docs: Add sha256 sums for the 10.4.4 release

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
(cherry picked from commit 54da987baec25fbaf6975c93f197ddbfe65a303c)

Add release notes for the 10.4.4 release

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
(cherry picked from commit 62eb27ac8bbd979796f50e253f6c786d7d791242)

nir: Fix broken fsat recognizer.

We've probably never seen this ridiculous pattern in the wild, so it
didn't matter.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>

nir: Slightly simplify algebraic code generation by reusing a struct.

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>

tgsi/ureg: Add missing some missing opcodes opcode_tmp.h

I wanted all of these for NIR-to-TGSI.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>

tgsi/ureg: Move ureg_dst_register() to the header.

I wanted to use it for nir-to-tgsi.  The equivalent ureg_src_register() is
also located here.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>

gallium/u_tests: test a NULL buffer sampler view

Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>

gallium/u_tests: test a NULL constant buffer

This expects (0,0,0,0), though it can be changed to something else or allow
more than one set of values to be considered correct.

This is currently the radeonsi behavior.

Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>

gallium/u_tests: test a NULL texture sampler view

v2: allow one of the two values

gallium/u_tests: restructure the only test, refactor out reusable code

Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>

gallium: run gallium tests if GALLIUM_TESTS=1 is set

Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>

gallium/postprocessing: fix crash at context destruction

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

r600g/sb: fix a bug in constants folding optimisation pass

    ADD     R6.y.1,    R5.w.1, ~1|3f800000
    ADD     R6.y.2,    |R6.y.1|, -0.0001|b8d1b717

was wrongly being converted to

    ADD     R6.y.1,    R5.w.1, ~1|3f800000
    ADD     R6.y.2,    R5.w.1, -1.0001|bf800347

because abs() modifier was ignored.

Signed-off-by: Xavier Bouchoux <xavierb@gmail.com>
Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>

r600g: fix abs() support on ALU 3 source operands instructions

Since alu does not support abs() modifier on source operands, spill
and apply the modifiers to a temp register when needed.

Signed-off-by: Xavier Bouchoux <xavierb@gmail.com>
Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>

r300g: small code cleanup (v2)

v2: incorporated changes from Marek Olšák

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: David Heidelberg <david@ixit.cz>

glsl: GLSL ES identifiers cannot exceed 1024 characters

v2 (Ian Romanick)
- Move the check to the lexer before rallocing a copy of the large string.

Fixes the following 2 dEQP tests:
dEQP-GLES3.functional.shaders.keywords.invalid_identifiers.max_length_vertex
dEQP-GLES3.functional.shaders.keywords.invalid_identifiers.max_length_fragment

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

i965: Fix INTEL_DEBUG=shader_time for SIMD8 VS (and GS).

We were incorrectly attributing VS time to FS8 on Gen8+, which now use
fs_visitor for vertex shaders.

We don't hit this for geometry shaders yet, but we may as well add
support now - the fix is obvious, and we'll just forget later.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

i965/fs: Use inst->eot rather than opcodes in register allocation.

Previously, we special cased FB writes and URB writes in the register
allocation code.  What we really wanted was to handle any message with
EOT set.

This saves us from extending the list with new opcodes in the future.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>

i965/fs: Delete is_last_send(); just check inst->eot.

This helper function basically just checks inst->eot, but also asserts
that only opcodes we expect to terminate threads have EOT set.  As far
as I'm aware, we've never had such a bug.

Removing it means that we don't have to extend the list for new opcodes.
Cherryview and Skylake introduce an optimization where sampler messages
can have EOT set; scalar GS/HS/DS will likely introduce new opcodes as
well.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>

st/mesa: Don't use PIPE_USAGE_STREAM for GL_PIXEL_UNPACK_BUFFER_ARB

The latter currently implies CPU read access, so only PIPE_USAGE_STAGING
can be expected to be fast.

Mesa demos src/tests/streaming_rect on Kaveri (radeonsi):

Unpatched:  42 frames in  1.023 seconds = 41.056 FPS
Patched:   615 frames in  1.000 seconds = 615.000 FPS

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=88658
Cc: "10.3 10.4" <mesa-stable@lists.freedestkop.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

st/nine: Implement dummy vbo behaviour when vs is missing inputs

Use a dummy vertex buffer object when vs inputs have no corresponding
entries in the vertex declaration. This dummy buffer will give to the
shader float4(0,0,0,0).

This fixes several artifacts on some games.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
Signed-off-by: Tiziano Bacocco <tizbac2@gmail.com>

gallium/targets/d3dadapter9: Free card device

The drm fd wasn't released, causing a crash
for wine tests on nouveau, which seems to have
a bug when a lot of device descriptors are open.

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

gallium/targets/d3dadapter9: Release the pipe_screen at destruction.

We weren't releasing hal and ref, causing some issues (threads not released, etc)

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

gallium/targets/d3dadapter9: Fix device detection for render-nodes

When on a render node the unique ioctl doesn't work.

This patch drops the code to detect the device, which relied
on an ioctl, and replaces it by the mesa loader function.
The mesa loader function is more complete and won't fail for render-nodes.

Alternatively we could also have used the pipe cap to
determine the vendor and device id from the driver.

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Dummy sampler should have a=1

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Fix update_framebuffer binding cbufs the pixel shader wouldn't render to

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Clear: better behave if rt_mask is different to the one of the framebuffer bound

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Fix multisampling support detection

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Fix enabled lights in stateblocks

Reviewed-by: Axel Davy <axel.davy@ens.fr>
Signed-off-by: Tiziano Bacocco <tizbac2@gmail.com>

st/nine: Fix depth stencil formats bindings flags.

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Fix gpu memory leak in swapchain

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: SetResourceResize should track nr_samples too

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: D3DRS_FILLMODE set to 0 is D3DFILL_SOLID

Reviewed-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: David Heidelberg <david@ixit.cz>
Signed-off-by: Tiziano Bacocco <tizbac2@gmail.com>

st/nine: Setting D3DRS_ALPHAFUNC to 0 means D3DCMP_NEVER

Reviewed-by: Axel Davy <axel.davy@ens.fr>
Signed-off-by: Tiziano Bacocco <tizbac2@gmail.com>

st/nine: Implement fallback behaviour when rts and ds don't match

This seems to be the behaviour on Win. Previous behaviour led
to different issues depending on the driver.

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Fix present_buffers allocation

If has_present_buffers was false at first,
but after a device reset, it turns true (for
example if we begin to render to a multisampled
back buffer), there was a crash due to present_buffers
being uninitialised.
This patch fixes it.

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Check for aligned offset in each vertex element

Fixes wine test test_vertex_declaration_alignment()

Reviewed-by: Axel Davy <axel.davy@ens.fr>
Signed-off-by: Patrick Rudolph <siro@das-labor.org>

st/nine: Fix bufferoverflow in {Get|Set}PixelShaderConstantF

Previous code wasn't checking against the correct limit: 224
for sm3 hardware, but 256.

Fixes wine test test_pixel_shader_constant()

Reviewed-by: Axel Davy <axel.davy@ens.fr>
Signed-off-by: Patrick Rudolph <siro@das-labor.org>

st/nine: Set [out] argument to NULL for some functions

Wine tests, and probably some apps, check for errors by checking for NULL
instead of error codes.
Fixes wine test test_surface_blocks()

Reviewed-by: Axel davy <axel.davy@ens.fr>
Signed-off-by: Patrick Rudolph <siro@das-labor.org>

st/nine: Remove duplicated debug message

Likely a rebase error

Reviewed-by: Axel Davy <axel.davy@ens.fr>
Signed-off-by: Patrick Rudolph <siro@das-labor.org>

st/nine: Return E_FAIL for unused vertexdeclaration type

Add returncode E_FAIL.
Return E_FAIL for any vertexdeclaration element with type unused.

Reviewed-by: Axel Davy <axel.davy@ens.fr>
Signed-off-by: Patrick Rudolph <siro@das-labor.org>

st/nine: Missing sanity check for CALLOC return E_OUTOFMEMORY if allocation of usage_map fails

Reviewed-by: Axel Davy <axel.davy@ens.fr>
Signed-off-by: Patrick Rudolph <siro@das-labor.org>

st/nine: Implement ATOC hack

ATOC is an hack for Alpha to coverage
that is supported by NV and Intel.

You need to check the support for it
with CheckDeviceFormat.

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Implement AMD alpha to coverage

This D3D hack is supposed to be supported
by all AMD SM2+ cards. Apps use it without
checking if they are on AMD.

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Add D3DFMT_DF16 support

This depth buffer format, like D3DFMT_INTZ, can be used to read
the depth buffer values when bound to a shader.

Some apps may use this format to get better performance when
they don't need the precision of INTZ (24 bits for depth, 8 for
stencil, whereas DF16 is just 16 bits for depth)

We don't add support for DF24 yet, because it implies support
for FETCH4, which we don't support for now.

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Change the value of some advertised caps

These values are taken from wine.

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: NineDevice9_SetClipPlane: pPlane must be non-NULL

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Implement fallback for D3DFMT_D24S8, D3DFMT_D24X8 and D3DFMT_INTZ

Some drivers support PIPE_FORMAT_S8_UINT_Z24_UNORM,
some others PIPE_FORMAT_Z24_UNORM_S8_UINT, some both.

It doesn't matter which one we use, since the d3d formats
they map to aren't lockable (app can read it directly).

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Refactor format d3d9 to pipe conversion

Move the checks of whether the format is supported
into a common place.
The advantage is that allows to handle when a d3d9
format can be mapped to several formats, and that
cards don't support all of them.

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Refactor nine_d3d9_to_pipe_format_map

The order of the format is changed to have
an increasing ordering of the d3d9 format values.

Some missing formats are added and matched to PIPE_FORMAT_NONE

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Improve CheckDeviceFormat debug output

Because the debug output of this function was cut in two parts,
sometimes the second part wasn't print when we would return earlier,
whereas we would like to get it.

The reason of the separation was that it's only at the end of the function
we can print what we map to the d3d9 arguments, but we can always retrieve
that info by hand.

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Implement RESZ hack

This D3D hack allows to resolve a multisampled
depth buffer into a single sampled one.

Note that the implementation is slightly incorrect.
When querying the content of D3DRS_POINTSIZE,
it should return the resz code if it has been set.
This behaviour will be implemented when state changes
will be reworked. For now the current behaviour is ok,
since apps use the D3DCREATE_PUREDEVICE flag when creating
the device, which means they won't read states and in exchange
get better performance.

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: fix early basetexture destruction

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Do not leak private data in volume9.

This->data was allocated by nine, but not freed.

Reviewed-by: Axel Davy <axel.davy@ens.fr>
Signed-off-by: Patrick Rudolph <siro@das-labor.org>

st/nine: Check block alignment for compressed textures in NineSurface9_CopySurface

Reviewed-by: Axel Davy <axel.davy@ens.fr>
Signed-off-by: Patrick Rudolph <siro@das-labor.org>

st/nine: Commit sampler views again if srgb state changed.

This fixes a wine test and some minor visual issues on some games.

The patch is not optimal, there is probably a more efficient way to
fix this issue, but the code there already has some innefficiencies.
There is plans to rewrite that part of the code to make it more
efficient.

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Fix use of D3DSP_NOSWIZZLE

D3DSP_NOSWIZZLE already contains the shift.
Detected with Clang.

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Reviewed-by: David Heidelberg <david@ixit.cz>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Check for the correct number of constants.

This removes unneeded hack for Anno 1404.
This app is not checking the number of supporting
constants, and rely on the shader compilation to fail
if it puts too many constants.

This patch also checks for the correct number of constants for ps.

Note that we don't check the official limitations for old vs and ps
versions. The restrictions were fixed, unlike for the number of vertex
shader constants for later versions. Likely apps use the correct number,
and it's not a problem for us if it wants use more.

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Introduce failure handling for shader parsing.

Instead of crashing on buggy shaders, we should return an error.
This patch introduces this behaviour in the case of invalid constant
access

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Print warnings for r500 when shader is likely to go wrong

r500 hasn't enough float constants for vs to fill all needs.
Overlapping issues can happen with complex shaders.
The fix would be to recompile shaders to include the integer
and boolean constants, instead of reserving slots for them.

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Declare constants only up to the maximum needed.

Previously 276 constants were declared everytime.

This patch makes shaders declare constants up to the maximum
constant needed and moves the moment we print the TGSI
shader after the moment we declare the constants.

This is needed for r500, since when indirect addressing is used,
it cannot reduce the amount of constants needed, and that it is
restricted to 256 constant slots.

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Refactor how user constbufs sizes are calculated

Count explicitly the slots for float, int and bool constants,
and deduce the constbuf size in nine_shader.

Reviewed-by: Tiziano Bacocco <tizbac2@gmail.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

st/nine: Explicit nine requirements

This patch raises nine requirements and disables nine for old
hw that don't match them.

Currently for these cards only games that don't have tight requirements
would work well with nine. However nine is missing several checks
regarding these limitations.
To make code and future patches less heavy, dropping support for these old
card seems a good solution.

That makes r500 the only dx9 generation cards supported by nine. It seems the one
with the less limitations for nine. Still not everything is ok, and we'll have
for example to implement shader recompilation for these cards to include
integer and boolean constants in the shader.
Eventually when this is done, we can reintroduce support for older cards.

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

gallium: Add MULTISAMPLE_Z_RESOLVE cap

Resolving a multisampled depth texture into
a single sampled texture is supported on >= SM4.1
hw. It is possible some previous hw support it.

The ability was tested on radeonsi and nvc0.
Apparently is is also supported for radeon >= r700.

This patch adds the MULTISAMPLE_Z_RESOLVE cap and
add it to the drivers. It is advertised for drivers
for which it is sure the ability is supported.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Axel Davy <axel.davy@ens.fr>

GL: Update glext.h to Revision 29735 (20150202).

Khronos modified glext.h to get rid of GL_TEXTURE_BINDING, a special enum
added for ARB_direct_state_access.  This enum was ruled unimplementable.

Reviewed-by: Brian Paul <brianp@vmware.com>
Tested-by: Laura Ekstrand <laura@jlekstrand.net>

llvmpipe: Trivially advertise PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT.

Nothing special needs to be done.

Even though llvmpipe copies constant (ie uniform) buffers internally, the
application is supposed to flush and sync, so all should work.

All bufferstorage piglit tests pass.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>

i965: Remove now unnecessary Gen8 CMP destination type override.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Set CMP's destination type to src0's type.

Allows CMP instructions with float sources to be compacted and coissued.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965/fs: Implement the WaCMPInstFlagDepClearedEarly work-around.

Prevents piglit regressions from the next patch.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

gallium/util: Don't implement u_bit_scan64 on MSVC.

As ffsll doesn't exist in MSVC yet, and u_bit_scan64 is only used by
radeonsi which is never built with MSVC.

This is just a stop-gap fix to unbreak MSVC build until we refactor these
mathematical portability wrappers into src/util.

Trivial.

gallium/util: Define ffsll on MinGW.

Trivial.

(Fixing MSVC will be far less so, as _BitScanForward64 is only supported on x64.)

radeonsi: implement polygon stippling

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: add polygon stipple texture slot

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: deduce rasterizer primitive type at the beginning of draw_vbo

I will need this for polygon stippling.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: allow 64 descriptors per array

We need a slot for the stipple texture and the pixel shader already uses
32 textures (16 API slots + 16 FMASK slots).

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: add support for sampler views where resource = NULL

The hardware obeys swizzles even if the resource is NULL.
This will be used by set_polygon_stipple.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: add support for NULL texture sampler views that return (0,0,0,1)

This used to hang.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: fix a crash when binding a NULL sampler view list

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: move the buffer descriptor to the end of the image descriptor

This will allow supporting NULL textures.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: don't use tgsi_parse_context to get processor type

Also remove unused "tokens".

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: fix instanced arrays with non-zero start instance

Fixes piglit ARB_base_instance/arb_base_instance-drawarrays.

Cc: 10.3 10.4 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

r600g,radeonsi: don't append to streamout buffers that haven't been used yet

The FILLED_SIZE counter is uninitialized at the beginning, so we can't use it.
Instead, use offset = 0, which is what we always do when not appending.

This unexpectedly fixes spec/ARB_texture_multisample/sample-position/*.
Yes, the test does use transform feedback.

Cc: 10.3 10.4 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

gallium: set PIPE_MAX_SAMPLERS to 18

For drivers that use higher slots not to crash in tgsi_shader_info.

Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

gallium/u_pstipple: add ability to specify a fixed texture unit

E.g. r600g can use slot 17, which is outside of the API range.

Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

gallium/util: add u_bit_scan64

Same as u_bit_scan, but for uint64_t.

Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

tgsi: add tgsi_get_processor_type helper from radeon

Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

i965/fs: Fix saturate on MAD and LRP with the NIR backend.

Fixes misrendering in "Witcher 2" with INTEL_USE_NIR=1, and probably
many other programs.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

mesa: Fix _mesa_format_convert fallback path when src is not an array format

When a rebase swizzle is provided and we call _mesa_swizzle_and_convert
after unpacking the source format we were always passing normalized=false.
We should pass true or false depending on the formats involved in the
conversion for the byte and float paths (the integer path cannot ever be
normalized).

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Tested-by: Mark Janes <mark.a.janes@intel.com>

st/osmesa: Fix osbuffer->textures indexing

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=88930
Cc: 10.4 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>

i965/nir: use redundant phi optimization

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Tested-by: Jason Ekstrand <jason.ekstrand@intel.com>
Signed-off-by: Connor Abbott <cwabbott0@gmail.com>

nir: add an optimization to remove useless phi nodes

This removes phi nodes whose sources all point to the same thing.

Shader-db results:

total NIR instructions in shared programs: 2045293 -> 2041209 (-0.20%)
NIR instructions in affected programs:     126564 -> 122480 (-3.23%)
helped:                                615
HURT:                                  0

total FS instructions in shared programs: 4321840 -> 4320392 (-0.03%)
FS instructions in affected programs:     24622 -> 23174 (-5.88%)
helped:                                138
HURT:                                  0

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Tested-by: Jason Ekstrand <jason.ekstrand@intel.com>
Signed-off-by: Connor Abbott <cwabbott0@gmail.com>

nir/validate: Ensure that phi sources are SSA-only

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir/validate: Validate that only float ALU outputs are saturated

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir/lower_source_mods: Don't lower saturate for non-float outputs

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

i965/fs_nir: Get rid of get_alu_src

Originally, get_alu_src was supposed to handle resolving swizzles and
things like that.  However, now that basically every instruction we have
only takes scalar sources, we don't really need it anymore.  The only case
where it's still marginally useful is for the mov and vecN operations that
are left over from SSA form.  We can handle those cases as a special case
easily enough.  As a side-effect, we don't need the vec_to_movs pass
anymore.

v2 Jason Ekstrand <jason.ekstrand@intel.com>:
 - Rework the way we detect if we need an extra copy for swizzling.  The
   old code involved a pile of confusing switch fall-throughs; we now use a
   loop.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965/fs: Use NIR's scalarizing abilities and stop handling vectors

Now that we can scalarize with NIR, there's no need for all this code
anymore.  Let's get rid of it and just do scalar operations.

v2: run copy prop before lowering phi nodes

v3: Get rid of the "emit(...)->saturate = foo" pattern

v4: Run alu_to_scalar as an optimization pass

total instructions in shared programs: 5998321 -> 5974070 (-0.40%)
instructions in affected programs:     732075 -> 707824 (-3.31%)
helped:                                3137
HURT:                                  191
GAINED:                                18
LOST:                                  0

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

nir: Add a pass to lower vector phi nodes to scalar phi nodes

v2 Jason Ekstrand <jason.ekstrand@intel.com>:
 - Add better comments
 - Use nir_ssa_dest_init and nir_src_for_ssa more places
 - Fix some void * casts

v3 Jason Ekstrand <jason.ekstrand@intel.com>:
 - Rework the way we determine whether or not to sccalarize a phi node to
   make the recursion non-bogus
 - Treat load_const instructions as scalarizable

v4 Jason Ekstrand <jason.ekstrand@intel.com>:
 - Allow uniform and input loads to be scalarizable

v5 Jason Ekstrand <jason.ekstrand@intel.com>:
 - Also consider loads of inputs (varying, uniform, or ubo) to be
   scalarizable.  We were already doing this for load_var on uniforms and
   inputs.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965/fs: Add support for constant propagating into sources with modifiers.

All but 16 of the programs helped were ARB fp programs.

total instructions in shared programs: 5949286 -> 5945470 (-0.06%)
instructions in affected programs:     275162 -> 271346 (-1.39%)
helped:                                1197
GAINED:                                1

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

i965/vec4: Use abs/negate functions in const propagation.

No changes in shader-db.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

i965: Add function to take the abs of immediates.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>