AuroraMiddleware/SPIRV-Cross
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add option to control use of temp vars for OpVectorShuffle.

Browse Source
This commit is contained in:
Bill Hollings 2017-12-31 18:55:04 -05:00
parent 95910ddd5a
commit 1845f31397
2 changed files with 15 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
spirv_glsl.cpp
View File

@ -5567,11 +5567,13 @@ void CompilerGLSL::emit_instruction(const Instruction &instruction)
shuffle = true;
string expr;
bool trivial_forward;
bool should_fwd, trivial_forward;
if (shuffle)
{
trivial_forward = !expression_is_forwarded(vec0) && !expression_is_forwarded(vec1);
bool allow_fwd = !options.force_temp_use_for_two_vector_shuffles;
should_fwd = allow_fwd && should_forward(vec0) && should_forward(vec1);
trivial_forward = allow_fwd && !expression_is_forwarded(vec0) && !expression_is_forwarded(vec1);
// Constructor style and shuffling from two different vectors.
vector<string> args;
@ -5586,6 +5588,7 @@ void CompilerGLSL::emit_instruction(const Instruction &instruction)
}
else
{
should_fwd = should_forward(vec0);
trivial_forward = !expression_is_forwarded(vec0);
// We only source from first vector, so can use swizzle.
@ -5600,7 +5603,7 @@ void CompilerGLSL::emit_instruction(const Instruction &instruction)
// A shuffle is trivial in that it doesn't actually *do* anything.
// We inherit the forwardedness from our arguments to avoid flushing out to temporaries when it's not really needed.
emit_op(result_type, id, expr, should_forward(vec0) && should_forward(vec1), trivial_forward);
emit_op(result_type, id, expr, should_fwd, trivial_forward);
break;
}
@ -6099,8 +6102,8 @@ void CompilerGLSL::emit_instruction(const Instruction &instruction)
register_read(ops[1], ops[2], should_forward(ops[2]));
break;
// OpAtomicStore unimplemented. Not sure what would use that.
// OpAtomicLoad seems to only be relevant for atomic counters.
// OpAtomicStore unimplemented. Not sure what would use that.
// OpAtomicLoad seems to only be relevant for atomic counters.
case OpAtomicIIncrement:
forced_temporaries.insert(ops[1]);

7
spirv_glsl.hpp
View File

@ -90,6 +90,13 @@ public:
// If disabled on older targets, binding decorations will be stripped.
bool enable_420pack_extension = true;
// If true, all two-vector OpVectorShuffles will be forced to use a temporary variable.
// When multiple consecutive two-vector shuffles are performed, using temporary variables
// can avoid an exponential expansion of code generated for vector constructors.
// Since each constructor requires 4 arguments, shuffles upon shuffles can grow
// the number of constructor arguments and code generated by a power of 4.
bool force_temp_use_for_two_vector_shuffles = false;
enum Precision
{
DontCare,