i965/wm: emit null buffer surfaces when null buffers are attached

Otherwise we can expect odd things to happen if, for example, we ask
for the size of the attached buffer from shader code, since that
might query this value from the surface we uploaded and get random
results.

Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

i965/fs/nir: implement nir_intrinsic_get_buffer_size

v2:
- Remove inst->regs_written assignment as the instruction only
  writes to one register.

Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

i965/fs: Implement FS_OPCODE_GET_BUFFER_SIZE

Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

i965/vec4/nir: implement nir_intrinsic_get_buffer_size

Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

i965/vec4: Implement VS_OPCODE_GET_BUFFER_SIZE

Notice that Skylake needs to include a header in the sampler message
so it will need some tweaks to work there.

Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

nir: Implement ir_unop_get_buffer_size

This is how backends provide the buffer size required to compute
the size of unsized arrays in the previous patch

Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

glsl: implement unsized array length

v2:
- Reduce the number of lines over 80 character line width
  limit. (Thomas Hellan)

v3:
- Inject the formula to compute the array length in the IR, backends
  only need to provide the buffer size (Curro)
- Create an auxiliary function to simplify code (Jordan Justen)
- Rename variables (Jordan Justen)

Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

glsl: Add parser/compiler support for unsized array's length()

The unsized array length is computed with the following formula:

array.length() =
   max((buffer_object_size - offset_of_array) / stride_of_array, 0)

Of these, only the buffer size needs to be provided by the backends, the
frontend already knows the values of the two other variables.

This patch identifies the cases where we need to get the length of an
unsized array, injecting ir_unop_ssbo_unsized_array_length expressions
that will be lowered (in a later patch) to inject the formula mentioned
above.

It also adds the ir_unop_get_buffer_size expression that drivers will
implement to provide the buffer length.

v2:
- Do not define a triop that will force backends to implement the
  entire formula, they should only need to provide the buffer size
  since the other values are known by the frontend (Curro).

v3:
- Call state->has_shader_storage_buffer_objects() in ast_function.cpp instead
  of using state->ARB_shader_storage_buffer_object_enable (Tapani).

Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

glsl: Add unsized array support to glsl_type::std140_size()

Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

glsl: fix indention in glsl_types.cpp

No functional changes.

Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

glsl: add support for unsized arrays in shader storage blocks

They only can be defined in the last position of the shader
storage blocks.

When an unsized array is used in different shaders, it might be
converted in different sized arrays, avoid get a linker error
in that case.

v2:
- Rework error condition and error messages (Timothy Arceri)

v3:
- Move OpenGL ES check to its own patch.

Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

glsl: return error if unsized arrays are found in OpenGL ES

Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

i965/fs: Do not split buffer variables

Buffer variables are the same as uniforms, only that read/write, so we want
the same treatment.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

i965: handle visiting of ir_var_shader_storage variables

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

i965: Upload Shader Storage Buffer Object surfaces

Since these are a special kind of UBOs we emit them together reusing the
same infrastructure, however, we use a RAW surface so we can reuse
existing untyped read/write/atomic messages which include a pixel mask
header that we need to set to obtain correct behavior with helper
invocations of the fragment shader.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

i965: Set MaxShaderStorageBuffers for compute shaders

v2:
- Set it after the driver's MaxShaderStorageBuffers value assignment.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

i965: set ARB_shader_storage_buffer_object related constant values

v2:
- Add tessellation shader constants assignment

v3:
- Set MaxShaderStorageBufferBindings to 36.

Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

i965: Implement DriverFlags.NewShaderStorageBuffer

We use the same dirty state for SSBOs and UBOs because they share the
same infrastructure.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

i965: Use 64-byte offset alignment for shader storage buffers

This should be a cacheline (64 bytes) so that we can safely have the
CPU and GPU writing the same SSBO on non-cachecoherent systems (our
Atom CPUs). With UBOs, the GPU never writes, so there's no
problem. For an SSBO, the GPU and the CPU can be updating disjoint
regions of the buffer simultaneously and that will break if the
regions overlap the same cacheline.

v2:
- Use cacheline size (64 bytes) instead of 16 bytes (Kristian).
- Update commit log and add a comment in the code explaining
  why we use cacheline size (Ben).

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

mesa: set MAX_SHADER_STORAGE_BUFFERS to 16.

v2:
- Set the value to 16 and drop the comment. (Kristian)

Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

glsl: add packed varyings to program resource list

This makes sure that user is still able to query properties about
variables that have gotten packed by lower_packed_varyings pass.

Fixes following OpenGL ES 3.1 test:
   ES31-CTS.program_interface_query.separate-programs-vertex

v2: fix 'name included in packed list' check (Ilia Mirkin)
v3: iterate over instances of name using strtok_r (Ilia Mirkin)

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>

mesa: add packed_varyings list to gl_shader

This is required to store information about packed varyings, currently
these variables get lost and cannot be retrieved later in sensible way
for program interface queries. List will be utilized by next patch.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>

i965/cs: Implement DispatchComputeIndirect support

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

mesa/cs: Implement glDispatchComputeIndirect

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

mesa/cs: Support GL_DISPATCH_INDIRECT_BUFFER

v2:
 * Use _mesa_has_compute_shaders (Ilia)

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

mesa/cs: Add _mesa_validate_DispatchCompute

Move API validation to _mesa_validate_DispatchCompute in
api_validate.c.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

mesa: fix mipmap generation for immutable, compressed textures

If the immutable compressed texture didn't have the full mip pyramid,
this didn't work, because it tried to generate mip levels for non-existing
levels. _mesa_prepare_mipmap_level() would correctly handle this by returning
FALSE if the mip level didn't exist, however we actually created the
non-existing mip level right before that because we used _mesa_get_tex_image()
before calling _mesa_prepare_mipmap_level(). It would then proceed to crash
(we allocated the mip level, which is a bad idea on an immutable texture,
but didn't initialize the values, leading to assertion failures or segfaults).
Fix this by using _mesa_select_tex_image() instead and call it after
_mesa_prepare_mipmap_level(), as that function will allocate missing mip levels
for non-immutable textures already.
This fixes a (2 year old) crash with astromenace which was hack-fixed in ubuntu
packages instead: http://bugs.debian.org/718680 (I guess most apps do full mip
chains - I believe this app not doing it is actually unintentional, always one
level less than full mip chain...).

Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>

glsl: Expose gl_MaxTess{Control,Evaluation}AtomicCounters.

... with only ARB_shader_atomic_counters.

I expected to see interactions with ARB_tessellation_shader in the
ARB_shader_atomic_counters spec, but they do not exist. It seems that we
should unconditionally expose these variables in the presence of
ARB_shader_atomic_counters:

   gl_MaxTessControlAtomicCounters
   gl_MaxTessEvaluationAtomicCounters

This partially reverts commit da7adb99e8. The commit also affected
gl_MaxTessControlImageUniforms and gl_MaxTessEvaluationImageUniforms
similarly but the ARB_shader_image_load_store spec does list an
interaction with ARB_tessellation_shader.

Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92095
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965/vec4: check swizzle before discarding a uniform on a 3src operand

Without this commit, copy propagation is discarded if it involves
a uniform with an instruction that has 3 sources. But 3 sourced
instructions can access scalar values.

For example, this is what vec4_visitor::fix_3src_operand() is already
doing:

   if (src.file == UNIFORM && brw_is_single_value_swizzle(src.swizzle))
      return src;

Shader-db results (unfiltered) on NIR:
total instructions in shared programs: 6259650 -> 6241985 (-0.28%)
instructions in affected programs:     812755 -> 795090 (-2.17%)
helped:                                7930
HURT:                                  0

Shader-db results (unfiltered) on IR:
total instructions in shared programs: 6445822 -> 6441788 (-0.06%)
instructions in affected programs:     296630 -> 292596 (-1.36%)
helped:                                2533
HURT:                                  0

v2:
- Updated commit message, using Matt Turner suggestions
- Move the check after we've created the final value, as Jason
  Ekstrand suggested
- Clean up the condition

v3:
- Move the check back to the original place, to keep things
  tidy, as suggested by Jason Ekstrand

v4:
- Fixed missing is_single_value_swizzle() as pointed by Jason Ekstrand

Reviewed-by: Matt Turner <mattst88@gmail.com>

android: radeonsi: fix sid_tables.h missing LOCAL_MODULE_CLASS

Signed-off-by: Marek Olšák <marek.olsak@amd.com>

gallium/radeon: remove the percentage symbol from HUD temperature

The HUD adds '%' if max == 100.

Signed-off-by: Benjamin Bellec <b.bellec@gmail.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>

gallium/u_blitter: handle allocation failures

Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: handle dummy constant buffer allocation failure

Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: don't forget to update scratch relocations for LS, HS, ES shaders

Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: skip drawing if updating the scratch buffer fails

Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: skip drawing if PS fails to compile or upload

Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: skip drawing if VS, TCS, TES, GS fail to compile or upload

Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: handle fixed-func TCS shader create failure

Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: handle shader precompile failures

Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: skip drawing if GS ring allocations fail

Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: skip drawing if the tess factor ring allocation fails

Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: add malloc fail paths to si_create_shader_state

Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: report alloc failure from si_shader_binary_read

Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

gallium/radeon: add a fail path for depth MSAA texture readback

Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

gallium/radeon: handle buffer alloc failures in r600_draw_rectangle

Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

gallium/radeon: handle buffer_map staging buffer failures better

Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: handle constant buffer alloc failures

Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: handle index buffer alloc failures

Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

st/mesa: fix front buffer regression after dropping st_validate_state in Blit

Broken by: d082c5324914212f76e45be497229c7a0681f706
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92072

Cc: 10.6 11.0 <mesa-stable@lists.freedesktop.org>
Tested-by: Ilia Mirkin <imirkin@alum.mit.edu>

wayland: Add copyright notice for wayland-egl.c

Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>

i965: Respect stride and subreg_offset for ATTR registers

When we assign hw regs to attributes, we don't incorporate the stride
and subreg_offset from the fs_reg. It's rarely used, but the integer
multiplication lowering uses unusual stride and subreg_offset
combination breaks when one source is an attribute.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91970
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
Reviewed-by: Matt Turner <mattst88@gmail.com>

mesa: rework Driver.CopyImageSubData() and related code

Previously, core Mesa's _mesa_CopyImageSubData() created temporary textures
to wrap renderbuffer sources/destinations.  This caused a bit of a mess in
the Mesa/gallium state tracker because we had to basically undo that
wrapping.

Instead, change ctx->Driver.CopyImageSubData() to take both gl_renderbuffer
and gl_texture_image src/dst pointers (one being null, the other non-null)
so the driver can handle renderbuffer vs. texture as needed.

For the i965 driver, we basically moved the code that wrapped textures
around renderbuffers from copyimage.c down into the met and driver code.

The old code in copyimage.c also made some questionable calls to
_mesa_BindTexture(), etc. which weren't undone at the end.

v2 (Jason Ekstrand): Rework the intel bits
v3 (Brian Paul): Update the temporary st_CopyImageSubData() function.

Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Tested-by: Kai Wasserbäch <kai@dev.carbon-project.org>
Tested-by: Nick Sarnie <commendsarnex@gmail.com>

st/xa: Fixups for PIPE_FORMAT_R8_UNORM A8 usage v2.

Check for PIPE_FORMAT_R8_UNORM when setting up the copy shader.
Also re-enable the dest alpha blending with A8 destination that
actually turned out to be correct.

Verified using rendercheck that the composite operators
overreverse, in, out, atop, atopreverse and xor seem to work fine
with a8 destiation.

v2: Fix a copy-paste error.

Reported-by: Jose Fonseca <jfonseca@vmware.com>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>

st/mesa: keep track of saturated writes when eliminating dead code

It doesn't matter whether a write is saturated or not, in another
implementation it might even have been a separate opcode. This code was
most likely copied from the copy-propagation pass (where one does have
to distinguish saturation).

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

glsl: correctly detect inactive UBO arrays

Previously the code was trying to get the packing type from the array not the
interface.

Cc: Ian Romanick <ian.d.romanick@intel.com>
Cc: Antia Puentes <apuentes@igalia.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>

i965: add ARB_texture_barrier support

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>

i965/gs: Fix extra level of indentation left by the previous commit.

I left a bunch of code indented a level in the previous patch to make
the diff easier to read.  But now we should fix that.

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

i965/gs: Use new NIR intrinsics.

By performing the vertex counting in NIR, we're able to elide a ton of
useless safety checks around every EmitVertex() call:

total instructions in shared programs: 3952 -> 3720 (-5.87%)
instructions in affected programs:     3491 -> 3259 (-6.65%)
helped:                                11
HURT:                                  0

Improves performance in Gl32GSCloth by 0.671742% +/- 0.142202% (n=621)
on Haswell GT3e at 1024x768.

This should also make it easier to implement Broadwell's "Static Vertex
Count" feature someday.

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

nir: Add new GS intrinsics that maintain a count of emitted vertices.

This patch also introduces a lowering pass to convert the simple GS
intrinsics to the new ones.  See the comments above that for the
rationale behind the new intrinsics.

This should be useful for i965; it's a generic enough mechanism that I
could see other drivers potentially using it as well, so I don't feel
too bad about putting it in the generic code.

v2:
- Use nir_after_block_before_jump for the cursor (caught by Jason
  Ekstrand - I'd mistakenly used nir_after_block when rebasing this
  code onto the new NIR control flow API).
- Remove the old emit_vertex intrinsic at the end, rather than in
  the middle (requested by Jason).
- Use state->... directly rather than locals (requested by Jason).
- Report progress from nir_lower_gs_intrinsics() (requested by me).
- Remove "Authors:" section from file comment (requested by
  Michael Schellenberger Costa).

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

nir: Add unit tests for control flow graphs.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Acked-by: Jason Ekstrand <jason.ekstrand@intel.com>
Acked-by: Connor Abbott <cwabbott0@gmail.com>

nir/cf: Fix dominance metadata in the dead control flow pass.

The NIR control flow modification API churns the block structure,
splitting blocks, stitching them back together, and so on.  Preserving
information about block dominance is hard (and probably not worthwhile).

This patch makes nir_cf_extract() throw away all metadata, like we do
when adding/removing jumps.

We then make the dead control flow pass compute dominance information
right before it uses it.  This is necessary because earlier work by the
pass may have invalidated it.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>

nir/cf: Fix unlink_block_successors to actually unlink the second one.

Calling unlink_blocks(block, block->successors[0]) will successfully
unlink the first successor, but then will shift block->successors[1]
down to block->successor[0].  So the successors[1] != NULL check will
always fail.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>

nir/cf: Alter block successors before adding a fake link.

Consider the case of "while (...) { break }".  Or in NIR:

        block block_0 (0x7ab640):
        ...
        /* succs: block_1 */
        loop {
                block block_1:
                /* preds: block_0 */
                break
                /* succs: block_2 */
        }
        block block_2:

Calling nir_handle_remove_jump(block_1, nir_jump_break) will remove the break.
Unfortunately, it would mangle the predecessors and successors.

Here, block_2->predecessors->entries == 1, so we would create a fake
link, setting block_1->successors[1] = block_2, and adding block_1 to
block_2's predecessor set.  This is illegal: a block cannot specify the
same successor twice.  In particular, adding the predecessor would have
no effect, as it was already present in the set.

We'd then call unlink_block_successors(), which would delete the fake
link and remove block_1 from block_2's predecessor set.  It would then
delete successors[0], and attempt to remove block_1 from block_2's
predecessor set a second time...except that it wouldn't be present,
triggering an assertion failure.

The fix appears to be simple: simply unlink the block's successors and
recreate them to point at the correct blocks first.  Then, add the fake
link.  In the above example, removing the break would cause block_1 to
have itself as a successor (as it becomes an infinite loop), so adding
the fake link won't cause a duplicate successor.

v2: Add comments (requested by Connor Abbott) and fix commit message.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>

nir/cf: Conditionally do block_add_normal_succs() in unlink_jump();

There is a bug where we mess up predecessors/successors due to the
ordering of unlinking/recreating edges/adding fake edges.  In order to
fix that, I need everything in one routine.

However, calling block_add_normal_succs() isn't safe from
cleanup_cf_node() - it would crash trying to insert phi undefs.
So unfortunately I need to add a parameter.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>

nir/cf: Don't break outer-block successors in split_block_beginning().

Consider the following NIR:

   block block_0;
   /* succs: block_1 block_2 */
   if (...) {
      block block_1;
      ...
   } else {
      block block_2;
   }

Calling split_block_beginning() on block_1 would break block_0's
successors:  link_block() sets both successors of a block, so calling
link_block(block_0, new_block, NULL) would throw away the second
successor, leaving only /* succ: new_block */.  This is invalid: the
block before an if statement must have two successors.

Changing the call to link_block(pred, new_block, pred->successors[0])
would correctly leave both successors in place, but because unlink_block
may shift successor[1] to successor[0], it may not preserve the original
order.  NIR maintains a convention that successor[0] must point to the
"then" block, while successor[1] points to the "else" block, so we need
to take care to preserve this ordering.

This patch creates a new function that swaps out one successor for
another, preserving the ordering.  It then uses this to fix the issue.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>

nir/cf: Make a helper function for removing a predecessor.

I need to do this in a second place, and I'd rather make a helper
function than cut and paste the code.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>

nir: Validate that a block doesn't have two identical successors.

This is invalid, and causes disasters if we try to unlink successors:
removing the first will work, but removing the second copy will fail
because the block isn't in the successor's predecessor set any longer.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>

nir/lower_vec_to_movs: Don't emit unneeded movs

It's possible that, if a vecN operation is involved in a phi node, that we
could end up moving from a register to itself.  If swizzling is involved,
we need to emit the move but.  However, if there is no swizzling, then the
mov is a no-op and we might as well not bother emitting it.

Shader-db results on Haswell:

   total instructions in shared programs: 6262536 -> 6259558 (-0.05%)
   instructions in affected programs:     184780 -> 181802 (-1.61%)
   helped:                                838
   HURT:                                  0

Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

nir/lower_vec_to_movs: Properly handle source modifiers on vecN ops

I don't know of any piglit tests that are currently broken.  However, there
is nothing stopping a vecN instruction from getting source modifiers and
lower_vec_to_movs is run after we lower to source modifiers.

Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i915: Make hw_prim[] const

The table used to map the GL primitive to the hw primitive never
changes so make it const.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

t_dd_dmatmp: Make the render_tab[]s const

These tables hold function pointers and they never change so
make them const.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: Remove unused HAVE_TRI_STRIP_1 defines

Defined to 0 in a few places, but it's not used anywhere.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

t_dd_dmatmp: Constify dmasz

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

t_dd_dmatmp: Silence comparison between signed and unsigned integer expression warnings

../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:83:28: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
          nr = MIN2(currentsz, count - j);
                            ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:83:55: warning: signed and unsigned type in conditional expression [-Wsign-compare]
          nr = MIN2(currentsz, count - j);
                                                       ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:116:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
       nr = MIN2(currentsz, count - j);
                         ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:116:52: warning: signed and unsigned type in conditional expression [-Wsign-compare]
       nr = MIN2(currentsz, count - j);
                                                    ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:140:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
       nr = MIN2(currentsz, count - j);
                         ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:140:52: warning: signed and unsigned type in conditional expression [-Wsign-compare]
       nr = MIN2(currentsz, count - j);
                                                    ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h: In function 'intel_render_line_loop_verts':
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:174:28: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
          nr = MIN2(currentsz, count - j);
                            ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:174:55: warning: signed and unsigned type in conditional expression [-Wsign-compare]
          nr = MIN2(currentsz, count - j);
                                                       ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:224:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
       nr = MIN2(currentsz, count - j);
                         ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:224:52: warning: signed and unsigned type in conditional expression [-Wsign-compare]
       nr = MIN2(currentsz, count - j);
                                                    ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:255:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
       nr = MIN2(currentsz, count - j);
                         ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:255:52: warning: signed and unsigned type in conditional expression [-Wsign-compare]
       nr = MIN2(currentsz, count - j);
                                                    ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:281:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
       nr = MIN2(currentsz, count - j + 1);
                         ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:281:56: warning: signed and unsigned type in conditional expression [-Wsign-compare]
       nr = MIN2(currentsz, count - j + 1);
                                                        ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h: In function 'intel_render_poly_verts':
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:313:28: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
          nr = MIN2(currentsz, count - j + 1);
                            ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:313:59: warning: signed and unsigned type in conditional expression [-Wsign-compare]
          nr = MIN2(currentsz, count - j + 1);
                                                           ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:365:28: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
          nr = MIN2(currentsz, count - nr);
                            ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:365:56: warning: signed and unsigned type in conditional expression [-Wsign-compare]
          nr = MIN2(currentsz, count - nr);
                                                        ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:83:28: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
          nr = MIN2(currentsz, count - j);
                            ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:83:55: warning: signed and unsigned type in conditional expression [-Wsign-compare]
          nr = MIN2(currentsz, count - j);
                                                       ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:116:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
       nr = MIN2(currentsz, count - j);
                         ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:116:52: warning: signed and unsigned type in conditional expression [-Wsign-compare]
       nr = MIN2(currentsz, count - j);
                                                    ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:140:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
       nr = MIN2(currentsz, count - j);
                         ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:140:52: warning: signed and unsigned type in conditional expression [-Wsign-compare]
       nr = MIN2(currentsz, count - j);
                                                    ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h: In function 'radeon_dma_render_line_loop_verts':
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:174:28: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
          nr = MIN2(currentsz, count - j);
                            ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:174:55: warning: signed and unsigned type in conditional expression [-Wsign-compare]
          nr = MIN2(currentsz, count - j);
                                                       ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:224:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
       nr = MIN2(currentsz, count - j);
                         ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:224:52: warning: signed and unsigned type in conditional expression [-Wsign-compare]
       nr = MIN2(currentsz, count - j);
                                                    ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:255:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
       nr = MIN2(currentsz, count - j);
                         ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:255:52: warning: signed and unsigned type in conditional expression [-Wsign-compare]
       nr = MIN2(currentsz, count - j);
                                                    ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:281:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
       nr = MIN2(currentsz, count - j + 1);
                         ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:281:56: warning: signed and unsigned type in conditional expression [-Wsign-compare]
       nr = MIN2(currentsz, count - j + 1);
                                                        ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h: In function 'radeon_dma_render_poly_verts':
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:313:28: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
          nr = MIN2(currentsz, count - j + 1);
                            ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:313:59: warning: signed and unsigned type in conditional expression [-Wsign-compare]
          nr = MIN2(currentsz, count - j + 1);
                                                           ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:365:28: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
          nr = MIN2(currentsz, count - nr);
                            ^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:365:56: warning: signed and unsigned type in conditional expression [-Wsign-compare]
          nr = MIN2(currentsz, count - nr);
                                                        ^

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

t_dd_dmatmp: Use stdbool.h

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

t_dd_dmatmp: General indentation and formatting fixes

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

t_dd_dmatmp: Indentation and formatting fixes after HAVE_ELTS change

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

t_dd_dmatmp: Remove HAVE_ELTS support

Two drivers use this file, and neither supports ELTs.

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

t_dd_dmatmp: Indentation and formatting fixes after HAVE_TRI_FANS change

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

t_dd_dmatmp: Require HAVE_TRI_FANS

Two drivers use this file, and both support triangle fans.

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

t_dd_dmatmp: Indentation and formatting fixes after HAVE_TRI_STRIPS change

v2: Fix '- nr' typo noticed by Marius.

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com> [v1]

t_dd_dmatmp: Require HAVE_TRI_STRIPS

Two drivers use this file, and both support triangle strips.

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

t_dd_dmatmp: Require HAVE_TRIANGLES

Two drivers use this file, and both support triangles.

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

t_dd_dmatmp: Indentation and formatting fixes after HAVE_LINE_STRIPS change

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

t_dd_dmatmp: Require HAVE_LINE_STRIPS

Two drivers use this file, and both support line strips.

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

t_dd_dmatmp: Indentation and formatting fixes after HAVE_LINES change

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

t_dd_dmatmp: Require HAVE_LINES

Two drivers use this file, and both support lines.

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

t_dd_dmatmp: Indentation and formatting fixes after HAVE_QUADS change

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

t_dd_dmatmp: Remove HAVE_QUADS support

Two drivers use this file, and neither supports quads.

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

t_dd_dmatmp: Remove HAVE_QUAD_STRIPS support

Two drivers use this file, and neither supports quad strips.

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

t_dd_dmatmp: Use addition instead of subtraction in loop bounds

This is used everywhere else in this file because it avoids problems
when count is zero (due to trimming).

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=38109
Reviewed-by: Brian Paul <brianp@vmware.com>
Cc: Marius Predut <marius.predut@intel.com>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>

t_dd_dmatmp: Pull out common 'count -= count & 3' code

This was missing in the HAVE_TRIANGLES path, and that could cause
incorrect rendering.

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=38109
Reviewed-by: Brian Paul <brianp@vmware.com>
Cc: Marius Predut <marius.predut@intel.com>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>

t_dd_dmatmp: Use '& 3' instead of '% 4' everywhere

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>

t_dd_dmatmp: Clean up improper code formatting from previous patch

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>

t_dd_dmatmp: Make "count" actually be the count

The value passed in count previously was "vertex after the last vertex
to be processed."  Calling that "count" was misleading and kind of mean.
Looking at the code, many functions immediately do "count-start" to get
back the true count.  That's just silly.

If it is better for the loops to be 'for (j = start; j < (start +
count); j++)', GCC will do that transformation.

NOTE: There is some strange formatting left by this patch.  That was
done to make it more obvious that the before and after code is
equivalent.  These will be fixed in the next patch.

No piglit regressions on i915 (G33) or radeon (Radeon 7500).

v2: Fix a remaining (count-start) in render_quad_strip_verts.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com> [v1]
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>

i965/vec4: Don't coalesce regs in Gen6 MATH ops if reswizzle/writemask needed

Gen6 MATH instructions can not execute in align16 mode, so swizzles or
writemasking are not allowed.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92033
Reviewed-by: Matt Turner <mattst88@gmail.com>

mesa: Fix GL_FRAMEBUFFER_ATTACHMENT_OBJECT_TYPE for default framebuffer.

From section 9.2. Binding and Managing Framebuffer Objects:

"Upon successful return from Get*FramebufferAttachmentParameteriv, if
pname is FRAMEBUFFER_ATTACHMENT_OBJECT_TYPE, then params will contain
one of NONE, FRAMEBUFFER_DEFAULT, TEXTURE, or RENDERBUFFER, identifying
the type of object which contains the attached image."

And then it clarifies further:

"If the value of FRAMEBUFFER_ATTACHMENT_OBJECT_TYPE is NONE, then
either no framebuffer is bound to target; or the default framebuffer is
bound, attachment is DEPTH or STENCIL, and the number of depth or stencil
bits, respectively, is zero"

Currently, if the default framebuffer is bound, we always return
GL_FRAMEBUFFER_DEFAULT for FRAMEBUFFER_ATTACHMENT_OBJECT_TYPE, but
according to the spec, when GL_DEPTH or GL_STENCIL attachments are
the ones being queried, we should return GL_NONE if they don't exist.

Fixes the following dEQP test:
dEQP-GLES3.functional.state_query.fbo.framebuffer_attachment_x_size_initial

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: "10.6" <mesa-stable@lists.freedesktop.org>

glsl: bail out early in _mesa_ShaderSource if no shaderobj

Patch fixes a crash in conformance test that tries out different
invalid arguments for glShaderSource and glGetShaderSource:

   ES2-CTS.gtf.GL.glGetShaderSource.getshadersource_programhandle

This is a regression from commit:
   04e201d0c02cd30ace5c6fe80e9f021ebb733682

Additions in v2 also fix following failing deqp test:
   dEQP-GLES[2|3].functional.negative_api.shader.shader_source

v2: cleanup function, do check earlier (Iago Toral)

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>

i965/vec4: Detect and delete useless MOVs.

With NIR:

instructions in affected programs:     111508 -> 109193 (-2.08%)
helped:                                507

Without NIR:

instructions in affected programs:     28763 -> 28474 (-1.00%)
helped:                                186

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

prog_to_nir: Use nir_op_dph

Shader-db results on HSW:

   instructions in affected programs:     72 -> 56 (-22.22%)

Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>