freedreno/ir3: add some ubo range related asserts

And a comment..  since we are mixing units of bytes/dwords/vec4,
hopefully this will avoid some unit confusion.

Signed-off-by: Rob Clark <robdclark@chromium.org>

diff --git a/src/freedreno/ir3/ir3_context.c b/src/freedreno/ir3/ir3_context.c
index f822e9e..da1e148 100644 (file)
--- a/src/freedreno/ir3/ir3_context.c
+++ b/src/freedreno/ir3/ir3_context.c
@@ -123,7 +123,10 @@ ir3_context_init(struct ir3_compiler *compiler,
         *
         * Immediates go last mostly because they are inserted in the CP pass
         * after the nir -> ir3 frontend.
+        *
+        * Note UBO size in bytes should be aligned to vec4
         */
+       debug_assert((ctx->so->shader->ubo_state.size % 16) == 0);
        unsigned constoff = align(ctx->so->shader->ubo_state.size / 16, 4);
        unsigned ptrsz = ir3_pointer_size(ctx);
 

diff --git a/src/freedreno/ir3/ir3_shader.c b/src/freedreno/ir3/ir3_shader.c
index 46eba2a..d1d7488 100644 (file)
--- a/src/freedreno/ir3/ir3_shader.c
+++ b/src/freedreno/ir3/ir3_shader.c
@@ -131,7 +131,8 @@ void * ir3_shader_assemble(struct ir3_shader_variant *v, uint32_t gpu_id)
         * the compiler (to worst-case value) since we don't know in
         * the assembler what the max addr reg value can be:
         */
-       v->constlen = MIN2(255, MAX2(v->constlen, v->info.max_const + 1));
+       v->constlen = MAX2(v->constlen, v->info.max_const + 1);
+       debug_assert(v->constlen < 256);
 
        fixup_regfootprint(v, gpu_id);
 

diff --git a/src/gallium/drivers/freedreno/ir3/ir3_gallium.c b/src/gallium/drivers/freedreno/ir3/ir3_gallium.c
index 65fb756..0f4427f 100644 (file)
--- a/src/gallium/drivers/freedreno/ir3/ir3_gallium.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_gallium.c
@@ -265,10 +265,13 @@ emit_user_consts(struct fd_context *ctx, const struct ir3_shader_variant *v,
                if (state->range[i].start < state->range[i].end &&
                        constbuf->enabled_mask & (1 << i)) {
 
+                       uint32_t size = state->range[i].end - state->range[i].start;
+                       uint32_t offset = cb->buffer_offset + state->range[i].start;
+                       debug_assert((state->range[i].offset % 16) == 0);
+                       debug_assert((size % 16) == 0);
+                       debug_assert((offset % 16) == 0);
                        ctx->emit_const(ring, v->type, state->range[i].offset / 4,
-                                                       cb->buffer_offset + state->range[i].start,
-                                                       (state->range[i].end - state->range[i].start) / 4,
-                                                       cb->user_buffer, cb->buffer);
+                                                       offset, size / 4, cb->user_buffer, cb->buffer);
                }
        }
 }