radeonsi: apply a mask to gl_SampleMaskIn in the PS prolog

[mesa.git] / src / gallium / state_trackers / clover / core / kernel.cpp

diff --git a/src/gallium/state_trackers/clover/core/kernel.cpp b/src/gallium/state_trackers/clover/core/kernel.cpp
index bce3b525b139efdb8723ad805d2bb28bad710cce..471670532317f28077193f3871341c4ad9233f49 100644 (file)
--- a/src/gallium/state_trackers/clover/core/kernel.cpp
+++ b/src/gallium/state_trackers/clover/core/kernel.cpp
@@ -51,7 +51,7 @@ kernel::launch(command_queue &q,
                const std::vector<size_t> &grid_offset,
                const std::vector<size_t> &grid_size,
                const std::vector<size_t> &block_size) {
-   const auto m = program().binary(q.device());
+   const auto m = program().build(q.device()).binary;
    const auto reduced_grid_size =
       map(divides(), grid_size, block_size);
    void *st = exec.bind(&q, grid_offset);
@@ -76,6 +76,7 @@ kernel::launch(command_queue &q,
                               exec.g_buffers.data(), g_handles.data());
 
    // Fill information for the launch_grid() call.
+   info.work_dim = grid_size.size();
    copy(pad_vector(q, block_size, 1), info.block);
    copy(pad_vector(q, reduced_grid_size, 1), info.grid);
    info.pc = find(name_equals(_name), m.syms).offset;
@@ -89,6 +90,8 @@ kernel::launch(command_queue &q,
                              exec.sviews.size(), NULL);
    q.pipe->bind_sampler_states(q.pipe, PIPE_SHADER_COMPUTE, 0,
                                exec.samplers.size(), NULL);
+
+   q.pipe->memory_barrier(q.pipe, PIPE_BARRIER_GLOBAL_BUFFER);
    exec.unbind();
 }
 
@@ -139,7 +142,7 @@ kernel::args() const {
 
 const module &
 kernel::module(const command_queue &q) const {
-   return program().binary(q.device());
+   return program().build(q.device()).binary;
 }
 
 kernel::exec_context::exec_context(kernel &kern) :
@@ -157,9 +160,9 @@ kernel::exec_context::bind(intrusive_ptr<command_queue> _q,
    std::swap(q, _q);
 
    // Bind kernel arguments.
-   auto &m = kern.program().binary(q->device());
+   auto &m = kern.program().build(q->device()).binary;
    auto margs = find(name_equals(kern.name()), m.syms).args;
-   auto msec = find(type_equals(module::section::text), m.secs);
+   auto msec = find(type_equals(module::section::text_executable), m.secs);
    auto explicit_arg = kern._args.begin();
 
    for (auto &marg : margs) {
@@ -177,7 +180,7 @@ kernel::exec_context::bind(intrusive_ptr<command_queue> _q,
          break;
       }
       case module::argument::grid_offset: {
-         for (cl_uint x : pad_vector(*q, grid_offset, 1)) {
+         for (cl_uint x : pad_vector(*q, grid_offset, 0)) {
             auto arg = argument::create(marg);
 
             arg->set(sizeof(x), &x);