panfrost: Unify bifrost_scratchpad with mali_shared_memory

It looks like these are the same structure, so this allows us to reuse
mali_shared_memory across architectures, and dispels with the
Bifrost-specific mystery of the scratchpads... nothing so mysterious
after all, just stack.

Signed-off-by: Alyssa Rosenzweig <[email protected]>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3835>

4 files changed, 18 insertions, 52 deletions

diff --git a/src/gallium/drivers/panfrost/pan_compute.c b/src/gallium/drivers/panfrost/pan_compute.c
index 33618cb6997..f4c28c36429 100644
--- a/src/gallium/drivers/panfrost/pan_compute.c
+++ b/src/gallium/drivers/panfrost/pan_compute.c
@@ -131,7 +131,7 @@ panfrost_launch_grid(struct pipe_context *pipe,
                 .shared_workgroup_count = ~0
         };
 
-        payload->postfix.framebuffer =
+        payload->postfix.shared_memory =
                 panfrost_upload_transient(batch, &shared, sizeof(shared));
 
         /* Invoke according to the grid info */
diff --git a/src/gallium/drivers/panfrost/pan_context.c b/src/gallium/drivers/panfrost/pan_context.c
index 417c900c961..f6f9858e074 100644
--- a/src/gallium/drivers/panfrost/pan_context.c
+++ b/src/gallium/drivers/panfrost/pan_context.c
@@ -141,6 +141,9 @@ panfrost_clear(
         panfrost_batch_clear(batch, buffers, color, depth, stencil);
 }
 
+/* TODO: Bifrost requires just a mali_shared_memory, without the rest of the
+ * framebuffer */
+
 static void
 panfrost_attach_vt_framebuffer(struct panfrost_context *ctx)
 {
@@ -162,7 +165,7 @@ panfrost_attach_vt_framebuffer(struct panfrost_context *ctx)
         }
 
         for (unsigned i = 0; i < PIPE_SHADER_TYPES; ++i)
-                ctx->payloads[i].postfix.framebuffer = batch->framebuffer.gpu;
+                ctx->payloads[i].postfix.shared_memory = batch->framebuffer.gpu;
 }
 
 /* Reset per-frame context, called on context initialisation as well as after
@@ -172,7 +175,7 @@ void
 panfrost_invalidate_frame(struct panfrost_context *ctx)
 {
         for (unsigned i = 0; i < PIPE_SHADER_TYPES; ++i)
-                ctx->payloads[i].postfix.framebuffer = 0;
+                ctx->payloads[i].postfix.shared_memory = 0;
 
         if (ctx->rasterizer)
                 ctx->dirty |= PAN_DIRTY_RASTERIZER;
diff --git a/src/panfrost/include/panfrost-job.h b/src/panfrost/include/panfrost-job.h
index 8027abebf34..d0d56810ecc 100644
--- a/src/panfrost/include/panfrost-job.h
+++ b/src/panfrost/include/panfrost-job.h
@@ -1056,17 +1056,6 @@ struct bifrost_tiler_only {
         u64 zero8;
 } __attribute__((packed));
 
-struct bifrost_scratchpad {
-        u32 zero;
-        u32 flags; // = 0x1f
-        /* This is a pointer to a CPU-inaccessible buffer, 16 pages, allocated
-         * during startup. It seems to serve the same purpose as the
-         * gpu_scratchpad in the SFBD for Midgard, although it's slightly
-         * larger.
-         */
-        mali_ptr gpu_scratchpad;
-} __attribute__((packed));
-
 struct mali_vertex_tiler_postfix {
         /* Zero for vertex jobs. Pointer to the position (gl_Position) varying
          * output from the vertex shader for tiler jobs.
@@ -1100,11 +1089,10 @@ struct mali_vertex_tiler_postfix {
         u64 viewport;
         u64 occlusion_counter; /* A single bit as far as I can tell */
 
-        /* Note: on Bifrost, this isn't actually the FBD. It points to
-         * bifrost_scratchpad instead. However, it does point to the same thing
-         * in vertex and tiler jobs.
-         */
-        mali_ptr framebuffer;
+        /* On Bifrost, this points directly to a mali_shared_memory structure.
+         * On Midgard, this points to a framebuffer (either SFBD or MFBD as
+         * tagged), which embeds a mali_shared_memory structure */
+        mali_ptr shared_memory;
 } __attribute__((packed));
 
 struct midgard_payload_vertex_tiler {
diff --git a/src/panfrost/pandecode/decode.c b/src/panfrost/pandecode/decode.c
index 41b843f7a3b..ef115dfaa0a 100644
--- a/src/panfrost/pandecode/decode.c
+++ b/src/panfrost/pandecode/decode.c
@@ -1809,29 +1809,6 @@ pandecode_uniforms(mali_ptr uniforms, unsigned uniform_count)
         free(ptr);
 }
 
-static void
-pandecode_scratchpad(uintptr_t pscratchpad, int job_no, char *suffix)
-{
-
-        struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(pscratchpad);
-
-        struct bifrost_scratchpad *PANDECODE_PTR_VAR(scratchpad, mem, pscratchpad);
-
-        if (scratchpad->zero) {
-                pandecode_msg("XXX: scratchpad zero tripped");
-                pandecode_prop("zero = 0x%x\n", scratchpad->zero);
-        }
-
-        pandecode_log("struct bifrost_scratchpad scratchpad_%"PRIx64"_%d%s = {\n", pscratchpad, job_no, suffix);
-        pandecode_indent++;
-
-        pandecode_prop("flags = 0x%x", scratchpad->flags);
-        MEMORY_PROP(scratchpad, gpu_scratchpad);
-
-        pandecode_indent--;
-        pandecode_log("};\n");
-}
-
 static const char *
 shader_type_for_job(unsigned type)
 {
@@ -2137,14 +2114,15 @@ pandecode_vertex_tiler_postfix_pre(
                 .rt_count = 1
         };
 
-        if (is_bifrost)
-                pandecode_scratchpad(p->framebuffer & ~1, job_no, suffix);
-        else if (p->framebuffer & MALI_MFBD)
-                fbd_info = pandecode_mfbd_bfr((u64) ((uintptr_t) p->framebuffer) & FBD_MASK, job_no, false, job_type == JOB_TYPE_COMPUTE);
+        if (is_bifrost) {
+                pandecode_log_cont("\t/* %X %/\n", p->shared_memory & 1);
+                pandecode_compute_fbd(p->shared_memory & ~1, job_no);
+        } else if (p->shared_memory & MALI_MFBD)
+                fbd_info = pandecode_mfbd_bfr((u64) ((uintptr_t) p->shared_memory) & FBD_MASK, job_no, false, job_type == JOB_TYPE_COMPUTE);
         else if (job_type == JOB_TYPE_COMPUTE)
-                pandecode_compute_fbd((u64) (uintptr_t) p->framebuffer, job_no);
+                pandecode_compute_fbd((u64) (uintptr_t) p->shared_memory, job_no);
         else
-                fbd_info = pandecode_sfbd((u64) (uintptr_t) p->framebuffer, job_no, false, gpu_id);
+                fbd_info = pandecode_sfbd((u64) (uintptr_t) p->shared_memory, job_no, false, gpu_id);
 
         int varying_count = 0, attribute_count = 0, uniform_count = 0, uniform_buffer_count = 0;
         int texture_count = 0, sampler_count = 0;
@@ -2309,7 +2287,7 @@ pandecode_vertex_tiler_postfix_pre(
                 /* MRT blend fields are used whenever MFBD is used, with
                  * per-RT descriptors */
 
-                if (job_type == JOB_TYPE_TILER && p->framebuffer & MALI_MFBD) {
+                if (job_type == JOB_TYPE_TILER && p->shared_memory & MALI_MFBD) {
                         void* blend_base = (void *) (s + 1);
 
                         for (unsigned i = 0; i < fbd_info.rt_count; i++) {
@@ -2759,9 +2737,6 @@ pandecode_fragment_job(const struct pandecode_mapped_memory *mem,
 
         bool is_mfbd = s->framebuffer & MALI_MFBD;
 
-        /* Bifrost theoretically may retain support for SFBD on compute jobs,
-         * but for graphics workloads with a FRAGMENT payload, use MFBD */
-
         if (!is_mfbd && is_bifrost)
                 pandecode_msg("XXX: Bifrost fragment must use MFBD\n");