[Mesa-dev] [PATCH 1/3] glsl: Add new barrier functions for compute shaders

[Jordan Justen]

When these functions are called in GLSL code, we create an intrinsic
function call:

 * groupMemoryBarrier => __intrinsic_group_memory_barrier
 * memoryBarrierAtomicCounter => __intrinsic_memory_barrier_atomic_counter
 * memoryBarrierBuffer => __intrinsic_memory_barrier_buffer
 * memoryBarrierImage => __intrinsic_memory_barrier_image
 * memoryBarrierShared => __intrinsic_memory_barrier_shared

Signed-off-by: Jordan Justen <jordan.l.justen at intel.com>
---

Notes:
    git://people.freedesktop.org/~jljusten/mesa cs-shader-barrier-funcs-v1
    http://patchwork.freedesktop.org/bundle/jljusten/cs-shader-barrier-funcs-v1
    
    Tested with the OpenGLES 3.1 CTS test suite on my cs branch. With
    these patches reverted, these tests fail:
    
     * ES31-CTS.compute_shader.work-group-size
     * ES31-CTS.compute_shader.shared-simple
     * ES31-CTS.compute_shader.shared-struct
     * ES31-CTS.compute_shader.atomic-case1
     * ES31-CTS.compute_shader.shared-indexing

 src/glsl/builtin_functions.cpp | 134 ++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 133 insertions(+), 1 deletion(-)

diff --git a/src/glsl/builtin_functions.cpp b/src/glsl/builtin_functions.cpp
index 509a57b..2bb7bcb 100644
--- a/src/glsl/builtin_functions.cpp
+++ b/src/glsl/builtin_functions.cpp
@@ -459,9 +459,15 @@ fp64(const _mesa_glsl_parse_state *state)
 }
 
 static bool
+compute_shader(const _mesa_glsl_parse_state *state)
+{
+   return state->stage == MESA_SHADER_COMPUTE;
+}
+
+static bool
 barrier_supported(const _mesa_glsl_parse_state *state)
 {
-   return state->stage == MESA_SHADER_COMPUTE ||
+   return compute_shader(state) ||
           state->stage == MESA_SHADER_TESS_CTRL;
 }
 
@@ -715,6 +721,11 @@ private:
    ir_function_signature *_EndStreamPrimitive(builtin_available_predicate avail,
                                               const glsl_type *stream_type);
    B0(barrier)
+   B0(groupMemoryBarrier)
+   B0(memoryBarrierAtomicCounter)
+   B0(memoryBarrierBuffer)
+   B0(memoryBarrierImage)
+   B0(memoryBarrierShared)
 
    BA2(textureQueryLod);
    B1(textureQueryLevels);
@@ -787,6 +798,26 @@ private:
       builtin_available_predicate avail);
    ir_function_signature *_memory_barrier(
       builtin_available_predicate avail);
+   ir_function_signature *_group_memory_barrier_intrinsic(
+      builtin_available_predicate avail);
+   ir_function_signature *_group_memory_barrier(
+      builtin_available_predicate avail);
+   ir_function_signature *_memory_barrier_atomic_counter_intrinsic(
+      builtin_available_predicate avail);
+   ir_function_signature *_memory_barrier_atomic_counter(
+      builtin_available_predicate avail);
+   ir_function_signature *_memory_barrier_buffer_intrinsic(
+      builtin_available_predicate avail);
+   ir_function_signature *_memory_barrier_buffer(
+      builtin_available_predicate avail);
+   ir_function_signature *_memory_barrier_image_intrinsic(
+      builtin_available_predicate avail);
+   ir_function_signature *_memory_barrier_image(
+      builtin_available_predicate avail);
+   ir_function_signature *_memory_barrier_shared_intrinsic(
+      builtin_available_predicate avail);
+   ir_function_signature *_memory_barrier_shared(
+      builtin_available_predicate avail);
 
    ir_function_signature *_shader_clock_intrinsic(builtin_available_predicate avail,
                                                   const glsl_type *type);
@@ -968,6 +999,22 @@ builtin_builder::create_intrinsics()
                 _shader_clock_intrinsic(shader_clock,
                                         glsl_type::uvec2_type),
                 NULL);
+
+   add_function("__intrinsic_group_memory_barrier",
+                _group_memory_barrier_intrinsic(compute_shader),
+                NULL);
+   add_function("__intrinsic_memory_barrier_atomic_counter",
+                _memory_barrier_atomic_counter_intrinsic(compute_shader),
+                NULL);
+   add_function("__intrinsic_memory_barrier_buffer",
+                _memory_barrier_buffer_intrinsic(compute_shader),
+                NULL);
+   add_function("__intrinsic_memory_barrier_image",
+                _memory_barrier_image_intrinsic(compute_shader),
+                NULL);
+   add_function("__intrinsic_memory_barrier_shared",
+                _memory_barrier_shared_intrinsic(compute_shader),
+                NULL);
 }
 
 /**
@@ -2081,6 +2128,11 @@ builtin_builder::create_builtins()
                 _EndStreamPrimitive(gs_streams, glsl_type::int_type),
                 NULL);
    add_function("barrier", _barrier(), NULL);
+   add_function("groupMemoryBarrier", _group_memory_barrier(compute_shader), NULL);
+   add_function("memoryBarrierAtomicCounter", _memory_barrier_atomic_counter(compute_shader), NULL);
+   add_function("memoryBarrierBuffer", _memory_barrier_buffer(compute_shader), NULL);
+   add_function("memoryBarrierImage", _memory_barrier_image(compute_shader), NULL);
+   add_function("memoryBarrierShared", _memory_barrier_shared(compute_shader), NULL);
 
    add_function("textureQueryLOD",
                 _textureQueryLod(texture_query_lod, glsl_type::sampler1D_type,  glsl_type::float_type),
@@ -5294,6 +5346,86 @@ builtin_builder::_shader_clock(builtin_available_predicate avail,
    return sig;
 }
 
+ir_function_signature *
+builtin_builder::_group_memory_barrier_intrinsic(builtin_available_predicate avail)
+{
+   MAKE_INTRINSIC(glsl_type::void_type, avail, 0);
+   return sig;
+}
+
+ir_function_signature *
+builtin_builder::_group_memory_barrier(builtin_available_predicate avail)
+{
+   MAKE_SIG(glsl_type::void_type, avail, 0);
+   body.emit(call(shader->symbols->get_function("__intrinsic_group_memory_barrier"),
+                  NULL, sig->parameters));
+   return sig;
+}
+
+ir_function_signature *
+builtin_builder::_memory_barrier_atomic_counter_intrinsic(builtin_available_predicate avail)
+{
+   MAKE_INTRINSIC(glsl_type::void_type, avail, 0);
+   return sig;
+}
+
+ir_function_signature *
+builtin_builder::_memory_barrier_atomic_counter(builtin_available_predicate avail)
+{
+   MAKE_SIG(glsl_type::void_type, avail, 0);
+   body.emit(call(shader->symbols->get_function("__intrinsic_memory_barrier_atomic_counter"),
+                  NULL, sig->parameters));
+   return sig;
+}
+
+ir_function_signature *
+builtin_builder::_memory_barrier_buffer_intrinsic(builtin_available_predicate avail)
+{
+   MAKE_INTRINSIC(glsl_type::void_type, avail, 0);
+   return sig;
+}
+
+ir_function_signature *
+builtin_builder::_memory_barrier_buffer(builtin_available_predicate avail)
+{
+   MAKE_SIG(glsl_type::void_type, avail, 0);
+   body.emit(call(shader->symbols->get_function("__intrinsic_memory_barrier_buffer"),
+                  NULL, sig->parameters));
+   return sig;
+}
+
+ir_function_signature *
+builtin_builder::_memory_barrier_image_intrinsic(builtin_available_predicate avail)
+{
+   MAKE_INTRINSIC(glsl_type::void_type, avail, 0);
+   return sig;
+}
+
+ir_function_signature *
+builtin_builder::_memory_barrier_image(builtin_available_predicate avail)
+{
+   MAKE_SIG(glsl_type::void_type, avail, 0);
+   body.emit(call(shader->symbols->get_function("__intrinsic_memory_barrier_image"),
+                  NULL, sig->parameters));
+   return sig;
+}
+
+ir_function_signature *
+builtin_builder::_memory_barrier_shared_intrinsic(builtin_available_predicate avail)
+{
+   MAKE_INTRINSIC(glsl_type::void_type, avail, 0);
+   return sig;
+}
+
+ir_function_signature *
+builtin_builder::_memory_barrier_shared(builtin_available_predicate avail)
+{
+   MAKE_SIG(glsl_type::void_type, avail, 0);
+   body.emit(call(shader->symbols->get_function("__intrinsic_memory_barrier_shared"),
+                  NULL, sig->parameters));
+   return sig;
+}
+
 /** @} */
 
 /******************************************************************************/
-- 
2.6.2