Transform this into b2f(and(a, b)).
total instructions in shared programs: 6190291 -> 6189225 (-0.02%)
instructions in affected programs: 267247 -> 266181 (-0.40%)
helped: 866
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
There were some bugs, and the code was really difficult to follow. We
would optimize
min(max(x, b), 1.0) into max(sat(x), b)
but not pay attention to the order of min/max and also do
max(min(x, b), 1.0) into max(sat(x), b)
Corrects four shaders from Champions of Regnum that do
min(max(x, 1), 10)
and corrects rendering of Mass Effect under VMware Workstation.
Cc: "10.4 10.5" <mesa-stable@lists.freedesktop.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=89180
Reviewed-by: Abdiel Janulgue <abdiel.janulgue@linux.intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Lots of shaders divide by exp2(...) which we turn into a multiplication
by the reciprocal. We can avoid the reciprocal by simply negating exp2's
argument.
total instructions in shared programs: 5947154 -> 5946695 (-0.01%)
instructions in affected programs: 118661 -> 118202 (-0.39%)
helped: 380
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Cuts an instruction from two shaders in Tesseract, by allowing the
(x+y) cmp 0 -> x cmp -y optimization to take place.
instructions in affected programs: 1198 -> 1194 (-0.33%)
Reviewed-by: Eric Anholt <eric@anholt.net>
The optimization in commit d056863b covers these cases, which were the
first optimizations I added to the GLSL compiler.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Helps a small number of vertex shaders in the games Dungeon Defenders
and Shank, as well as an internal benchmark.
instructions in affected programs: 2801 -> 2719 (-2.93%)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
v2: - Output max(saturate(x),b) instead of saturate(max(x,b))
- Make sure we do component-wise comparison for vectors (Ian Romanick)
v3: - Add missing condition where the outer constant value is > 0.0 and
inner constant is 1.0.
- Fix comments to show that the optimization is a commutative operation
(Matt Turner)
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Abdiel Janulgue <abdiel.janulgue@linux.intel.com>
v2: - Output min(saturate(x),b) instead of saturate(min(x,b)) suggested by Ilia Mirkin
- Make sure we do component-wise comparison for vectors (Ian Romanick)
v3: - Add missing condition where the outer constant value is zero and
inner constant is < 1
- Fix comments to reflect we are doing a commutative operation (Matt Turner)
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Abdiel Janulgue <abdiel.janulgue@linux.intel.com>
v2: - Check that the base type is float (Ian Romanick)
v3: - Make sure comments reflect that we are doing a commutative operation
- Add missing condition where the inner constant is 1.0 and outer constant is 0.0
- Make indexing of operands easier to read (Matt Turner)
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Abdiel Janulgue <abdiel.janulgue@linux.intel.com>
Mesa has an optimization that converts expressions like "v.x + v.y + v.z
+ v.w" into dot(v, 1.0). And therein lies the rub: the other operand to
the dot-product is always a float... even if the vector is an ivec or
uvec. This results in an assertion failure in ir_builder.
If the base type of the operand is not float, don't try the
optimization. Dot-product is not valid on integer data.
Fixes piglit vs-integer-reduction.shader_test and OpenGL ES conformance
test ES2-CTS.gtf.GL2Tests.glGetUniform.glGetUniform.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Christoph Brill <egore911@gmail.com>
The next patch will introduce an optimization that only works when
integers are not represented as floating point values.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cuts a small handful of instructions in Serious Sam 3:
instructions in affected programs: 4692 -> 4666 (-0.55%)
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
opt_algebraic was translating lrp(x, 0, a) into add(x, -mul(x, a)).
Unfortunately, this references "x" twice, which is invalid in the IR,
leading to assertion failures in the validator.
Normally, cloning IR solves this. However, "x" could actually be an
arbitrary expression tree, so copying it could result in huge piles
of wasted computation. This is why we avoid reusing subexpressions.
Instead, transform it into mul(x, add(1.0, -a)), which is equivalent
but doesn't need two references to "x".
Fixes a regression since d5fa8a9562, which isn't in any stable
branches. Fixes 18 shaders in shader-db (bastion and yofrankie).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
v2: Fix pasteo of an extra abs being inserted (caught by many). Rewrite
to drop the silly switch statement.
Reviewed-by: Matt Turner <mattst88@gmail.com> (v1)
On Haswell, POW takes 24 cycles, while EXP2 only takes 14. Plus, using
POW requires putting 2.0 in a register, while EXP2 doesn't.
I believe that EXP2 will be faster than POW on basically all GPUs, so
it makes sense to optimize it.
Looking at the savage2 subset of shader-db:
total instructions in shared programs: 113225 -> 113179 (-0.04%)
instructions in affected programs: 2139 -> 2093 (-2.15%)
instances of 'math pow': 795 -> 749 (-6.14%)
instances of 'math exp': 389 -> 435 (11.8%)
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
The comment was stale, because the lowering in question wasn't happening
in lower_instructions.cpp. Presumably if the lowering ever moves there,
we can plumb the lowering mask through to opt_algebraic.
total instructions in shared programs: 1618696 -> 1616810 (-0.12%)
instructions in affected programs: 243018 -> 241132 (-0.78%)
GAINED: 0
LOST: 0
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
I noticed this in a shader in Unigine Heaven that was spilling. While it
doesn't really reduce register pressure, it shaves a few instructions
anyway (7955 -> 7882).
v2: Fix turning "0 >> x" into "x" instead of "0" (caught by Erik
Faye-Lund).
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
While ir_builder is slightly less efficient, we're only increasing the
work when there's actual optimization being done, and it's way more
readable code.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Matt and I had each screwed up these common required patterns recently, in
ways that wouldn't have been noticed for a long time if not for code
review. Just enforce it in the caller so that we don't rely on code
review catching these bugs.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Two extra instructions in some heroesofnewerth shaders, but a win for
everything else.
total instructions in shared programs: 1531352 -> 1530815 (-0.04%)
instructions in affected programs: 121898 -> 121361 (-0.44%)
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
