glsl: Use array deref for access to vector components

We've assumed that we could lower per-component vector access from

  vec[i] = scalar

to

  vec = ir_triop_vector_insert(vec, scalar, i)

but with SSBOs (and compute shader SLM and tesselation outputs) this is
no longer valid. If a vector is "externally visible", multiple threads
can write independent components simultaneously. With lowering to
ir_triop_vector_insert, each thread read the entire vector, changes one
component, then writes out the entire vector. This is racy.

Instead of generating a ir_binop_vector_extract when we see v[i], we
generate ir_dereference_array. We then add a lowering pass to lower the
ir_dereference_array to ir_binop_vector_extract for rvalues and for to
vector_insert for lvalues in a separate lowering pass.

The resulting IR is the same as before, but we now have a window between
ast->ir conversion and the lowering pass where v[i] appears in the IR as
an array deref. This lets us run lowering passes that lower the vector
access to I/O (eg for SSBO load/store) before we lower the per-component
access to full vector writes.

Reviewed-by: Jordan Justen <[email protected]>
Signed-off-by: Kristian Høgsberg Kristensen <[email protected]>

1 files changed, 13 insertions, 1 deletions

diff --git a/src/glsl/lower_ubo_reference.cpp b/src/glsl/lower_ubo_reference.cpp
index 24806ac6ce9..b74aa3d0630 100644
--- a/src/glsl/lower_ubo_reference.cpp
+++ b/src/glsl/lower_ubo_reference.cpp
@@ -390,7 +390,19 @@ lower_ubo_reference_visitor::setup_for_load_or_store(ir_variable *var,
       case ir_type_dereference_array: {
          ir_dereference_array *deref_array = (ir_dereference_array *) deref;
          unsigned array_stride;
-         if (deref_array->array->type->is_matrix() && *row_major) {
+         if (deref_array->array->type->is_vector()) {
+            /* We get this when storing or loading a component out of a vector
+             * with a non-constant index. This happens for v[i] = f where v is
+             * a vector (or m[i][j] = f where m is a matrix). If we don't
+             * lower that here, it gets turned into v = vector_insert(v, i,
+             * f), which loads the entire vector, modifies one component and
+             * then write the entire thing back.  That breaks if another
+             * thread or SIMD channel is modifying the same vector.
+             */
+            array_stride = 4;
+            if (deref_array->array->type->is_double())
+               array_stride *= 2;
+         } else if (deref_array->array->type->is_matrix() && *row_major) {
             /* When loading a vector out of a row major matrix, the
              * step between the columns (vectors) is the size of a
              * float, while the step between the rows (elements of a