 5:18 PM, Jason Ekstrand wrote:
> The shader-subroutine code creates uniforms of type SUBROUTINE for
> subroutines that are then read as integers in th
On Tue, Nov 10, 2015 at 7:02 PM, Ilia Mirkin wrote:
> On Tue, Nov 10, 2015 at 6:44 PM, Eric Anholt wrote:
>> Ilia Mirkin writes:
>>
>>> In nv50, and in the python script that Rob circulated, we do:
>>>
>>>bld.mkCmp(OP_SET, CC_GE, TYPE_U32, (s = bld.getSSA()), TYPE_U32, m, b);
>>>
>>> Do the
On Tue, Nov 10, 2015 at 7:49 PM, Jason Ekstrand wrote:
> On Tue, Nov 10, 2015 at 4:18 PM, Connor Abbott wrote:
>> Since subroutine types are really just integers, should we just make
>> glsl_type_is_vector_or_scalar() return true for them?
>
> That's an interesting ques
Reviewed-by: Connor Abbott
On Thu, Nov 12, 2015 at 10:02 PM, Jason Ekstrand wrote:
> Previously, we walked through a given deref_node's copies and, after
> lowering the copy away, removed it from both the source and destination
> copy sets. This commit changes this to only remo
On Thu, Nov 12, 2015 at 9:36 PM, Jason Ekstrand wrote:
> bump! Where are we at on this?
Go ahead and push it with my r-b. I still think
glsl_type::is_vector_or_scalar() should return true for subroutine
types, but that might require some more fixing and testing.
>
> On Tue, Nov 10, 2015 at 6:16
Reviewed-by: Connor Abbott
On Fri, Nov 13, 2015 at 12:53 AM, Jason Ekstrand wrote:
> This was a remnant of an early attempt to handle output reads in
> vars_to_ssa. That attempt was abandon a long time ago but these few lines
> were aparently left in the pass and managed to eva
On Fri, Nov 13, 2015 at 9:38 AM, Ilia Mirkin wrote:
> On Fri, Nov 13, 2015 at 9:25 AM, Emil Velikov
> wrote:
>> Hello Hans,
>>
>> Not to muddy the waters or anything, have you thought about the NIR
>> integration that Rob was thinking about ?
>> I'm pretty sure he'll be happy to have extra peopl
I think it would be better if we kept track of the type of the
constant instead. That would also allow us to simplify the constant
construction code in, err, something else...
On Wed, Nov 11, 2015 at 8:23 PM, Jason Ekstrand wrote:
> From: Rob Clark
>
> This will simplify things somewhat in clone
On Sat, Nov 14, 2015 at 11:55 AM, Jason Ekstrand wrote:
> On Sat, Nov 14, 2015 at 8:19 AM, Connor Abbott wrote:
>> I think it would be better if we kept track of the type of the
>> constant instead. That would also allow us to simplify the constant
>> construction code in
suppose, but for now this makes clone possible.
glsl_get_array_length() will do the right thing for everything except
matrices and vectors/scalars, so it should be possible to get the
number of elements in only a few lines.
>
> BR,
> -R
>
> On Sat, Nov 14, 2015 at 11:19 AM, Connor A
Whoops.
Reviewed-by: Connor Abbott
On Sat, Nov 14, 2015 at 8:49 PM, Matt Turner wrote:
> ---
> src/glsl/nir/glsl_to_nir.cpp | 6 +++---
> 1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/src/glsl/nir/glsl_to_nir.cpp b/src/glsl/nir/glsl_to_nir.cpp
> inde
Series is
Reviewed-by: Connor Abbott
Although I'm not as familiar now with the code touched in the last patch.
On Thu, Nov 12, 2015 at 3:13 PM, Jason Ekstrand wrote:
> The subject says it all. This little series adds texture swizzle support
> to nir_lower_tex and makes the i965
The equivalent of the last patch for the hash table. I'm not aware of
any issues this fixes.
Signed-off-by: Connor Abbott
---
src/util/hash_table.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/src/util/hash_table.c b/src/util/hash_table.c
index 3247593..466519f 1
What happens if you have something like foo = vec3(foo.z, bar.x,
foo.x)? I don't think emitting vector mov's for only the contiguous
components is enough.
On Thu, Jan 22, 2015 at 4:51 PM, Eric Anholt wrote:
> It now emits vector MOVs instead of a series of individual MOVs, which
> should be usefu
Argh, nevermind, I was reading it wrong...
On Thu, Jan 22, 2015 at 8:18 PM, Connor Abbott wrote:
> What happens if you have something like foo = vec3(foo.z, bar.x,
> foo.x)? I don't think emitting vector mov's for only the contiguous
> components is enough.
>
> On Thu
On Wed, Jan 21, 2015 at 11:03 PM, Jason Ekstrand wrote:
> Connor wrote *exactly* the same patch:
>
> http://lists.freedesktop.org/archives/mesa-dev/2015-January/074522.html
>
> This version can have my R-B too. And, fwiw, I like your commit message
> better.
Indeed... let's push this version, si
We can do this now that we have opcode info available in Python.
Signed-off-by: Connor Abbott
---
src/glsl/nir/nir_algebraic.py | 38 ++
src/glsl/nir/nir_opt_algebraic.py | 6 ++
2 files changed, 32 insertions(+), 12 deletions(-)
diff --git a/src
Signed-off-by: Connor Abbott
---
src/glsl/nir/nir_opt_constant_folding.c | 179
1 file changed, 21 insertions(+), 158 deletions(-)
diff --git a/src/glsl/nir/nir_opt_constant_folding.c
b/src/glsl/nir/nir_opt_constant_folding.c
index 878436b..e8a361f 100644
--- a
Signed-off-by: Connor Abbott
---
src/glsl/nir/.gitignore | 1 +
1 file changed, 1 insertion(+)
create mode 100644 src/glsl/nir/.gitignore
diff --git a/src/glsl/nir/.gitignore b/src/glsl/nir/.gitignore
new file mode 100644
index 000..6d954fe
--- /dev/null
+++ b/src/glsl/nir/.gitignore
taken from
ir_constant_expressions.cpp.
v2: use Python formatting and get rid of regex's
v2.5:
- fixup Makefile changes
- remove unused global wr(), rename wr() to wrap() (Jason)
- remove optional arguments in wr() and a few functions in unop_reduce()
and binop_reduce() that were unnecessary
Signed-off-by
rom object (Dylan)
- don't use assert like it's a function (Dylan)
- style fixes for fnoise, use xrange (Dylan)
- use iterkeys() in nir_opcodes_h.py (Dylan)
- use pydoc-style comments (Jason)
- don't make fmin/fmax commutative and associative yet (Jason)
Signed-off-by: Con
r-opcodes-cleanup-v2
Connor Abbott (5):
nir: add generated file to .gitignore
nir: use Python to autogenerate opcode information
nir: add new constant folding infrastructure
nir/constant_folding: use the new constant folding infrastructure
nir/opt_algebraic: be more careful about const
On Fri, Jan 23, 2015 at 1:07 PM, Jason Ekstrand wrote:
>
>
> On Thu, Jan 22, 2015 at 8:32 PM, Connor Abbott wrote:
>>
>> Before, we used a system where a file, nir_opcodes.h, defined some macros
>> that
>> were included to generate the enum values and th
> diff --git a/src/glsl/nir/.gitignore b/src/glsl/nir/.gitignore
> index 4c28193..261f64f 100644
> --- a/src/glsl/nir/.gitignore
> +++ b/src/glsl/nir/.gitignore
> @@ -1,3 +1,4 @@
> nir_opt_algebraic.c
> nir_opcodes.c
> nir_opcodes.h
> +nir_constant_expressions.c
> diff -
On Fri, Jan 23, 2015 at 5:34 PM, Eric Anholt wrote:
> Connor Abbott writes:
>
>> Argh, nevermind, I was reading it wrong...
>>
>> On Thu, Jan 22, 2015 at 8:18 PM, Connor Abbott wrote:
>>> What happens if you have something like foo = vec3(foo.z, bar.x,
>>
Other than the one comment fix below,
Reviewed-by: Connor Abbott
On Fri, Jan 23, 2015 at 7:17 PM, Jason Ekstrand wrote:
> Add a required field to the Opcode class, const_expr, that contains an
> expression or statement that computes the result of the opcode given known
> constant inp
On Sat, Jan 24, 2015 at 1:00 AM, Jason Ekstrand wrote:
> ---
> src/glsl/Makefile.sources | 1 +
> src/glsl/nir/nir.h | 2 +
> src/glsl/nir/nir_lower_phis_to_scalar.c | 238
>
> 3 files changed, 241 insertions(+)
> create mo
On Sat, Jan 24, 2015 at 12:22 PM, Jason Ekstrand wrote:
>
> On Jan 24, 2015 8:18 AM, "Connor Abbott" wrote:
>>
>> On Sat, Jan 24, 2015 at 1:00 AM, Jason Ekstrand
>> wrote:
>> > ---
>> > src/glsl/Makefile.sources | 1 +
r per-component instructions to say how many source
components are being used.
Signed-off-by: Connor Abbott
---
src/glsl/nir/nir.h | 15 +++
1 file changed, 15 insertions(+)
diff --git a/src/glsl/nir/nir.h b/src/glsl/nir/nir.h
index 0ef83a1..efcaa9d 100644
--- a/src/glsl/nir/nir.h
+++
the last patch.
Signed-off-by: Connor Abbott
---
Whoops, sorry I missed this during review...
src/glsl/nir/nir_opt_constant_folding.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/src/glsl/nir/nir_opt_constant_folding.c
b/src/glsl/nir/nir_opt_constant_folding.c
index
Reviewed-by: Connor Abbott
On Mon, Jan 26, 2015 at 12:43 PM, Jason Ekstrand wrote:
> Previously, we called the abs() function in math.h. However, this involves
> unnecessarily going through double. This commit changes it to use integers
> directly with a ternary.
> ---
>
Indeed, I guess I misunderstood the purpose of this opcode when I was
writing the constant expression. I see now that it's supposed to be
used for scalarizing unpack_half_2x16. Goes to show how useful this
self-documenting stuff is, I guess.
Reviewed-by: Connor Abbott
On Mon, Jan 26, 2015
Reviewed-by: Connor Abbott
On Fri, Jan 23, 2015 at 7:59 PM, Jason Ekstrand wrote:
> This avoids the overhead of copying structures and better matches the newly
> added nir_alu_src_copy and nir_alu_dest_copy.
> ---
>
> This should be obvious, but this applies on top of Eric An
On Mon, Jan 26, 2015 at 2:21 PM, Eric Anholt wrote:
> I just notied: When do I need to all nir_metadata_preserve()? I didn't
> in my pass.
It dirties all the metadata (topologically-sorted block indices,
dominance info, liveness info, etc.) except the stuff you tell it to
preserve via the flags.
On Mon, Jan 26, 2015 at 2:31 PM, Eric Anholt wrote:
> Connor Abbott writes:
>
>> On Fri, Jan 23, 2015 at 5:34 PM, Eric Anholt wrote:
>>> Connor Abbott writes:
>>>
>>>> Argh, nevermind, I was reading it wrong...
>>>>
>>>> On Thu,
On Sun, Jan 25, 2015 at 7:51 PM, Jason Ekstrand wrote:
>
>
> On Sun, Jan 25, 2015 at 8:56 AM, Connor Abbott wrote:
>>
>> Unlike with non-SSA ALU instructions, where if they're per-component
>> you have to look at the writemask to know which source channels are
>
The rest of the series seems reasonable although I'm not an expert,
but this one is definitely
Reviewed-by: Connor Abbott
On Wed, Jan 28, 2015 at 3:46 PM, Jason Ekstrand wrote:
> ---
> src/glsl/nir/nir_opcodes.py | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
&
Hi Eric,
A general suggestion:
I think that we should rename these functions to "nir_op_create()" so
that we can make a more specific "nir_op()" function that takes and
returns a nir_ssa_def *, so that we can do stuff like:
nir_builder *b = ...
nir_ssa_def *new_index = nir_iadd(b, nir_imul(b, o
I mentioned this on IRC, but it would be good to add the ability to
append the sequence of instructions before/after an instruction as
well as at the beginning or end of a basic block. We would need to
store the current basic block, the current instruction, and an enum
consisting of "insert before
I think this logic should be flipped around... you should have a flag
'lower_sub' which is false for vc4 but true for i965, and then if the
flag is true we lower sub to add+neg and if false we try and
reconstruct sub (and do other things like pulling neg's out of
multiplies). That way, eventually w
rate have to do this
already. Rather, what probably saves us is that we already recognize
this pattern in GLSL IR. Other than that,
Reviewed-by: Connor Abbott
> ---
> src/glsl/nir/nir_opt_algebraic.py | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/
On Sun, Feb 1, 2015 at 6:54 PM, Eric Anholt wrote:
> Connor Abbott writes:
>
>> On Sun, Feb 1, 2015 at 4:17 PM, Eric Anholt wrote:
>>> We've probably never seen this ridiculous pattern in the wild, so it
>>> didn't matter.
>>
>> I don't
A few comments here...
1. We should really split this up into a patch that adds
nir_shader_compiler_options (which would be empty at this point) and
the nir_algebraic infrastructure for conditionalizing transforms
(actually, maybe we should split these up into two patches) and a
patch which adds l
This removes phi nodes whose sources all point to the same thing.
Only compile tested.
Signed-off-by: Connor Abbott
---
src/glsl/Makefile.sources | 1 +
src/glsl/nir/nir.h | 2 +
src/glsl/nir/nir_opt_remove_phis.c | 111 +
3
Signed-off-by: Connor Abbott
---
src/mesa/drivers/dri/i965/brw_fs_nir.cpp | 2 ++
1 file changed, 2 insertions(+)
diff --git a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
index de0d780..4305e14 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
+++ b
On Tue, Feb 3, 2015 at 1:54 AM, Connor Abbott wrote:
> This removes phi nodes whose sources all point to the same thing.
>
> Only compile tested.
>
> Signed-off-by: Connor Abbott
> ---
> src/glsl/Makefile.sources | 1 +
> src/glsl/nir/nir.h
On Tue, Feb 3, 2015 at 1:49 PM, Jason Ekstrand wrote:
> For whatever reason, I thought we might have non-ssa phi sources. However,
> I looked again and I don't think it's possible. Also, this patch applies
> with no piglit regressions.
Yeah, the only reason I made it so that phi instructions co
On Tue, Feb 3, 2015 at 3:27 PM, Kenneth Graunke wrote:
> On Tuesday, February 03, 2015 07:10:20 AM you wrote:
>> On Feb 3, 2015 2:35 AM, "Kenneth Graunke" wrote:
>> > Caught by lit_sat.shader_test.
>> >
>> > Signed-off-by: Kenneth Graunke
>> > ---
>> > src/mesa/drivers/dri/i965/brw_fs_nir.cpp |
Reviewed-by: Connor Abbott
On Tue, Feb 3, 2015 at 3:43 PM, Jason Ekstrand wrote:
> ---
> src/glsl/nir/nir_lower_to_source_mods.c | 4
> 1 file changed, 4 insertions(+)
>
> diff --git a/src/glsl/nir/nir_lower_to_source_mods.c
> b/src/glsl/nir/nir_lower_to_source_mods
Reviewed-by: Connor Abbott
Funny, I thought we already did this... whoops.
On Tue, Feb 3, 2015 at 3:43 PM, Jason Ekstrand wrote:
> ---
> src/glsl/nir/nir_validate.c | 8
> 1 file changed, 8 insertions(+)
>
> diff --git a/src/glsl/nir/nir_validate.c b/src/glsl/nir
On Tue, Feb 3, 2015 at 5:20 PM, Eric Anholt wrote:
> Connor Abbott writes:
>
>> I think this logic should be flipped around... you should have a flag
>> 'lower_sub' which is false for vc4 but true for i965, and then if the
>> flag is true we lower sub
While we're cleaning this up, we should go a little farther - we
should also see if anything is using the SSA result of any of these
instructions, and if so make it point to an undef_instr instead to
avoid dangling pointers. This doesn't come up yet but it will when we
start cleaning up constant co
Reviewed-by: Connor Abbott
On Thu, Feb 5, 2015 at 5:28 PM, Jason Ekstrand wrote:
> This is both simpler and more correct. The old code didn't properly index
> load_const instructions.
> ---
> src/glsl/nir/nir.c | 26 --
> 1 file changed, 4 inserti
Other than one small comment below, 3 and 4 are
Reviewed-by: Connor Abbott
On Thu, Feb 5, 2015 at 5:28 PM, Jason Ekstrand wrote:
> ---
> src/glsl/nir/nir_validate.c | 87
> -
> 1 file changed, 79 insertions(+), 8 deletions(-)
>
>
Could use a comment here explaining that nir_instr_prev() may return
NULL above if the only instruction in the block is a jump instruction.
On Thu, Feb 5, 2015 at 5:29 PM, Jason Ekstrand wrote:
> ---
> src/glsl/nir/nir_from_ssa.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff -
This should go before the previous patch, or else that patch won't
work as intended, and also it confused me a little by using a behavior
(the returning NULL) before introducing it.
On Thu, Feb 5, 2015 at 5:29 PM, Jason Ekstrand wrote:
> ---
> src/glsl/nir/nir.h | 28 ++--
Reviewed-by: Connor Abbott
On Thu, Feb 5, 2015 at 5:29 PM, Jason Ekstrand wrote:
> ---
> src/glsl/nir/nir.h | 3 +++
> src/glsl/nir/nir_dominance.c | 9 -
> 2 files changed, 11 insertions(+), 1 deletion(-)
>
> diff --git a/src/glsl/nir/nir.h b/src/glsl
On Thu, Feb 5, 2015 at 5:29 PM, Jason Ekstrand wrote:
> ---
> src/glsl/Makefile.sources | 1 +
> src/glsl/nir/nir.h | 2 +
> src/glsl/nir/nir_opt_gcm.c | 499
> +
> 3 files changed, 502 insertions(+)
> create mode 100644 src/glsl/nir/nir_
Maybe we should run this after DCE so that we don't do unnecessary
work by scheduling stuff that we then delete.
On Thu, Feb 5, 2015 at 5:29 PM, Jason Ekstrand wrote:
> ---
> src/mesa/drivers/dri/i965/brw_fs_nir.cpp | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/src/mesa/drivers/dri/
On the one hand, this really should use the metadata framework since
it is metadata. On the other hand, this would get invalidated *all the
time* so it doesn't seem so useful to have a dirty flag for it. Maybe
add a comment explaining this? Other than that,
Connor Abbott
On Thu, Feb 5, 20
On Thu, Feb 5, 2015 at 9:33 PM, Jason Ekstrand wrote:
>
> On Feb 5, 2015 5:07 PM, "Connor Abbott" wrote:
>>
>> On Thu, Feb 5, 2015 at 5:29 PM, Jason Ekstrand
>> wrote:
>> > ---
>> > src/glsl/Makefile.sources | 1 +
>> > src/gl
On Fri, Feb 6, 2015 at 9:42 AM, Francisco Jerez wrote:
> Right now virtual GRF book-keeping and allocation is performed in each
> visitor class separately (among other hundred different things),
> leading to duplicated logic in each visitor and preventing layering as
> it forces any code that mani
I'd rather keep it the way it was, since that's how it's presented as
pseudocode in the paper and someone reading this is more likely to be
confused by a function with the exact same name as in the paper but a
different implementation.
On Fri, Feb 6, 2015 at 5:12 PM, Jason Ekstrand wrote:
> Befor
; + return child->dom_pre_index >= parent->dom_pre_index &&
> + child->dom_pre_index <= parent->dom_post_index;
It doesn't matter too much, but I think doing child->dom_post_index <=
parent->dom_post_index will make it less likely fo
Reviewed-by: Connor Abbott
On Fri, Feb 6, 2015 at 5:12 PM, Jason Ekstrand wrote:
> ---
> src/glsl/nir/nir_from_ssa.c | 9 ++---
> 1 file changed, 2 insertions(+), 7 deletions(-)
>
> diff --git a/src/glsl/nir/nir_from_ssa.c b/src/glsl/nir/nir_from_ssa.c
> index 2e7ad
On Fri, Feb 6, 2015 at 7:22 PM, Jason Ekstrand wrote:
>
>
> On Fri, Feb 6, 2015 at 5:38 PM, Connor Abbott wrote:
>>
>> On Fri, Feb 6, 2015 at 5:12 PM, Jason Ekstrand
>> wrote:
>> > This is mostly thanks to Connor. The idea is to do a depth-first search
>
Hi Eric,
Sorry for not responding to this in a while. I'm replying to this
comment instead of the new version because it has the relevant
discussion.
On Mon, Feb 2, 2015 at 7:30 PM, Eric Anholt wrote:
> Connor Abbott writes:
>
>> I mentioned this on IRC, but it would be good t
Reviewed-by: Connor Abbott
On Sat, Feb 7, 2015 at 12:16 AM, Eric Anholt wrote:
> No change on shader-db on i965.
> ---
> src/glsl/nir/nir_opt_algebraic.py | 9 +
> 1 file changed, 9 insertions(+)
>
> diff --git a/src/glsl/nir/nir_opt_algebraic.py
> b/src/glsl/nir
On Sat, Feb 7, 2015 at 12:16 AM, Eric Anholt wrote:
> NIR instruction count results on i965:
> total instructions in shared programs: 1261954 -> 1261937 (-0.00%)
> instructions in affected programs: 455 -> 438 (-3.74%)
>
> One in yofrankie, two in tropics. Apparently i965 had also optimized a
Maybe value.f[0] since it's an array and not a pointer? Other than that,
Reviewed-by: Connor Abbott
On Sat, Feb 7, 2015 at 12:16 AM, Eric Anholt wrote:
> We want the size of a float per component, not the size of a whole vec4.
>
> NIR instructions on i965:
> total instr
tests. Note that for
pigliting it might be wise to turn off opt_algebraic since in some cases (but
not all) it will do the same optimization so we don't get a chance to exercise
the code.
This series is also available at
git://people.freedesktop.org/~cwabbott0/mesa nir-reassociate-consts
Connor
expression tree.
Signed-off-by: Connor Abbott
---
src/glsl/nir/nir_search.c | 34 +-
1 file changed, 33 insertions(+), 1 deletion(-)
diff --git a/src/glsl/nir/nir_search.c b/src/glsl/nir/nir_search.c
index 79cfdba..af41578 100644
--- a/src/glsl/nir/nir_search.c
nd @ which we already use, suggestions welcome...) or
'!name@property1,property2'.
Signed-off-by: Connor Abbott
---
src/glsl/nir/nir_algebraic.py | 68 +--
src/glsl/nir/nir_search.c | 40 +
src/glsl/nir/n
ll have to think
carefully to make sure they don't fight.
Signed-off-by: Connor Abbott
---
src/glsl/nir/nir_opt_algebraic.py | 10 ++
1 file changed, 10 insertions(+)
diff --git a/src/glsl/nir/nir_opt_algebraic.py
b/src/glsl/nir/nir_opt_algebraic.py
index a5fe19a..f4466b4 100644
---
Yeah, I find it really surprising that GLSL doesn't have per-compoent
and or, and xor, given that it already has per-component not. I didn't
even realize this was the case until I just looked at the spec. At the
very least, we can relax the restriction as long as st/glsl_to_tgsi
will work. NIR shou
So last time you posted this, I had some suggestions but I realized
that in light of the fact that we want to do value numbering, it
didn't seem like my suggestions were any good. But now that I've
thought about it a little bit, it seems to me like a better plan might
be to pull instructions out of
On Mon, Feb 9, 2015 at 5:04 PM, Jason Ekstrand wrote:
>
>
> On Mon, Feb 9, 2015 at 3:01 PM, Connor Abbott wrote:
>>
>> So last time you posted this, I had some suggestions but I realized
>> that in light of the fact that we want to do value numbering, it
>> did
On Mon, Feb 9, 2015 at 6:11 PM, Jason Ekstrand wrote:
>
>
> On Mon, Feb 9, 2015 at 12:01 PM, Connor Abbott wrote:
>>
>> So last time you posted this, I had some suggestions but I realized
>> that in light of the fact that we want to do value numbering, it
>> did
For some reason, the v2 version of this disappeared from my inbox, so I'll say
Reviewed-by: Connor Abbott
I'm not convinced this is enough for when we optimize constant
if-statement conditions, but I'll leave that be for now.
On Thu, Feb 5, 2015 at 5:28 PM, Jason Ekstrand wrot
On Tue, Feb 10, 2015 at 1:32 PM, Eric Anholt wrote:
> Connor Abbott writes:
>
>> On Sat, Feb 7, 2015 at 12:16 AM, Eric Anholt wrote:
>>> NIR instruction count results on i965:
>>> total instructions in shared programs: 1261954 -> 1261937 (-0.00%)
>>>
Reviewed-by: Connor Abbott
On Tue, Feb 10, 2015 at 11:08 AM, Emil Velikov wrote:
> Cc: "10.5"
> Signed-off-by: Emil Velikov
> ---
> src/glsl/Makefile.sources | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/src/glsl/Makefile.sources b/src/glsl/M
Reviewed-by: Connor Abbott
On Tue, Feb 10, 2015 at 2:24 PM, Matt Turner wrote:
> ---
> src/glsl/nir/glsl_to_nir.cpp | 14 +++---
> 1 file changed, 7 insertions(+), 7 deletions(-)
>
> diff --git a/src/glsl/nir/glsl_to_nir.cpp b/src/glsl/nir/glsl_to_nir.cpp
> inde
Reviewed-by: Connor Abbott
On Tue, Feb 10, 2015 at 2:24 PM, Matt Turner wrote:
> ---
> src/glsl/nir/nir.c | 16
> src/glsl/nir/nir_dominance.c | 6 +++---
> src/glsl/nir/nir_lower_atomics.c | 3 +--
>
On Wed, Feb 11, 2015 at 2:36 PM, Eric Anholt wrote:
> Connor Abbott writes:
>
>> On Tue, Feb 10, 2015 at 1:32 PM, Eric Anholt wrote:
>>> Connor Abbott writes:
>>>
>>>> On Sat, Feb 7, 2015 at 12:16 AM, Eric Anholt wrote:
>>>>> NIR instr
On Wed, Feb 11, 2015 at 3:17 PM, Eric Anholt wrote:
> Connor Abbott writes:
>
>> On Wed, Feb 11, 2015 at 2:36 PM, Eric Anholt wrote:
>>> Connor Abbott writes:
>>>
>>>> On Tue, Feb 10, 2015 at 1:32 PM, Eric Anholt wrote:
>>>>> Connor Abb
Reviewed-by: Connor Abbott
On Mon, Feb 9, 2015 at 11:24 PM, Jason Ekstrand wrote:
> Being able to find the least common anscestor in the dominance tree is a
> useful thing that we may want to do in other passes. In particular, we
> need it for GCM.
>
> v2: Handle NULL inputs b
Given that this commit only changes NIR code, I think it should be
prefixed by "nir: ", e.g. "nir: Drop dependency on mtypes.h in nir.h."
Once that's done, this gets my r-b.
On Wed, Feb 11, 2015 at 7:48 PM, Eric Anholt wrote:
> One less new directory necessary for gallium code that wants to inter
101 - 200 of 1264 matches
Mail list logo
