g along these lines or rather just
have it per draw. Not sure which one would have less overhead neither...
You should get some more feedback from other driver's authors but otherwise
Reviewed-by: Roland Scheidegger <srol...@vmware.com>
___
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
Am 12.04.2016 um 00:30 schrieb Brian Paul:
> Instead of using an array indexed by SYSTEM_VALUE_x, just use a
> switch statement. This fixes a regression caused by inserting new
> SYSTEM_VALUE_ enums but not updating the mapping to TGSI semantics.
> ---
>
Am 11.04.2016 um 23:50 schrieb Jose Fonseca:
> On 25/03/16 23:12, Jason Ekstrand wrote:
>> These weren't added before because they are actually calculated values
>> that
>> are computed from other inputs. However, in order to handle them in
>> nir_lower_system_values, it's nice for them to have a
Am 11.04.2016 um 21:05 schrieb Matt Turner:
> This will make adding SSE2 code a lot cleaner.
> ---
> src/mesa/drivers/dri/i965/intel_tiled_memcpy.c | 42
> +++---
> 1 file changed, 24 insertions(+), 18 deletions(-)
>
> diff --git
es -= 4;
> }
> return dst;
>
FWIW mesa core has truckloads of similar per-byte load/swap/store
functions for format conversions, which could be optimized similarly (or
be done, hint hint, with sse2...).
But nonethelss, looking good here.
For the series (with some issue mentioned for 2
SYSTEM_VALUE_INSTANCE_INDEX here.
(fwiw I find the names confusing - we have:
SYSTEM_VALUE_VERTEX_ID - includes baseindex
SYSTEM_VALUE_VERTEX_ID_ZERO_BASE - without baseindex
SYSTEM_VALUE_INSTANCE_ID - without baseinstance
SYSTEM_VALUE_INSTANCE_INDEX - includes baseinsta
I suspect you simply ran into an issue with old mesa and new llvm.
Unfortunately some usage of the c++ abi of llvm is unavoidable, and the
abi is unstable, thus breaking quite often. And we do not generally
backport such fixes to older mesa releases (if only mostly because
nobodoy would actually
Am 21.03.2016 um 03:08 schrieb Dave Airlie:
> This appears to set some values into some temp register that we never
> read from or look at again,
>
> should it be setting something that ExecMask gets set to at machine run time?
>
> just noticed it while trying to work out the various mask for
ry?
>
> Cheers
> Jason
>
> On 17/03/2016 12:36 pm, "Jason Anderssen" <janders...@exactal.com> wrote:
>
>> Mesa version is 11.1.2
>> LVM - 3.4
>> Ill see what I can do regarding a gdb backtrace.
>>
>> Cheers
>> Jason
>>
&g
Not that I really care what you can or can't build (and I won't comment
on build changes), what are those features lacking in llvmpipe, beside
from anisotropic filtering (which I always considered essentially
useless for a software renderer, albeit interesting if you're curious
about the math
Ah nice, I wasn't aware this would now work.
For the series:
Reviewed-by: Roland Scheidegger <srol...@vmware.com>
Am 02.04.2016 um 16:13 schrieb Jose Fonseca:
> Only provide a fallback for LLVM 3.3.
>
> One less dependency on LLVM C++ interface.
> ---
> src/gallium/auxil
gestions if anyone has a better idea. But right
> now this, plus maybe eventually ripping out the old way, seems like
> the best plan.
>
> BR,
> -R
>
> On Fri, Apr 1, 2016 at 8:03 PM, Roland Scheidegger <srol...@vmware.com> wrote:
>> I'll admit I'm not an expert on this but
Am 03.04.2016 um 10:11 schrieb Axel Davy:
> d3d 9 needs COLOR0 to be 1.0 on all channels when
> undefined. 0.0 for the others is fine.
> GL behaviour is undefined.
>
> Signed-off-by: Axel Davy
> ---
> src/gallium/drivers/r600/r600_state.c | 4
> 1 file changed, 4
t; + util_snprintf(intrinsic, sizeof intrinsic, "llvm.fabs.v%uf%u",
> type.length, type.width);
> + return lp_build_intrinsic_unary(builder, intrinsic, vec_type, a);
> }
>
> if(type.width*type.length == 128 && util_cpu_caps.has_ssse3) {
>
Looks like fabs was ne
}
> +
> + util_snprintf(intrinsic, sizeof intrinsic, "%s.v%uf%u",
> +intrinsic_root, type.length, type.width);
> +
> + return lp_build_intrinsic_unary(builder, intrinsic, bld->vec_type, a);
> + }
> else /* (util_cpu_caps.has_altivec)
Am 03.04.2016 um 02:31 schrieb Rob Clark:
> On Sat, Apr 2, 2016 at 7:55 PM, Roland Scheidegger <srol...@vmware.com> wrote:
>> I don't really have anything against the new interface (though don't
>> really feel qualified for a proper review), but it is indeed the
>> exis
Am 03.04.2016 um 17:05 schrieb Rob Clark:
> On Sat, Apr 2, 2016 at 8:31 PM, Rob Clark <robdcl...@gmail.com> wrote:
>> On Sat, Apr 2, 2016 at 7:55 PM, Roland Scheidegger <srol...@vmware.com>
>> wrote:
>>> I don't really have anything against the new int
ize, "%s.%c%u", name_root, c, width);
> + }
> +}
> +
> +
> LLVMValueRef
> lp_declare_intrinsic(LLVMModuleRef module,
> const char *name,
> diff --git a/src/gallium/auxiliary/gallivm/lp_bld_intr.h
> b/src/
Am 04.04.2016 um 14:04 schrieb Axel Davy:
> On 03/04/2016 20:41, Roland Scheidegger wrote :
>> Am 03.04.2016 um 10:11 schrieb Axel Davy:
>>> d3d 9 needs COLOR0 to be 1.0 on all channels when
>>> undefined. 0.0 for the others is fine.
>>> GL behaviour is un
Am 29.03.2016 um 22:39 schrieb Jose Fonseca:
> On 29/03/16 21:22, Roland Scheidegger wrote:
>> I actually think "is_tex" is a bit confusing (because there's lots of
>> other parts in the code where we don't really distinguish between
>> "texture" and
ally it
actually meant there'd be a texture token but we have that with sample
instructions as well...)
Roland
For the series:
Reviewed-by: Roland Scheidegger <srol...@vmware.com>
Am 29.03.2016 um 21:41 schrieb Brian Paul:
> Texture sample instructions specify a sampler unit and texture t
l_context *ctx,
> - struct gl_texture_object *texObj, GLuint level,
> - GLsizei width, GLsizei height, GLsizei depth,
> - GLsizei border, GLenum intFormat, mesa_format
> format);
> -
> void
>
nsigned num_coords = tgsi_util_get_texture_coord_dim(target, );
> + unsigned num_coords = tgsi_util_get_texture_coord_dim(target);
> LLVMValueRef coords[4];
> LLVMValueRef tmp;
> int chan;
> @@ -3387,8 +3386,8 @@ static void tex_fetch_args(
> unsigned target = inst->Texture.Texture;
> LLVMValueRef coords[5], derivs[6];
> LLVMValueRef address[16];
> - int ref_pos;
> - unsigned num_coords = tgsi_util_get_texture_coord_dim(target, _pos);
> + unsigned num_coords = tgsi_util_get_texture_coord_dim(target);
> + int ref_pos = tgsi_util_get_shadow_ref_src_index(target);
> unsigned count = 0;
> unsigned chan;
> unsigned num_deriv_channels = 0;
>
Reviewed-by: Roland Scheidegger <srol...@vmware.com>
___
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
Am 31.03.2016 um 04:17 schrieb Ian Romanick:
> Arg. :(
>
> It turns out that shaders in the game PlaneShift use float suffixes in
> GLSL 1.10 shaders... at least the version of the shaders in our private
> shader-db repo do. Once some one check to make sure the game still does
> this, we will
r llvm versions had
tendencies to lower things to scalars when things weren't directly
doable with vector operations...).
Hopefully it works on non-x86 backends too but we can't test this, so
I'd say we just wait for the bug reports ;-).
Reviewed-by: Roland Scheidegger <srol...@vmware.com>
_
I'll admit I'm not an expert on this but I got a bad feeling on it.
Do you really need another per-context fence_finish function?
This looks to me like rather than improving the existing api, it throws
another one at the same problem, which is to be sort of used in parallel
making things confusing
Am 04.04.2016 um 17:27 schrieb Axel Davy:
> Le 04/04/2016 15:27, Roland Scheidegger a écrit :
>> Am 04.04.2016 um 14:04 schrieb Axel Davy:
>>> On 03/04/2016 20:41, Roland Scheidegger wrote :
>>>> Am 03.04.2016 um 10:11 schrieb Axel Davy:
>>>>> d3d
tLevel);
> - }
> -
> - dstImage = _mesa_get_tex_image(ctx, texObj, target, dstLevel);
> - if (!dstImage) {
> - _mesa_error(ctx, GL_OUT_OF_MEMORY, "generating mipmaps");
> - return;
> - }
> -
> - /* Free old image data */
Am 13.04.2016 um 12:12 schrieb Jose Fonseca:
> On 13/04/16 06:51, Jose Fonseca wrote:
>> On 13/04/16 04:00, srol...@vmware.com wrote:
>>> From: Roland Scheidegger <srol...@vmware.com>
>>>
>>> We used to use sse roundps intrinsic directly, but s
Am 13.04.2016 um 03:35 schrieb eocallag...@alterapraxis.com:
> On 2016-04-11 22:27, Roland Scheidegger wrote:
>> Am 10.04.2016 um 09:41 schrieb Edward O'Callaghan:
>>> All the piglits pass for these two as-is. However, some of the piglits
>>> require SSBO support to ru
char intrinsic[32];
> + lp_format_intrinsic(intrinsic, sizeof intrinsic, "llvm.fabs",
> vec_type);
> + return lp_build_intrinsic_unary(builder, intrinsic, vec_type, a);
> + }
> }
>
> if(type.width*type.length == 128 && util_cpu_caps.has_sss
ap to use Nans? We need to
> use BGRA8 and something else (e.g., RGBA32F) to covert both AoS/SoA paths.
Sounds like a good idea (I suppose though it's really only interesting
for sw based drivers).
Roland
>
> Jose
>
>
>
> On 22/04/16 14:33, srol...@vmware.com wrote:
&g
");
> }
> yylval->real = _mesa_strtof(yytext, NULL);
> return FLOATCONSTANT;
>
Sine I "voted" for warning instead of making it a driconf option, looks
fine to me.
Reviewed-by:
I don't quite understand why this is necessary.
Couldn't you just handle such failures in the driver easily? I can't
quite see why informing the state tracker of it really helps.
Roland
Am 20.04.2016 um 17:43 schrieb Nicolai Hähnle:
> From: Nicolai Hähnle
>
> Even when
linedocs/gcc-4.3.6/gcc/Other-Builtins.html */
> +# define HAVE___BUILTIN_BSWAP32 1
> +# define HAVE___BUILTIN_BSWAP64 1
> +#endif
> +
> +#if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 5)
> +# define HAVE___BUILTIN_UNREACHABLE 1
> +#endif
&g
Am 22.04.2016 um 18:22 schrieb Nicolai Hähnle:
> On 22.04.2016 08:56, Roland Scheidegger wrote:
>> I don't quite understand why this is necessary.
>> Couldn't you just handle such failures in the driver easily?
>
> How? When we need a new query buffer due to
Am 26.04.2016 um 23:18 schrieb Dave Airlie:
> On 27 April 2016 at 06:07, Roland Scheidegger <srol...@vmware.com> wrote:
>> Am 26.04.2016 um 06:42 schrieb Dave Airlie:
>>> From: Dave Airlie <airl...@redhat.com>
>>>
>>> This enables ARB_
Am 27.04.2016 um 18:13 schrieb Jose Fonseca:
> On 27/04/16 02:46, Roland Scheidegger wrote:
>> Am 27.04.2016 um 03:05 schrieb Dave Airlie:
>>> On 27 April 2016 at 11:00, Dave Airlie <airl...@gmail.com> wrote:
>>>>>> So far I've set the execm
Am 27.04.2016 um 22:45 schrieb Jose Fonseca:
> On 27/04/16 18:48, Roland Scheidegger wrote:
>> Am 27.04.2016 um 18:13 schrieb Jose Fonseca:
>>> On 27/04/16 02:46, Roland Scheidegger wrote:
>>>> Am 27.04.2016 um 03:05 schrieb Dave Airlie:
>>>>>
gt; #define SP_MAX_TEXTURE_CUBE_LEVELS 13 /* 4K x 4K */
>
>
>
d3d10 actually requiers 2048 too.
For the series:
Reviewed-by: Roland Scheidegger <srol...@vmware.com>
___
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
Am 27.04.2016 um 03:05 schrieb Dave Airlie:
> On 27 April 2016 at 11:00, Dave Airlie wrote:
So far I've set the execmask to 1 active channel, I'm contemplating
changing that
though and using less machines.
>>> Ah yes, I think that would indeed be desirable.
>>
>>
Am 28.04.2016 um 17:52 schrieb Chih-Wei Huang:
> I clearly understand this point.
> Actually I've spent several days to try to
> make it work on the master branch.
> That's why it was delayed -- I supposed to send them
> in the last week.
> > However, the master branch is always broken for
Am 22.04.2016 um 19:28 schrieb Nicolai Hähnle:
> On 22.04.2016 11:52, Roland Scheidegger wrote:
>> Am 22.04.2016 um 18:22 schrieb Nicolai Hähnle:
>>> On 22.04.2016 08:56, Roland Scheidegger wrote:
>>>> I don't quite understand why this is necessary.
>>>&
Am 24.05.2016 um 08:41 schrieb mathias.froehl...@gmx.net:
> From: Mathias Fröhlich
>
> The aim is to replace the CoordReplace array by
> a bitfield. Until all drivers are converted,
> establish the bitfield in paralell to the
> CoordReplace array.
>
> Signed-off-by:
Typically, if you expose the cap bits so that support for some extension
is announced you'd consider it done.
That does not necessarily mean that all the corresponding piglit tests
are passing, but of course you should generally not announce support for
features which don't really work (but you
if (!os_get_total_physical_memory(_memory))
> return 0;
>
> +#ifdef PIPE_ARCH_X86
> + /* cap to 2 GB on 32 bits system */
> + system_memory = MIN2(system_memory, 2048);
> +#endif
> +
>return (int)(system_memory >> 20);
> }
>
Am 17.05.2016 um 20:16 schrieb Rob Herring:
> On Tue, May 17, 2016 at 11:35 AM, Axel Davy wrote:
>> On 16/05/2016 23:28, Rob Herring wrote:
>>>
>>> On Sun, May 15, 2016 at 5:45 AM, Axel Davy wrote:
And cap to 2 GB on 32 bits.
Fixes
On 05/14/2016 04:24 PM, Ilia Mirkin wrote:
On Sat, May 14, 2016 at 10:23 AM, Roland Scheidegger <srol...@vmware.com> wrote:
Am 14.05.2016 um 14:55 schrieb Marek Olšák:
Dave,
It should be noted that clip distances can be disabled by
pipe_rasterizer_state::clip_plane_enable, but cull dis
gt; Our hardware only has 2 vec4 outputs. Each component can be configured
> to be "clip distance", "cull distance", or "disabled" independently.
>
Ok.
Roland
>
> On Sat, May 14, 2016 at 12:43 AM, Roland Scheidegger <srol...@vmware.com>
> wro
Am 13.05.2016 um 06:41 schrieb Dave Airlie:
> This is just the core patches, as I think the lowering was pretty
> broken in the last couple of reposts.
>
> The lowering now lowers to one array of 8 or whatever. I need
> to recheck the gallium and llvmpipe bits on top of this, as I think
>
my
opinion is since we generally don't allow different semantics within the
same reg, I see no good reason why we allow it here (and clip dists and
cull dists, albeit somewhat similar, are still different). So, if some
drivers wanted it in different regs and some in the same regs, I'd
prefer it to be dif
Am 13.05.2016 um 23:12 schrieb Dave Airlie:
> On 14 May 2016 at 01:12, Roland Scheidegger <srol...@vmware.com> wrote:
>> Am 13.05.2016 um 06:41 schrieb Dave Airlie:
>>> This is just the core patches, as I think the lowering was pretty
>>> broken in the last couple
ITEventListener *JEL = JITEventListener::createIntelJITEventListener();
>> + JIT->RegisterJITEventListener(JEL);
>> +#endif
>>if (JIT) {
>> *OutJIT = wrap(JIT);
>> return 0;
>> --
>> 1.9.1
>>
>
I know nothing about how that interface works, but looks reasonable
enough to me.
Acked-by: Roland Scheidegger <srol...@vmware.com>
___
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
Am 10.05.2016 um 06:50 schrieb Jose Fonseca:
> On 09/05/16 18:50, srol...@vmware.com wrote:
>> From: Roland Scheidegger <srol...@vmware.com>
>>
>> Those aren't really interesting, however outputting them is helpful when
>> trying to feed the IR to llvm llc (or op
Am 14.04.2016 um 21:53 schrieb Tim Rowley:
> ---
> .../drivers/swr/rasterizer/common/simdintrin.h | 51
> --
> src/gallium/drivers/swr/rasterizer/core/frontend.h | 12 +
> src/gallium/drivers/swr/rasterizer/core/state.h| 2 -
> 3 files changed, 59 insertions(+),
>
> @@ -793,6 +797,8 @@ st_DrawAtlasBitmaps(struct gl_context *ctx,
>
> pipe_resource_reference(, NULL);
>
> + pipe_sampler_view_reference(, NULL);
> +
> /* We uploaded modified constants, need to invalidate them. */
> st->dirty.mesa |= _NEW_PROGRAM_CONSTANTS;
> }
>
I don't thi
Am 14.04.2016 um 14:18 schrieb Oded Gabbay:
> This patch adds a new field, called "endian_format", to
> "struct pipe_resource". The new field is of type "enum pipe_endian" and
> can receive one of two values:
> - PIPE_ENDIAN_LITTLE
> - PIPE_ENDIAN_NATIVE
>
> PIPE_ENDIAN_NATIVE is initialized to
Am 18.04.2016 um 04:49 schrieb Dave Airlie:
> From: Dave Airlie
>
> This is a first attempt at adding support for dynamic indexing
> of samplers to llvmpipe. This is needed for ARB_gpu_shader5 support.
>
> This uses the sampler function generator to generate functions
> for
lvmpipe/lp_tex_sample.c
> @@ -91,7 +91,7 @@ static LLVMValueRef
> lp_llvm_texture_member(const struct lp_sampler_dynamic_state *base,
> struct gallivm_state *gallivm,
> LLVMValueRef context_ptr,
> - unsigned t
Makes me wonder why we didn't use enums in the first place.
But I can't think of any disadvantages.
Reviewed-by: Roland Scheidegger <srol...@vmware.com>
Am 16.04.2016 um 14:50 schrieb Marek Olšák:
> From: Marek Olšák <marek.ol...@amd.com>
>
> and remove number assignments
Am 16.04.2016 um 14:50 schrieb Marek Olšák:
> From: Marek Olšák
>
> we should use MESA_SHADER_* everywhere, but we're not ready for that yet
> ---
I think the idea initially was that tgsi is essentially self-contained:
you can parse the token stream without any "external"
Reviewed-by: Roland Scheidegger <srol...@vmware.com>
Am 16.04.2016 um 14:50 schrieb Marek Olšák:
> From: Marek Olšák <marek.ol...@amd.com>
>
> ---
> src/gallium/include/pipe/p_shader_tokens.h | 303
> -
> 1 file changed, 164 insertions(
Am 16.04.2016 um 14:50 schrieb Marek Olšák:
> From: Marek Olšák
>
> Use PIPE_SWIZZLE_* everywhere.
> Use X/Y/Z/W/0/1 instead of RED, GREEN, BLUE, ALPHA, ZERO, ONE.
> The new enum is called pipe_swizzle.
Using PIPE_SWIZZLE insteads sounds reasonable.
Not sure though how I
Am 16.04.2016 um 15:19 schrieb eocallag...@alterapraxis.com:
> On 2016-04-16 20:20, Marek Olšák wrote:
>> On Sat, Apr 16, 2016 at 8:04 AM, Michel Dänzer
>> wrote:
>>> On 16.04.2016 14:51, Michel Dänzer wrote:
On 16.04.2016 11:39, Tom Stellard wrote:
> The ds_bpermute
Am 17.04.2016 um 11:06 schrieb Jose Fonseca:
> On 16/04/16 22:30, srol...@vmware.com wrote:
>> From: Roland Scheidegger <srol...@vmware.com>
>>
>> llvm 3.7 sometimes simply miscompiles vector selects.
>
> Reviewed-by: Jose Fonseca <jfons...@vmware.com>
Am 18.04.2016 um 00:31 schrieb Emil Velikov:
> Hi gents,
>
> On 17 April 2016 at 10:06, Jose Fonseca <jfons...@vmware.com> wrote:
>> On 16/04/16 22:30, srol...@vmware.com wrote:
>>>
>>> From: Roland Scheidegger <srol...@vmware.com>
>>>
&g
Series looks great to me. Thanks for the cleanup.
Reviewed-by: Roland Scheidegger <srol...@vmware.com>
Am 18.04.2016 um 11:14 schrieb Jose Fonseca:
> And llvm::raw_string_ostream where not (LLVM 3.3).
>
> Thereby eliminating yet another dependency on unstable LLVM interfaces.
unsigned target,
> -LLVMValueRef context_ptr,
> -boolean is_sviewinfo,
> -enum lp_sampler_lod_property lod_property,
> -LLVMValueRef explicit_lod,
> -
ks good to me, though I'm not really all that familiar with
any of it.
Reviewed-by: Roland Scheidegger <srol...@vmware.com>
___
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
Am 18.04.2016 um 23:10 schrieb Dave Airlie:
> On 19 April 2016 at 03:40, Roland Scheidegger <srol...@vmware.com> wrote:
>> Am 18.04.2016 um 04:49 schrieb Dave Airlie:
>>> From: Dave Airlie <airl...@redhat.com>
>>>
>>> This is a first attempt at add
Am 19.04.2016 um 00:43 schrieb Ian Romanick:
> On 04/18/2016 02:05 PM, srol...@vmware.com wrote:
>> From: Roland Scheidegger <srol...@vmware.com>
>>
>> This was part of EXT_gpu_shader4 - as such it should have been supported
>> by glsl 130.
>> It was however
Am 13.04.2016 um 22:51 schrieb Christian Gmeiner:
> Signed-off-by: Christian Gmeiner
> ---
> src/gallium/auxiliary/tgsi/tgsi_lowering.c | 54
> ++
> src/gallium/auxiliary/tgsi/tgsi_lowering.h | 1 +
> 2 files changed, 55 insertions(+)
>
t pipe_screen *_screen,
> + enum pipe_shader_ir ir_type,
> + enum pipe_compute_cap param,
> + void *ret)
> +{
> + switch (param) {
> + case PIPE_COMPUTE_CAP_IR_TARGET:
> +
Am 12.07.2016 um 22:52 schrieb Marek Olšák:
> From: Marek Olšák
>
> Not sure if this is the right way to do it, but it seems to work.
> ---
> src/gallium/auxiliary/gallivm/lp_bld_misc.cpp | 9 +
> src/gallium/auxiliary/gallivm/lp_bld_misc.h | 3 +++
> 2 files
Am 14.07.2016 um 15:25 schrieb Rob Conde:
> Hello,
>
>It occurred to me that the llvmpipe might be a good basis for a GLSL
> debugger. Has anyone thought about this and how it might be approached?
>
>
> Rob Conde
I'm not entirely sure what exactly you want to debug, but if you just
want to
ons that looks good (plus, buffers and textures
are quite different indeed). (We'll have some code to adapt of course,
but this should be trivial.)
So, Acked-by:
Roland Scheidegger <srol...@vmware.com>
>
> Marek
>
> On Mon, Jul 18, 2016 at 2:25 PM, Marek Olšák <mar...@g
Am 29.07.2016 um 00:35 schrieb Matt Turner:
> ---
> src/gallium/auxiliary/Makefile.sources | 1 +
> src/gallium/auxiliary/util/u_bitcast.h | 57
> ++
> 2 files changed, 58 insertions(+)
> create mode 100644 src/gallium/auxiliary/util/u_bitcast.h
>
> diff --git
1UfJxBl7yv9I2FeiaeAYgMTLKRBc_I=yFSuLWBBbLWZDrPaxDVOCjO0cX88sn16LdG-wdamPNQ=vIzvE6OQ57S1mnD6755iC9AOD57xXqJKUGtsk_DKBhs=
>
>
> Look especially at footnote 85 of 6.5.2.3.
>
> Regards,
> Gustaw Smolarczyk
>
> 2016-07-29 15:22 GMT+02:00 Roland Scheidegger <srol...@vmware.com>:
>
Am 29.07.2016 um 06:55 schrieb Connor Abbott:
> On Thu, Jul 28, 2016 at 7:58 PM, Roland Scheidegger <srol...@vmware.com>
> wrote:
>> Am 29.07.2016 um 00:35 schrieb Matt Turner:
>>> ---
>>> src/gallium/auxiliary/Makefile.sources | 1 +
>>>
This looks very bogus to me.
How is this supposed to work if you immediately close the library again?
The functions are still going to get used.
This just segfaults right, left and center.
If you want to fix this for real, some util_format_s3tc_exit() is
probably needed, albeit I'm not entirely
Reviewed-by: Roland Scheidegger <srol...@vmware.com>
Am 14.08.2016 um 19:09 schrieb Ilia Mirkin:
> Signed-off-by: Ilia Mirkin <imir...@alum.mit.edu>
> Cc: "11.2 12.0" <mesa-sta...@lists.freedesktop.org>
> ---
> src/gallium/auxiliary/Makefile.source
I can't say I'm a big fan of this.
From an "api cleanness" point of view, defining things in elements makes
more sense imho.
This is due to the GL api though which uses generic buffer api to turn
buffers into textures which can be sampled (somewhat different to what
d3d10 does).
Can't you just say
.h
> @@ -73,6 +73,9 @@ lp_get_default_memory_manager();
> extern void
> lp_free_memory_manager(LLVMMCJITMemoryManagerRef memorymgr);
>
> +extern void
> +lp_add_attr_dereferenceable(LLVMValueRef val, uint64_t bytes);
> +
> #ifdef __cplusplus
> }
> #endif
>
Re
Am 12.07.2016 um 13:40 schrieb Nicolai Hähnle:
> From: Nicolai Hähnle
>
> This fixes a regression introduced by commit d8d6091a8.
>
> Heap allocations may be only 8-byte aligned on 32-bit system, and so having
> members with 16-byte alignment (such as in the case where
;
> };
>
I'm wondering if that's really needed. I have a difficult time to
imagine setting blend color is performance critical. And driver internal
you can obviously still align pipe_blend_color structs yourself.
But OTOH, why not...
Acked-by: Roland Scheidegger <srol...@vmware
Am 05.07.2016 um 11:36 schrieb Marek Olšák:
> From: Marek Olšák
>
> RadeonSI stats: Mostly 0% difference, but Valley shows a small improvement:
>
> ApplicationFilesSGPRs VGPRs SpillSGPR SpillVGPR Code
> SizeLDSMax Waves Waits
>
Am 06.07.2016 um 23:51 schrieb Tim Rowley:
> ---
> .../drivers/swr/rasterizer/jitter/builder_misc.cpp | 38
> --
> .../jitter/scripts/gen_llvm_ir_macros.py | 5 ---
> 2 files changed, 28 insertions(+), 15 deletions(-)
>
> diff --git
Yes, these commands are modelled after d3d10 clears (the only d3d10
clear methods are clear_rt/clear_dsv), so gallium at some point only had
the "ordinary" clear (useful for d3d9 and gl, but doesn't cover
everything neither). It is unfortunate all apis handle clear differently.
Making them
Am 09.08.2016 um 00:59 schrieb Marek Olšák:
> From: Marek Olšák
>
> ---
> src/gallium/auxiliary/util/u_clear.h| 5 +++--
> src/gallium/auxiliary/vl/vl_bicubic_filter.c| 2 +-
> src/gallium/auxiliary/vl/vl_compositor.c| 2 +-
>
d_alloca_undef, but do not zero-initialize the variable.
Like lp_build_alloca
For the series:
Reviewed-by: Roland Scheidegger <srol...@vmware.com>
> + */
> +LLVMValueRef
> +lp_build_alloca_undef(struct gallivm_state *gallivm,
> + LLVMTypeRef type,
> +
what we were doing with the union was float <-> uint32_t bitcasts
> and the remaining 10% can be done with a sinmple left-shift by 23.
simple
For the series:
Reviewed-by: Roland Scheidegger <srol...@vmware.com>
>
> Signed-off-by: Jason Ekstrand <ja...@jlekstrand.net>
Am 01.07.2016 um 00:59 schrieb Matt Turner:
> According to the referenced bug report, gcc-4.5 and newer do not inline
> memcmp(). I see no difference in performance of ipers with llvmpipe on a
> Sandybridge (which does not have "Enhanced REP MOVSB/STOSB") by removing
> this flag.
>
> I attempted
& mask) != mask ||
> blit->filter != PIPE_TEX_FILTER_NEAREST ||
> blit->scissor_enable ||
> + blit->num_window_rectangles > 0 ||
> blit->alpha_blend) {
>return FALSE;
> }
>
Reviewed-by: Roland Scheidegger <srol...@vmware
ation and added a lengthy comment explaining the
> reason for the alignment.
>
> Reported-by: Tim Rowley <timothy.o.row...@intel.com>
> Tested-by: Tim Rowley <timothy.o.row...@intel.com>
> Signed-off-by: Chuck Atkins <chuck.atk...@kitware.com>
> Acked-by: Roland Sc
35,6 @@ util_can_blit_via_copy_region(const struct pipe_blit_info
> *blit)
>return FALSE;
> }
>
> - if (blit->alpha_blend)
> - return FALSE;
> -
> return TRUE;
> }
>
>
Now that you've simplified this, it looks to me like it should a
MAT_A8_UINT, SVGA3D_FORMAT_INVALID,
> SVGA3D_FORMAT_INVALID, 0 },
> { PIPE_FORMAT_I8_UINT, SVGA3D_FORMAT_INVALID,
> SVGA3D_FORMAT_INVALID, 0 },
>
I don't doubt that it didn't work, but I don't quite understand the
reasoning - R32G32B32_FLOAT seems to be s
Am 01.02.2017 um 05:32 schrieb Andrew A.:
> Are there any known issues with using usampler2D on a GL_R32UI texture
> in llvmpipe from within a VS/FS? I'm getting expected results on
> Nvidia drivers (proprietary), but not with mesa llvmpipe (always get
> zero).
Not that I know of. There should be
good idea - basically the idea is that reference counting is used
everywhere, so it should be as cheap as possible. (Note that this is
used in a lot of places with an explicit NULL pointer as the second
argument, and noone has ever claimed it caused actual issues.)
Roland
>
> Cheers,
> Nicola
I'm not quite sure there's really a bug here?
As far as I can tell, these functions are architected specifically to
look like that - they do not actually dereference potential null
pointers, but only take the address in the end. This change seems to add
some overhead.
Roland
Am 07.02.2017 um
__((no_sanitize("undefined"))) ?
>
> On Wed, Feb 8, 2017 at 3:48 PM, Roland Scheidegger <srol...@vmware.com
> <mailto:srol...@vmware.com>> wrote:
>
> Am 08.02.2017 um 10:21 schrieb Nicolai Hähnle:
> > On 07.02.2017 23:00, Brian Paul wrote:
>
1401 - 1500 of 2156 matches
Mail list logo