updating phi sources.
There's one somewhat unrelated fix to stitch_blocks() snuck in, where we
weren't updating the successors correctly if the earlier block had a
jump, but a lot of the logic for it overlaps with the phi node changes
and it seemed hard to split it out so I've left it i
e. In that case, the value that any uses get
is undefined since the instruction never actually executes, so we can
just replace the instruction with an ssa_undef_instr.
v2: split out other fixes
Signed-off-by: Connor Abbott
---
src/glsl/nir/nir.c | 38 --
1
) that I'd rather not expose publically.
v2: use nir_cf_node_remove_after() instead of our own broken thing.
Signed-off-by: Connor Abbott
---
src/glsl/nir/nir.c | 26
src/glsl/nir/nir.h | 7 +++
src/glsl/nir/nir_opt_
Before, we might have left dangling predecessors from jumps that were
going to be removed.
v2: split out from "nir: insert ssa_undef instructions when cleaning up
defs/uses"
Signed-off-by: Connor Abbott
---
src/glsl/nir/nir.c | 10 --
1 file changed, 8 insertions(+), 2
Signed-off-by: Connor Abbott
---
src/glsl/nir/nir.c | 16
src/glsl/nir/nir.h | 2 ++
2 files changed, 18 insertions(+)
diff --git a/src/glsl/nir/nir.c b/src/glsl/nir/nir.c
index 79c4a4a..cbee507 100644
--- a/src/glsl/nir/nir.c
+++ b/src/glsl/nir/nir.c
@@ -2242,6 +2242,22
-by: Connor Abbott
---
src/glsl/nir/nir_opt_dead_cf.c | 121 +
1 file changed, 109 insertions(+), 12 deletions(-)
diff --git a/src/glsl/nir/nir_opt_dead_cf.c b/src/glsl/nir/nir_opt_dead_cf.c
index d71538c..a8d438f 100644
--- a/src/glsl/nir/nir_opt_dead_cf.c
This helps with testing the NIR dead cf pass.
Signed-off-by: Connor Abbott
---
src/glsl/glsl_parser_extras.cpp | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/src/glsl/glsl_parser_extras.cpp b/src/glsl/glsl_parser_extras.cpp
index be6713c..42e9b49 100644
--- a/src/glsl
v2: use nir_cf_node_remove_after().
v2: use foreach_list_typed() instead of hardcoding a list walk.
Signed-off-by: Connor Abbott
---
src/glsl/nir/nir_opt_dead_cf.c | 123 ++---
1 file changed, 115 insertions(+), 8 deletions(-)
diff --git a/src/glsl/nir
We were already doing this internally for iterating over a function
implementation, so just expose it directly.
Signed-off-by: Connor Abbott
---
src/glsl/nir/nir.c | 7 +++
src/glsl/nir/nir.h | 2 ++
2 files changed, 9 insertions(+)
diff --git a/src/glsl/nir/nir.c b/src/glsl/nir/nir.c
Doesn't do anything on the public shader-db.
Signed-off-by: Connor Abbott
---
src/glsl/Makefile.sources | 1 +
src/mesa/drivers/dri/i965/brw_nir.c | 2 ++
2 files changed, 3 insertions(+)
diff --git a/src/glsl/Makefile.sources b/src/glsl/Makefile.sources
index d784a81..7bdd895 1
On Thu, May 21, 2015 at 12:54 PM, Matt Turner wrote:
> On Thu, May 21, 2015 at 9:40 AM, Connor Abbott wrote:
>> Before, when we were deleting a cf node that was a block, we were first
>> removing all the instructions and then calling cleanup_cf_node(), at
>> which
dead, and this patch combined with my nir-dead-cf-v4
branch will cause the loop to be deleted entirely.
No piglit regressions.
Signed-off-by: Connor Abbott
---
src/glsl/nir/nir_opt_remove_phis.c | 20
1 file changed, 20 insertions(+)
diff --git a/src/glsl/nir/nir_opt_remov
On Thu, May 21, 2015 at 8:05 PM, Rogovin, Kevin wrote:
>
>> This line is too long. (It will not fit in 80 columns in git log since git
>> log adds some spaces before each commit message line.)
>
> What is the accepted maximum length for a line in a commit message?
>
>>> - extension table
>>> -
On Thu, May 21, 2015 at 8:40 PM, Rogovin, Kevin wrote:
>
>
>> I suppose it could be useful, but I think we've been mostly successful at
>> just expecting people to recognize when what they're writing doesn't look
>> like the code around it.
>
> This is my point. Older code had different style/ex
On Thu, May 21, 2015 at 8:51 PM, Rogovin, Kevin wrote:
>
>> FWIW, the kernel standards for commit messages are at:
>> https://www.kernel.org/doc/Documentation/SubmittingPatches
>> Most of those rules apply to Mesa too. It says the body should be wrapped to
>> 75 chars (although I've been using 72
t, pooled s = 0.332763)
Signed-off-by: Connor Abbott
---
src/glsl/nir/nir_opt_cse.c | 134 -
1 file changed, 24 insertions(+), 110 deletions(-)
diff --git a/src/glsl/nir/nir_opt_cse.c b/src/glsl/nir/nir_opt_cse.c
index a6a4a21..7894147 100644
--- a/sr
We'll want to use these outside of nir_opt_cse.c. Rather than adding yet
another thing to the mess that is nir.c, create a new file and move
nir_srcs_equal() there too.
Signed-off-by: Connor Abbott
---
src/glsl/Makefile.sources| 1 +
src/glsl/nir/nir.c | 27 --
This can be used for both CSE and value numbering.
Signed-off-by: Connor Abbott
---
src/glsl/Makefile.sources | 2 +
src/glsl/nir/nir_instr_hash.c | 255 ++
src/glsl/nir/nir_instr_hash.h | 36 ++
3 files changed, 293 insertions(+)
create mode
Signed-off-by: Connor Abbott
---
src/glsl/nir/nir.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/src/glsl/nir/nir.h b/src/glsl/nir/nir.h
index f04195f..e3b7b17 100644
--- a/src/glsl/nir/nir.h
+++ b/src/glsl/nir/nir.h
@@ -722,7 +722,7 @@ nir_alu_instr_channel_used
Signed-off-by: Connor Abbott
---
src/glsl/nir/nir.h | 4 ++--
src/glsl/nir/nir_instr_compare.c | 8
2 files changed, 6 insertions(+), 6 deletions(-)
diff --git a/src/glsl/nir/nir.h b/src/glsl/nir/nir.h
index e3b7b17..19b1e18 100644
--- a/src/glsl/nir/nir.h
+++ b/src/glsl
almost all the
time spent in nir_optimize() to less than half (but still a significant
amount).
This series is also available at
git://people.freedesktop.org/~cwabbott0/mesa nir-cse-hash
Connor Abbott (5):
nir: split out instruction comparison functions
nir: constify nir_ssa_alu_instr_src_
On Fri, May 29, 2015 at 6:23 AM, Eero Tamminen
wrote:
> Hi,
>
> On 05/28/2015 10:19 PM, Thomas Helland wrote:
>>
>> One more thing;
>> Is there a limit where the loop body gets so large that we
>> want to decide that "gah, this sucks, no point in unrolling this"?
>> I imagine as the loops get larg
Both are
Reviewed-by: Connor Abbott
On Tue, Jun 2, 2015 at 5:37 AM, Timothy Arceri wrote:
> ---
> src/glsl/nir/glsl_to_nir.cpp | 5 +
> 1 file changed, 1 insertion(+), 4 deletions(-)
>
> diff --git a/src/glsl/nir/glsl_to_nir.cpp b/src/glsl/nir/glsl_to_nir.cpp
> inde
Hi,
Over the past few weeks or so, I've been sitting down and writing some
documentation on how NIR, and especially the core NIR datastructures,
works. At this point I think it's reasonably complete, although I've
probably forgotten something that I should mention, and one thing
that's missing for
time. The old code had it backwards, and so we
would underestimate the total time whenever we thought there would be a
pipeline stall by up to the issue time of the instruction.
Signed-off-by: Connor Abbott
---
I haven't tested this, although it shouldn't affect anything other than the
c
On Tue, Jun 9, 2015 at 3:06 PM, Ian Romanick wrote:
> I've sent out review for all of the patches that don't already have R-b
> on them. Before pushing, you should double check the whitespace. In
> particular, it's hard to notice tabs and trailing whitespace when
> reviewing. If you do 'git sho
On Tue, Sep 23, 2014 at 2:43 PM, Matt Turner wrote:
> On Sat, Sep 20, 2014 at 10:22 AM, Jason Ekstrand wrote:
>> This series does a bunch of refactoring of the i965 fs backend IR to add
>> concepts of register width and instruction execution size. There's more to
>> be done yet, but this gets us
On Wed, Sep 24, 2014 at 7:13 PM, Kristian Høgsberg wrote:
> On Wed, Sep 24, 2014 at 12:57 PM, Kenneth Graunke
> wrote:
>> On Wednesday, September 24, 2014 12:28:20 PM Kristian Høgsberg wrote:
>>
>>> On Skylake, the MOCS bits are an index into a table of 63 different,
>>
>>> configurable cache co
On Fri, Sep 26, 2014 at 9:02 AM, Iago Toral Quiroga wrote:
> Original patch by Petri Latvala :
>
> Add an optimization pass that drops min/max expression operands that
> can be proven to not contribute to the final result. The algorithm is
> similar to alpha-beta pruning on a minmax search, from t
Eliminate minmax expressions operating on constant vectors with mixed values
> by resolving them.
>
> No piglit regressions observed with Version 3.
>
> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=76861
> ---
>
> Version 3 passes all minmax unit tests sent in the or
On Sun, Sep 28, 2014 at 4:26 PM, Matt Turner wrote:
> These checks were intended for Gen 7 only. None of these restrictions
> apply to Gen 8.
> ---
> src/mesa/drivers/dri/i965/brw_fs_visitor.cpp | 6 +++---
> 1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/src/mesa/drivers/dri/i9
On Fri, Oct 3, 2014 at 10:03 PM, Jason Ekstrand wrote:
> Previously, we were allowing the register allocation code to do the
> computation for us in ra_set_finalize. However, the runtime for this
> computation is O(c^4 * g) where c is the number of classes and g is the
> number of GRF registers.
On Tue, Oct 7, 2014 at 4:33 AM, Jason Ekstrand wrote:
> On older GENs in SIMD16 mode, we were accidentally building too much
> interference into our register classes. Since everything is divided by 2,
> the reigster allocator thinks we have 64 base registers instead of 128.
> The actual GRF mappi
On Tue, Oct 14, 2014 at 12:54 AM, Kenneth Graunke wrote:
> From: Connor Abbott
>
> Now that we've made all the texture emit code mostly independent of GLSL
> IR, this isn't necessary any more.
>
> Signed-off-by: Connor Abbott
> Reviewed-by: Kenneth Graunke
>
On Fri, Oct 31, 2014 at 9:27 PM, Matt Turner wrote:
> Three-source instructions on i965 have an annoying property that they
> cannot use immediate operands. They've do have the alluring property
> that they perform multiple operations in basically the same number of
> cycles as any other instructi
On Fri, Oct 31, 2014 at 9:27 PM, Matt Turner wrote:
> ---
> src/mesa/drivers/dri/i965/Makefile.sources | 1 +
> src/mesa/drivers/dri/i965/brw_fs.cpp | 2 +
> src/mesa/drivers/dri/i965/brw_fs.h | 1 +
> .../drivers/dri/i965/brw_fs_combine_constants.cpp
On Sat, Nov 1, 2014 at 5:36 PM, Matt Turner wrote:
> On Fri, Oct 31, 2014 at 7:35 PM, Connor Abbott wrote:
>> On Fri, Oct 31, 2014 at 9:27 PM, Matt Turner wrote:
>>> ---
>>> src/mesa/drivers/dri/i965/Makefile.sources | 1 +
>>> src/mesa/drivers/dri
FWIW, Ian commented on my patch to add foreach_list_typed_reverse that
we shouldn't merge it until it had users so people wouldn't try to
remove it -- it seems like this has the same issue. So either merge
both of them at the same time, or neither until NIR gets merged...
Connor
On Fri, Nov 7, 2
On Thu, Nov 13, 2014 at 7:28 PM, Kristian Høgsberg wrote:
> This will be reused for the scalar VS pass.
>
> Signed-off-by: Kristian Høgsberg
> ---
> src/mesa/drivers/dri/i965/brw_fs.cpp | 132
> +++
> src/mesa/drivers/dri/i965/brw_fs.h | 1 +
> 2 files change
On Tue, Nov 25, 2014 at 1:19 AM, Jason Ekstrand wrote:
> Previously, the hash_table API required the user to do all of the hashing
> of keys as it passed them in. Since the hashing function is intrinsicly
> tied to the comparison function, it makes sense for the hash table to know
> about it. Al
nt struct.
>
> v1 Reviewed-by: Kenneth Graunke
> v1 Reviewed-by: Connor Abbott
>
> v2: Jason Ekstrand :
>Use "(__node)->__field.next != NULL" to check for the end of the list
>instead of the "&__next->__field != NULL". The former is far
On Wed, Mar 11, 2015 at 6:28 PM, Jason Ekstrand wrote:
> I think I mentioned this at least once before, but I don't think that this
> is a good long-term solution. The nir_lower_io function does two things
> primarily. 1) It assigns each variable a "device location" and 2) changes
> the variable
On Fri, Mar 13, 2015 at 6:37 PM, Thomas Helland
wrote:
> This should give better cache locality, less memory consumption,
> and should also be faster since we avoid a modulo operation.
> Also change table size to be power of two.
> This gives better performance as we can do bitmasking instead of
>
On Fri, Mar 13, 2015 at 6:37 PM, Thomas Helland
wrote:
> The same rationale applies here as for the hash table.
> Power of two size should give better performance,
> and using the algorithm hash = sh + i/2 + i*i/2
> should result in only distinct hash values when hitting collisions.
> Should give
On Sun, Mar 15, 2015 at 1:25 AM, Connor Abbott wrote:
> On Fri, Mar 13, 2015 at 6:37 PM, Thomas Helland
> wrote:
>> This should give better cache locality, less memory consumption,
>> and should also be faster since we avoid a modulo operation.
>> Also change table
On Sun, Mar 15, 2015 at 1:50 PM, Thomas Helland
wrote:
> 2015-03-15 16:47 GMT+01:00 Thomas Helland :
>> 2015-03-13 23:37 GMT+01:00 Thomas Helland :
>>> So here comes the second version of this series.
>>> I found a way to exercise the bug in the previous series.
>>> This makes the test fail where
Ok, so I think I managed to find the source of the bug. When inserting
elements into the set/hash table, we computed the initial probe
address *before* doing the rehash. In the case where inserting an
element led to a rehash, this meant that we'd use the wrong starting
address when actually inserti
n false;
>}
> }
>
> @@ -120,7 +125,8 @@ nir_opt_peephole_select_block(nir_block *block, void
> *void_state)
> nir_block *else_block = nir_cf_node_as_block(else_node);
>
> /* ... and those blocks must only contain move-to-phi. */
You should ch
On Mon, Mar 16, 2015 at 10:08 PM, Jason Ekstrand wrote:
> On Mon, Mar 16, 2015 at 6:52 PM, Connor Abbott wrote:
>> On Mon, Mar 16, 2015 at 9:23 PM, Jason Ekstrand wrote:
>>> ---
>>> src/glsl/nir/nir_opt_peephole_select.c | 54
>>> +++---
Instead of doing this, I think it would be better to add something to
nir_shader_compiler_options and do it in nir_opt_algebraic, similar to
what Eric has already done for other things. I've seen some shaders
where we transform a mul + a series of adds into a series of mad's,
which is good on gen6+
On Wed, Mar 18, 2015 at 1:56 AM, Matt Turner wrote:
> Thanks. Looked through stats and at some of the regressions.
>
> Some of the areas I noticed we were doing worse:
>
> We generate two CMPs for discard_if; only one without NIR. I think you
> had an idea about solving this.
>
> SEL peephole inte
On Wed, Mar 18, 2015 at 11:27 AM, Jason Ekstrand wrote:
> On Wed, Mar 18, 2015 at 7:22 AM, Connor Abbott wrote:
>> On Wed, Mar 18, 2015 at 1:56 AM, Matt Turner wrote:
>>> Thanks. Looked through stats and at some of the regressions.
>>>
>>> Some of the
On Wed, Mar 18, 2015 at 2:58 PM, Matt Turner wrote:
> On Thu, Mar 12, 2015 at 9:46 PM, Matt Turner wrote:
>> Here are some ideas I think might be reasonable GSoC ideas.
>>
>> - GLSL linking in NIR
>> - Would allow us to stop doing optimizations and other expensive
>> things on GLSL IR
>
> I
Reviewed-by: Connor Abbott
A case of premature optimization I guess... I think I did it for
linking, but we'll cross that bridge when we come to it.
On Wed, Mar 18, 2015 at 7:45 PM, Jason Ekstrand wrote:
> We never did a single hash table lookup in the entire NIR code base that I
&g
Overall, I think it might be better to do this as a SSA pass right
before lowering to source modifiers. I don't think we would be running
any optimizations after it that depend on the output being all 0's or
all 1's, but if you're concerned about that then we could add separate
gen5-compare compare
So in GLSL, add and mul can take a combination of a scalar and a
vector, and it'll splat the scalar automatically, and with this this
series that means that and/or can now take a combination of a scalar
and a vector as well, and we'll have to splat them as well. Are you
sure that the various backen
On Wed, Mar 18, 2015 at 9:10 PM, Jason Ekstrand wrote:
> On Wed, Mar 18, 2015 at 6:07 PM, Connor Abbott wrote:
>> Overall, I think it might be better to do this as a SSA pass right
>> before lowering to source modifiers. I don't think we would be running
>> any optimiza
c4 "packed" means a different thing
usually. Otherwise, the series is
Reviewed-by: Connor Abbott
On Wed, Mar 18, 2015 at 7:45 PM, Jason Ekstrand wrote:
> Previously, we just assigned variable locations in nir_lower_io. Now, we
> force the user to assign variable locations for u
On Wed, Mar 18, 2015 at 11:24 PM, Kenneth Graunke wrote:
> On Tuesday, March 17, 2015 07:17:19 PM Jason Ekstrand wrote:
>> ---
>> src/mesa/drivers/dri/i965/brw_fs_nir.cpp | 8 +++-
>> 1 file changed, 7 insertions(+), 1 deletion(-)
>>
>> diff --git a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
>
On Wed, Mar 18, 2015 at 11:35 PM, Jason Ekstrand wrote:
>
> On Mar 18, 2015 8:32 PM, "Matt Turner" wrote:
>>
>> On Wed, Mar 18, 2015 at 7:39 PM, Jason Ekstrand
>> wrote:
>> > On Wed, Mar 18, 2015 at 11:37 AM, Matt Turner
>> > wrote:
>> >> Transform this into b2f(and(a, b)).
>> >>
>> >> total in
On Thu, Mar 19, 2015 at 12:35 AM, Jason Ekstrand wrote:
> On Wed, Mar 18, 2015 at 9:31 PM, Matt Turner wrote:
>> On Wed, Mar 18, 2015 at 9:27 PM, Jason Ekstrand wrote:
>>> On Wed, Mar 18, 2015 at 8:59 PM, Connor Abbott wrote:
>>>> On Wed, Mar 18, 2015 at 11:35 P
Fixes a piglit regression
(shaders/glsl-fs-vec4-indexing-temp-dst-in-nested-loop-combined) with
my series for GVN.
Signed-off-by: Connor Abbott
---
I think this particular case might not trigger anymore after Matt's
patch to improve cmod prop since we're better at getting rid of
m
Reviewed-by: Connor Abbott
On Sun, Mar 22, 2015 at 9:49 PM, Jordan Justen
wrote:
> Signed-off-by: Jordan Justen
> Reviewed-by: Chris Forbes
> ---
> src/glsl/nir/glsl_to_nir.cpp | 4 +++-
> src/glsl/nir/nir_intrinsics.h | 1 +
> 2 files changed, 4 insertions(+), 1 deletion
; +
> + case nir_op_imov:
> + case nir_op_inot:
> +if (alu->dest.write_mask == 1) {
> + /* This is a single-source instruction. Just copy the
> + * resolve status from the source.
> +*/
> +
We could already handle vector arguments to these -- we just couldn't
handle mixed scalar and vector arguments. So change the commit message
to something like "nir: Handle mixed scalar/vector arguments to
logical and/or/xor." and fix the typo and then this is
Reviewed-by: Connor
On Mon, Mar 23, 2015 at 8:37 PM, Kenneth Graunke wrote:
> Shamelessly ripped off from Eric Anholt's tgsi_to_nir pass.
>
> Not compiled on SCons, like the rest of NIR.
>
> Signed-off-by: Kenneth Graunke
> ---
> src/mesa/Makefile.am|2 +
> src/mesa/Makefile.sources |
I think this is being both too aggressive and not aggressive enough.
First, the too aggressive part. Imagine if we have something like:
a = mul b, c;
d = add a, 1.0;
e = add a, 2.0;
f = add a, 3.0;
In this case, we don't want to turn this into a series of 3 mad's
since the 3 load immediate instru
We already have a lower_negate compiler option, so I think it would be
better to add the 1 line of code to lower sub -> neg predicated on
!lower_neg to nir_opt_algebraic.
On Mon, Mar 23, 2015 at 8:37 PM, Kenneth Graunke wrote:
> prog_to_nir uses this for OPCODE_SUB.
>
> Signed-off-by: Kenneth Gra
Reviewed-by: Connor Abbott
We should probably use this for Gen4-5 too.
On Wed, Mar 25, 2015 at 5:11 PM, Eric Anholt wrote:
> ---
>
> I missed this one when I was doing my other lowering bits. I don't
> have ffma, and I certainly don't have flrp.
>
> src/gallium/d
On Wed, Mar 25, 2015 at 7:21 PM, Kenneth Graunke wrote:
> Everything is already in place; we simply have to take the scalar code
> generation path. This gives us SIMD8 VS programs, instead of SIMD4x2.
>
> v2: Rebase on the patch that drops brw->gen >= 8.
>
> Signed-off-by: Kenneth Graunke
> Revi
Except for one comment on patch 6, patches 1-2 and 4-6 are
Reviewed-by: Connor Abbott
I don't know enough about Mesa IR to fully review patch 3, although on
a quick read-through I couldn't find anything to improve.
On Wed, Mar 25, 2015 at 7:21 PM, Kenneth Graunke wrote:
> Both
+), 153 deletions(-)
>
> --
> 2.3.4
>
> ___
> mesa-dev mailing list
> mesa-dev@lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev
I don't see any performance data on each commit, did
On Sun, Mar 29, 2015 at 5:54 PM, Thomas Helland
wrote:
> Does anyone have suggestions for good literature on compilers?
> Since GPU's and CPU's are a bit different there are probably
> books that are better suited than others for GPUs?
> I have what is probably Norway's biggest library on the subj
On Sun, Mar 29, 2015 at 9:51 PM, Connor Abbott wrote:
> On Sun, Mar 29, 2015 at 5:54 PM, Thomas Helland
> wrote:
>> Does anyone have suggestions for good literature on compilers?
>> Since GPU's and CPU's are a bit different there are probably
>> books that are
Series is
Reviewed-by: Connor Abbott
On Mon, Mar 30, 2015 at 1:12 PM, Eric Anholt wrote:
> Since we have patterns based on b2f, generate them if we see the b2f
> equivalent using an iand. This is common when generating NIR from TGSI.
> ---
> src/glsl/nir/nir_opt_algebraic.py | 1
On Tue, Mar 31, 2015 at 6:57 PM, Rob Clark wrote:
> From: Rob Clark
>
> Based on the algo from NV50LegalizeSSA::handleDIV() and handleMOD().
> See also trans_idiv() in freedreno/ir3/ir3_compiler.c (which was an
> adaptation of the nv50 code from Ilia).
>
> Just sending as an rfc right now, since
Unfortunately, we can't support unstructured COME FROM yet, since we
can't structurize arbitrary control flow graphs. Also TODO is adding
support for threading by having multiple come_from's point to the same
unconditional branch, as well as various other INTERCAL features.
Signed
Unfortunately, we can't support unstructured COME FROM yet, since we
can't structurize arbitrary control flow graphs. Also TODO is adding
support for threading by having multiple come_from's point to the same
unconditional branch, as well as various other INTERCAL features.
Signed
On Wed, Apr 1, 2015 at 2:53 PM, Brian Paul wrote:
> On 04/01/2015 12:49 PM, Brian Paul wrote:
>>
>> On 04/01/2015 12:02 PM, Eric Anholt wrote:
>>>
>>> Module: Mesa
>>> Branch: master
>>> Commit: 783ad697d25e754ab719ab6c715969c35dbe867b
>>> URL:
>>>
>>> https://urldefense.proofpoint.com/v2/url?u=ht
On Wed, Apr 1, 2015 at 4:44 PM, Eric Anholt wrote:
> Jason Ekstrand writes:
>
>> Module: Mesa
>> Branch: master
>> Commit: 793a94d6b5fc589ca8d7475347def4e222cd3d7c
>> URL:
>> http://cgit.freedesktop.org/mesa/mesa/commit/?id=793a94d6b5fc589ca8d7475347def4e222cd3d7c
>>
>> Author: Jason Ekstrand
On Wed, Apr 1, 2015 at 6:49 PM, Jason Ekstrand wrote:
> On Tue, Mar 31, 2015 at 4:10 PM, Ilia Mirkin wrote:
>> On Tue, Mar 31, 2015 at 6:57 PM, Rob Clark wrote:
>>> From: Rob Clark
>>>
>>> Signed-off-by: Rob Clark
>>> ---
>>> src/glsl/nir/nir_opcodes.py | 4 ++--
>>> 1 file changed, 2 inserti
I think it might be better here if we had a callback that backends can
fill in that tells you when an instruction can be pulled out by the
sel peephole. As Jason noted, you won't be able to do this for
everything (notably, output writes and variable writes) and we'll
probably need special handling
Reviewed-by: Connor Abbott
On Wed, Apr 1, 2015 at 7:17 PM, Jason Ekstrand wrote:
> ---
> src/glsl/nir/nir_print.c | 24
> 1 file changed, 12 insertions(+), 12 deletions(-)
>
> diff --git a/src/glsl/nir/nir_print.c b/src/glsl/nir/nir_print.c
> inde
On Wed, Apr 1, 2015 at 8:34 PM, Jason Ekstrand wrote:
> On Wed, Apr 1, 2015 at 9:59 AM, Francisco Jerez wrote:
>> Connor Abbott writes:
>>
>>> Unfortunately, we can't support unstructured COME FROM yet, since we
>>> can't structurize arbitra
On Thu, Apr 2, 2015 at 1:32 PM, Rob Clark wrote:
> On Thu, Apr 2, 2015 at 1:24 PM, Jason Ekstrand wrote:
>> On Thu, Apr 2, 2015 at 1:20 AM, Kenneth Graunke
>> wrote:
>>> On Wednesday, April 01, 2015 07:22:40 PM Connor Abbott wrote:
>>>> I think it might be b
Wow, that's pretty embarrassing... it's tough to test this code, so
I'm not too surprised it's this busted.
Reviewed-by: Connor Abbott
On Thu, Apr 2, 2015 at 2:35 PM, Rob Clark wrote:
> From: Rob Clark
>
> v2: discovered that i2b/b2i are also confused
On Fri, Apr 3, 2015 at 2:07 PM, Rob Clark wrote:
> From: Rob Clark
>
> If we want to use NIR from state trackers that don't already pull in the
> whole of glsl (ie. anything other than mesa state tracker), we need a
> separate more minimal libnir. Possibly NIR should be better split out
> from g
Hi Thomas,
Thanks for submitting a proposal! Some comments/answers below.
On Tue, Apr 7, 2015 at 3:34 PM, Thomas Helland
wrote:
> Hi,
>
> For those that don't know I've submitted a proposal for this years GSoC.
> I've proposed to implement value range propagation and loop unrolling in
> NIR.
> S
> Yes, copy propagation probably won't be so useful once we have value
> range propagation; the former is a special case of the latter.
Err, I meant constant propagation...
___
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop
On Sat, Apr 11, 2015 at 3:12 PM, Thomas Helland
wrote:
>> Yes, copy propagation probably won't be so useful once we have value
>> range propagation; the former is a special case of the latter. Note
>> that we have a nifty way of actually doing the constant folding
>> (nir_constant_expressions.py a
On Sat, Apr 11, 2015 at 2:13 PM, Matt Turner wrote:
> Readding mesa-dev@ to Cc.
>
> On Sat, Apr 11, 2015 at 11:06 AM, Darius Goad wrote:
>> Well, I did make a v2, but all that did was fix a bug. How does this not
>> come close?
>
> Well, for instance the i965 driver supports ~5 generations of
> h
Patches 1 and 2 are
Reviewed-by: Connor Abbott
Adding stricter assertions for tex sources makes sense to me now after
some thought/discussion, but if we're going to do it then we should go
all the way and make intrinsic srcs and dests and if srcs more strict
as well. Also, maybe you&#x
On Tue, Apr 14, 2015 at 11:43 AM, Ian Romanick wrote:
> From: Ian Romanick
>
> Shader-db results:
>
> GM45 NIR:
> total instructions in shared programs: 4082044 -> 4081919 (-0.00%)
> instructions in affected programs: 27609 -> 27484 (-0.45%)
> helped:44
>
> Iro
We weren't comparing the right number of components when checking
swizzles. Use nir_ssa_alu_instr_num_src_components() to do the right
thing.
I haven't piglited this yet, but it seems straightforward.
Cc: Ian Romanick
Signed-off-by: Connor Abbott
---
src/glsl/nir/nir_opt_
On Tue, Apr 14, 2015 at 3:55 PM, Jason Ekstrand wrote:
> On Tue, Apr 14, 2015 at 12:51 PM, Connor Abbott wrote:
>> We weren't comparing the right number of components when checking
>> swizzles. Use nir_ssa_alu_instr_num_src_components() to do the right
>
> I don't
On Tue, Apr 14, 2015 at 4:56 PM, Jason Ekstrand wrote:
> On Tue, Apr 14, 2015 at 1:39 PM, Connor Abbott wrote:
>> On Tue, Apr 14, 2015 at 3:55 PM, Jason Ekstrand wrote:
>>> On Tue, Apr 14, 2015 at 12:51 PM, Connor Abbott wrote:
>>>> We weren't comparing
On Tue, Apr 14, 2015 at 5:16 PM, Rob Clark wrote:
> On Tue, Apr 14, 2015 at 4:59 PM, Jason Ekstrand wrote:
>>> + /**
>>> +* Addressing mode for corresponding _indirect intrinsics:
>>> +*/
>>> + nir_addressing_mode var_addressing_mode;
>>> + nir_addressing_mode in
On Tue, Apr 14, 2015 at 7:08 PM, Rob Clark wrote:
> On Tue, Apr 14, 2015 at 6:24 PM, Connor Abbott wrote:
>> On Tue, Apr 14, 2015 at 5:16 PM, Rob Clark wrote:
>>> On Tue, Apr 14, 2015 at 4:59 PM, Jason Ekstrand
>>> wrote:
>>>>>>>>> +
Reviewed-by: Connor Abbott
On Wed, Apr 15, 2015 at 9:18 PM, Ian Romanick wrote:
> From: Ian Romanick
>
> Suggested by Jason on a different patch after some comments /
> questions by Ilia.
>
> Signed-off-by: Ian Romanick
> Cc: Jason Ekstrand
> ---
> src/glsl/nir
Signed-off-by: Connor Abbott
---
src/mesa/main/.gitignore | 1 +
1 file changed, 1 insertion(+)
diff --git a/src/mesa/main/.gitignore b/src/mesa/main/.gitignore
index 8256ad7..b5f786d 100644
--- a/src/mesa/main/.gitignore
+++ b/src/mesa/main/.gitignore
@@ -9,3 +9,4 @@ get_hash.h.tmp
701 - 800 of 1264 matches
Mail list logo
