r600g: don't enable tiling for STAGING and STREAM usage cases

Reviewed by: Christian König <christian.koenig@amd.com>

winsys/radeon: add usage parameter to cs_is_buffer_referenced

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

r600g: initialize DB_DEPTH_CONTROL at beginning of CS

Otherwise streamout with rasterizer discard will make the kernel upset
if the state tracker doesn't set a depth-stencil-alpha state.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

r600g: initialize VS_GS_OUT_PRIM_TYPE

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

r600g: initialize SX_SURFACE_SYNC

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

r600g: if pixel shader is NULL, bind a dummy one

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

r600g: add support for missing colorbuffer formats

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

r600g: remove support for SCALED colorbuffer formats

Unused by the current stack and APIs, therefore untestable.
It was used to facilitate the transition to integers.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

r600g: permit blitting between textures with STREAM and STAGING usage

Eventually I'd like to make every format blittable, so that the function
can go away.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

r600g: fix and improve rasterizer discard for r600-r700

For polygons, we have been using face culling with success, but that doesn't
work for points and lines.

Setting the point size and line width to 0 fixes it.

Also improve it even more by setting SCREEN_SCISSOR to a zero area.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

r600g: rework rasterizer discard for evergreen

Implement it right using STRMOUT_CONFIG.RAST_STREAM. This fixes rasterizer
discard with points and lines.

This also adds another derived state. It's a combination of rasterizer discard
and streamout enable.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

r600g: make some scissor regs invariant on r600-r700

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

r600g: make some scissor regs invariant on evergreen

We only need one scissor for the framebuffer.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

r600g: rework scissor for r600-r700

We must use VPORT_SCISSOR, because that's the only one we can use for multiple
scissor rectangles in ARB_viewport_array.

R700 can use the VPORT_SCISSOR_ENABLE bit, but R600 doesn't have that and must
emit a 8192x8192 rectangle if scissor is disabled.

This commit also cleanups magic numbers in create_rs_state.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

r600g: rework scissor for evergreen

VPORT_SCISSOR is the OpenGL scissor. How do I know? Because there are
16 of them just like GL4.1 has multiple scissor rectangles.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

r600g: remove obsolete todo comments

Also use XXX in the other ones, because it's the most used word for that
purpose in Mesa.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

r600g: correctly handle queries which don't need begin_query

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

r600g: don't suspend timer queries for u_blitter

Timer queries should be able to measure the time spent in u_blitter as well.

Queries are split into two groups: the timer ones and the others (streamout,
occlusion), because we should only suspend non-timer queries for u_blitter,
and later if the non-timer queries are suspended, the context flush should
only suspend and resume the timer queries.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

r600g: release query buffers in destroy_query

This fixes a memory leak introduced with the rework.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

r600g: move all query code into r600_query.c

And rename or inline functions where appropriate.
There is no reason to keep this stuff in r600_hw_context.c.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

Revert "gallium: add flag PIPE_TRANSFER_MAP_PERMANENTLY"

This reverts commit 0950086376b1c8b7fb89eda81ed7f2f06dee58bc.

It was decided to refactor the transfer API instead of adding workarounds
to address the performance issues.

i915: fix wrong rendering of gl_PointSize on Pineview

The current code would ignore the point size specified by gl_PointSize
builtin variable in vertex shader on Pineview. This patch servers as
fixing that.

This patch fixes the following issues on Pineview:
webglc: https://cvs.khronos.org/svn/repos/registry/trunk/public/webgl/sdk/tests/conformance/rendering/point-size.html
piglit: glsl-vs-point-size

NOTE: This is a candidate for stable release branches.

v2: pick Eric's nice tip for fixing this issue in hardware rendering.
v3: the last arg of EMIT_ATTR specify the size in _byte_. (Eric)

Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

tnl: let _TNL_ATTRIB_POINTSIZE do not depend on ctx->VertexProgram._Enabled

We may specify the point size in a glsl vertex shader.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=46311
piglit: glsl-vs-point-size

NOTE: This is a candidate for stable release branches.

Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

r300/compiler: Add support for inline literals

On R500 chips, shader instructions can take 7-bit (3-bit mantissa, 4-bit
exponent) floating point values as inputs in place of registers.

vl: fix shader in/out numbering

Fix all the other wrong numberings.

Signed-off-by: Christian König <deathsimple@vodafone.de>

egl: Also put a symlink from libEGL.so in lib/ for now.

This fixes the egl_gallium.so driver build when no system libEGL.so is
present, since it's relying on the lib/ to build against until it gets
converted to automake.

i915: Fix i830 polygon stipple from PBOs.

This is a direct port of the i915 patch in
a856da63247a4b403f6350914f732e14d1530ed1.

Fixes glean's pbo test.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=41372
Reviewed-by: Eric Anholt <eric@anholt.net>
NOTE: This is a candidate for release branches.

i915: Compute maximum number of verts using the actual batchbuffer size.

We were looking at the size of batch.map for how big the batchbuffer
was, but on 865 we just use a single-page batchbuffer due to hardware
limits.

v2: Removed check for sizeof map < bo->size, since that's always false.
    [change by anholt]
NOTE: This is a candidate for release branches.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=41495

i830: Compute initial number of vertices from remaining batch space

In order to prevent an overflow of the batch buffer when emitting
triangles, we need to limit the initial primitive to fit within the
current batch. To do we need to measure the remaining space and thence
compute the maximum number of vertices that fit into that space.

Reported-by: Kurt Roeckx <kurt@roeckx.be>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=41495
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Eric Anholt <eric@anholt.net>
NOTE: This is a candidate for release branches.

dri/i915: Fix off-by-one in i830 clip region size.

The hardware, like i915, uses an inclusive bounds on min and max for
the drawing rectangle, but we were providing a number for exclusive.
The number of bits used by the hardware only covers this value going
up to the maximum size, so when we programmed 2048 as the maximum
inclusive X, it saw a maximum X of 0 and clipped all rendering.  This
caused rendering failures in gnome-shell.

Fixes piglit fbo-maxsize.

v2: dropped changes to the blitter, which does use an exclusive x2, y2.
    [change by anholt]

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=45558
Reviewed-by: Eric Anholt <eric@anholt.net>
NOTE: This is a candidate for release branches.

intel: Don't enable GL_ARB_draw_instanced pre-gen4.

swtnl doesn't handle this extension.

i915: Fix piglit fbo-nodepth-test on i830.

This is a direct port of fc4fba52cf7e9616c70dd76b4d6bdba6582e157b from
i915, and fixes GPU hangs when running piglit.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=41372
Reviewed-by: Eric Anholt <eric@anholt.net>
NOTE: This is a candidate for release branches.

vl/compositor: fix shader in/out numbering

Michel pointed out that my assumption of a global
index namespace is incorrect and breaks r300g.

Signed-off-by: Christian König <deathsimple@vodafone.de>

st/vdpau: implement OutputSurfaceQueryPutBitsYCbCrCapabilities

Signed-off-by: Christian König <deathsimple@vodafone.de>

st/vdpau: implement OutputSurfacePutBitsYCbCr

v2: the csc_matrix is optional

Signed-off-by: Christian König <deathsimple@vodafone.de>

st/vdpau: implement VideoSurfaceGetBitsYCbCr

Signed-off-by: Christian König <deathsimple@vodafone.de>

st/vdpau: implement OutputSurfaceQueryGetPutBitsNativeCapabilities

Signed-off-by: Christian König <deathsimple@vodafone.de>

st/vdpau: implement OutputSurfaceGetBitsNative

Signed-off-by: Christian König <deathsimple@vodafone.de>

st/vdpau: implement OutputSurfacePutBitsNative

Signed-off-by: Christian König <deathsimple@vodafone.de>

st/vdpau: implement OutputSurfaceQueryPutBitsIndexedCapabilities

Signed-off-by: Christian König <deathsimple@vodafone.de>

st/vdpau: fix default swizzle for Output/Bitmap samplers

Signed-off-by: Christian König <deathsimple@vodafone.de>

st/vdpau: add support for per vertex colors

Signed-off-by: Christian König <deathsimple@vodafone.de>

st/vdpau: implement BitmapSurfaceQueryCapabilities

Signed-off-by: Christian König <deathsimple@vodafone.de>

st/vdpau: implement OutputSurfaceRenderBitmapSurface

Signed-off-by: Christian König <deathsimple@vodafone.de>

st/vdpau: implement BitmapSurfaceGetParameters

Signed-off-by: Christian König <deathsimple@vodafone.de>

st/vdpau: implement BitmapSurfacePutBitsNative

Signed-off-by: Christian König <deathsimple@vodafone.de>

st/vdpau: implement BitmapSurfaceCreate/Destroy

Signed-off-by: Christian König <deathsimple@vodafone.de>

st/vdpau: avoid an unnessasary copy

Delay the actual mixer rendering until we really
know which surface is the real destination.

Signed-off-by: Christian König <deathsimple@vodafone.de>

vl/csc: simplify matrix handling

A csc matrix is only 4x3 not 4x4, also define a VDPAU compatible type for it.

Signed-off-by: Christian König <deathsimple@vodafone.de>

vl/compositor: add support for per layer dst areas

Signed-off-by: Christian König <deathsimple@vodafone.de>

vl/compositor: add per vertex color suport

Used in subtitles, for example.

Signed-off-by: Christian König <deathsimple@vodafone.de>

vl/compositor: replace pipe_video_rect with u_rect

So we support things like flipping also.

Signed-off-by: Christian König <deathsimple@vodafone.de>

vl/compositor: split shaders and state

Signed-off-by: Christian König <deathsimple@vodafone.de>

Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."

This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.

llvm-3.1svn r151687 makes MemoryObject accessor members const again.

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>

Revert "st/mesa: don't unreference user attribs up front."

This reverts commit 72931ca4b9fb1002f5b62b74f7f7f32e94e80fde.

This commit caused a few piglit regressions (quad-invariance, draw-batch,
etc) with the vmware svga driver.

vl: move dirty area handling into winsys abstraction

Fixing uninitialized areas in SwapBuffers mode.

Signed-off-by: Christian König <deathsimple@vodafone.de>

vl: hide X latency by using asynchronous requests

Signed-off-by: Christian König <deathsimple@vodafone.de>

vl: use SwapBuffers instead of CopyBuffers

This should speed things up a bit, but also shows
some bugs with the kernel implementation.

v2: require xcb-dri2 version 1.8

Signed-off-by: Christian König <deathsimple@vodafone.de>

st/xvmc: remove dst buffer texture workaround

Signed-off-by: Christian König <deathsimple@vodafone.de>

vl: rewrite vl DRI backend using XCB

v2: also set array_size of texture

Signed-off-by: Christian König <deathsimple@vodafone.de>

egl: Don't include strings.h on Windows.

strings.h is needed for the declaration of strcasecmp(), but only on
non-Windows platforms.

mesa: let GL3 buf obj queries not depend on opengl major version

While the ARB_map_buffer_range extension spec says nothing about these
queries -- they were added in GL 3.0 --, it seems like this could be an
error in the extension spec.  This is one of the extensions, like
ARB_framebuffer_object, that "back ports" OpenGL 3.0 functionality to
previous versions.  These extensions are supposed to provide identical
functionality to OpenGL 3.0.  The other cases of mismatches have been
determined to be bugs in the extension specs.

And tools like apitrace rely on such queries to function properly.

Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Signed-off-by: José Fonseca <jfonseca@vmware.com>
Acked-by: Brian Paul <brianp@vmware.com>
Acked-by: Ian Romanick <ian.d.romanick@intel.com>

gallivm: Update comments and prototype of vector-selects.

No runtime behavior change. As vector selects are still not very well
supported by LLVM.

docs: remove mention of components no longer in main Mesa tree

st/mesa: remove old assertion

r300g: init chan=0 to silence warning

i965: Disable PrimitiveID upload.

We currently don't support gl_PrimitiveID, and I believe asking the
hardware to generate it results in vertex cache invalidations.

This could result in slowdowns for applications that use gl_InstanceID,
which would be counter-productive.  Just turn it off for now.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Enable the GL_ARB_draw_instanced extension.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Fix swizzles for system values such as gl_InstanceID.

visit(ir_variable *) sets dst_reg::writemask to the appropriate channel
for system values.  Unfortunately, visit(ir_dereference_variable *) then
calls swizzle_for_size, which for a float, sets the swizzle to .x.

This works for gl_VertexID, since we store it in the .x component (see
brw_draw_upload.c:732 - VID), but fails for gl_InstanceID (IID) since we
store it in the .y channel.

To fix this, avoid calling swizzle_for_size on ir_var_system_values.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

glsl: Make gl_InstanceID available with GL_ARB_draw_instanced extension

Originally ARB_draw_instanced only specified that ARB decorated name.
Since no vendor actually implemented that behavior and some apps use
the undecorated name, the extension now specifies that both names are
available.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Marek Olšák <maraeo@gmail.com>

mesa: Add missing error check for first < 0 in glDrawArraysInstanced().

Fixes piglit GL_ARB_draw_instanced/negative-arrays-first-negative.

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

mesa: Fix display lists for draw_elements_base_vertex with draw_instanced.

Fixes piglit GL_ARB_draw_elements_base_vertex/dlist-arb_draw_instanced

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

mesa: Fix display list handling for GL_ARB_draw_instanced.

When you called them in a display list compile before, you would just
end up calling through NULL.

Fixes piglit GL_ARB_draw_instanced/dlist.

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

mesa: Fix typo in comment.

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

i965: Fix Gen6+ dynamic state upper bound on older kernels.

Kernels prior to 271d81b84171d84723357ae6d172ec16b0d8139c (March 2011)
don't support relocations outside of the target buffer object.  Rather
than guarding this with a I915_PARAM_HAS_RELAXED_DELTA check, just
smash the bound to 0xfffff001 like we do on Ironlake.

This effectively gives us no upper bound check, just like we did prior
to commit 271d81b84171d84723357ae6d172ec16b0d8139c.

Daniel Vetter would also like to mention that this relies on the guard
page at the end of the GTT.

NOTE: This is a candidate for release branches.
Fixes a regression since 271d81b84171d84723357ae6d172ec16b0d8139c.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=46766
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>

egl/drivers: Convert to automake.

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

egl/main: Convert to automake.

The drivers/ walk-through-subdirs makefile is converted as well so I
didn't need to keep EGL_DRIVERS_DIRS along with the per-driver
HAVE_EGL_DRIVER_WHATEVER.

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

egl: Clean up some flow of EGL platform handling.

The default case code was set up in a separate way, while this makes
it more normal.  I wanted to add code to the explicit x11 platform and
default x11 platform cases in the next commit.

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

egl: Drop the non-builtin-drivers support.

The two drivers both did builtin mode, so drop the non-builtin build
system.

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

egl: Drop _EGL_MAIN entrypoint obfuscation.

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

glx: Convert to automake.

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

mesa: Push the shine table into the tnl module.

All users of the shine table outside of the tnl module
are gone. Move the implementation into the tnl module and
prefix the public functions with _tnl.

Reviewed-by: Alex Deucher <alexdeucher@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Mathias Froehlich <Mathias.Froehlich@web.de>

mesa: Do not invalidate shine tables in compute light positions.

Since the shine tables are now only used in the tnl lighting stage, where
they are validated through the tnl driver function NotifyMaterialChange
called in tnl/t_vb_light.c, we can not omit calling
_mesa_validate_all_lighting_tables (which only validates the shine tables)
in main/light.c.

Reviewed-by: Alex Deucher <alexdeucher@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Mathias Froehlich <Mathias.Froehlich@web.de>

mesa: Avoid revalidating shine tables in _tnl_RasterPos.

Now that _tnl_RasterPos no longer uses the shine tables, avoid
revalidating them.

Reviewed-by: Alex Deucher <alexdeucher@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Mathias Froehlich <Mathias.Froehlich@web.de>

mesa: Avoid using the shine table for _tnl_RasterPos

Use direct computation of pow for computing the shininess
in _tnl_RasterPos. Since the _tnl_RasterPos function is still
used by plenty drivers that do only need the shine table for
_tnl_RasterPos but do not make use of swtnl computations, this
enables pushing down the shine table computation and validation
into the tnl module, which will happen in a followup change.

Reviewed-by: Alex Deucher <alexdeucher@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Mathias Froehlich <Mathias.Froehlich@web.de>

mesa: Avoid explicit invalidation of shine tables.

Since the shine tables are implicitly invalidated by having
a different shininess value than the current one, we can
omit the explicit invalidation of the shine table.

Reviewed-by: Alex Deucher <alexdeucher@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Mathias Froehlich <Mathias.Froehlich@web.de>

mesa: Remove gl_light_attrib::_Flags.

This variable is only used locally in _mesa_update_lighting.

Reviewed-by: Alex Deucher <alexdeucher@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Mathias Froehlich <Mathias.Froehlich@web.de>

mesa: Remove _CosCutoffNeg from light state.

It is only used as a temporary variable during computation of
_CosCutoff. So, don't store it.

Reviewed-by: Alex Deucher <alexdeucher@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Mathias Froehlich <Mathias.Froehlich@web.de>

util: replace format equality test with compatibility test in blit code

This lets us use the resource_copy_region() path when blitting from
R8G8B8A8 to R8G8B8x8, for example.

v2: be smarter when src_format==dst_format

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>

util: better comment for util_is_format_compatible()

mesa: use arrayObj var in more places in client_state()

mesa: asst. fixes for texture arrays and borders

For 1D array textures, there is no border on the height dimension.
For 2D array textures, there is no border on the depth dimension.

mesa: check for no state change in VertexAttribDivisor()

Reviewed-by: José Fonseca <jfonseca@vmware.com>

util: fix assertions in u_blitter.c code

Assertions of the form assert(a && b) should be written as separate assertions
so that you can actually tell which part is false when there's a failure.

Reviewed-by: Marek Olšák <maraeo@gmail.com>

mesa: reorder things in mtypes.h

Move structs, enums, etc so they're in more logical order.  In particular,
the shader and transform feedback-related structs/enums were pretty
scattered around.

svga: fix max_lod clamping in update_tss_binding()

After biasing we need to clamp to be sure we don't exceed the number of
levels in the mipmap.  This fixes an assertion at svga_sampler_view.c:70

v2: simplify the biasing, clamping code per Jose's suggestion.

Reviewed-by: José Fonseca <jfonseca@vmware.com>

st/vdpau: fix use of *.o in Makefile.vdpau

Signed-off-by: Christian König <deathsimple@vodafone.de>

st/xvmc: fix use of *.o in Makefile.xvmc

Signed-off-by: Christian König <deathsimple@vodafone.de>

i965: Avoid blocking on the GPU for setting the HiZ op vertex data.

We need to allocate new space every time to avoid blocking on the last
HiZ op completing.  There are two easy ways to do this:
brw_state_batch() and intel_upload_data().  brw_state_batch() is
simpler and avoids another buffer allocation.

Improves Unigine Tropics performance 0.376416% +/- 0.148722% (n=7).

Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

glcpp: Don't strlen() the output for every token being printed.

The ralloc string appending functions were originally intended for
simple, non-hot-path uses like printing to an info log.

Cuts Unigine Tropics load time by around 20% (6 seconds).

v2: Avoid strlen() on every newline, too.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com> [v1]
Acked-by: José Fonseca <jfonseca@vmware.com> [v1]