v3d: Emit the VCM_CACHE_SIZE packet.

This is needed to ensure that we don't get blocked waiting for VPM space
with bin/render overlapping.

Cc: "18.2" <mesa-stable@lists.freedesktop.org>

src/broadcom/cle/v3d_packet_v33.xml

@@ -528,6 +528,16 @@
     <field name="number of attribute arrays" size="5" start="0" type="uint"/>
   </packet>
 
+  <packet code="71" name="VCM Cache Size" min_ver="41">
+    <field name="Number of 16-vertex batches for rendering" size="4" start="4" type="uint"/>
+    <field name="Number of 16-vertex batches for binning" size="4" start="0" type="uint"/>
+  </packet>
+
+  <packet code="73" name="VCM Cache Size" max_ver="33">
+    <field name="Number of 16-vertex batches for rendering" size="4" start="4" type="uint"/>
+    <field name="Number of 16-vertex batches for binning" size="4" start="0" type="uint"/>
+  </packet>
+
   <packet code="73" name="Transform Feedback Buffer" min_ver="41">
     <field name="Buffer Address" size="32" start="32" type="address"/>
     <field name="Buffer Size in 32-bit words" size="30" start="2" type="uint"/>

src/broadcom/common/v3d_device_info.h

@@ -27,13 +27,14 @@
 #include <stdint.h>
 
 /**
- * Struct for tracking features of the V3D chip. This is where we'll store
- * boolean flags for features in a specific version, but for now it's just the
- * version
+ * Struct for tracking features of the V3D chip across driver and compiler.
  */
 struct v3d_device_info {
         /** Simple V3D version: major * 10 + minor */
         uint8_t ver;
+
+        /** Size of the VPM, in bytes. */
+        int vpm_size;
 };
 
 #endif

src/broadcom/compiler/v3d_compiler.h

@@ -648,6 +648,9 @@ struct v3d_vs_prog_data {
 
         /* Total number of components written, for the shader state record. */
         uint32_t vpm_output_size;
+
+        /* Value to be programmed in VCM_CACHE_SIZE. */
+        uint8_t vcm_cache_size;
 };
 
 struct v3d_fs_prog_data {

src/broadcom/compiler/vir.c

@@ -756,10 +756,28 @@ uint64_t *v3d_compile_vs(const struct v3d_compiler *compiler,
         if (prog_data->uses_iid)
                 prog_data->vpm_input_size++;
 
-        /* Input/output segment size are in 8x32-bit multiples. */
+        /* Input/output segment size are in sectors (8 rows of 32 bits per
+         * channel).
+         */
         prog_data->vpm_input_size = align(prog_data->vpm_input_size, 8) / 8;
         prog_data->vpm_output_size = align(c->num_vpm_writes, 8) / 8;
 
+        /* Compute VCM cache size.  We set up our program to take up less than
+         * half of the VPM, so that any set of bin and render programs won't
+         * run out of space.  We need space for at least one input segment,
+         * and then allocate the rest to output segments (one for the current
+         * program, the rest to VCM).  The valid range of the VCM cache size
+         * field is 1-4 16-vertex batches, but GFXH-1744 limits us to 2-4
+         * batches.
+         */
+        assert(c->devinfo->vpm_size);
+        int sector_size = 16 * sizeof(uint32_t) * 8;
+        int vpm_size_in_sectors = c->devinfo->vpm_size / sector_size;
+        int half_vpm = vpm_size_in_sectors / 2;
+        int vpm_output_batches = half_vpm - prog_data->vpm_input_size;
+        assert(vpm_output_batches >= 2);
+        prog_data->vcm_cache_size = CLAMP(vpm_output_batches - 1, 2, 4);
+
         return v3d_return_qpu_insts(c, final_assembly_size);
 }
 

src/gallium/drivers/v3d/v3d_screen.c

@@ -585,6 +585,8 @@ v3d_get_device_info(struct v3d_screen *screen)
         uint32_t minor = (ident1.value >> 0) & 0xf;
         screen->devinfo.ver = major * 10 + minor;
 
+        screen->devinfo.vpm_size = (ident1.value >> 28 & 0xf) * 1024;
+
         switch (screen->devinfo.ver) {
         case 33:
         case 41:

src/gallium/drivers/v3d/v3dx_draw.c

@@ -306,6 +306,13 @@ v3d_emit_gl_shader_state(struct v3d_context *v3d,
                 }
         }
 
+        cl_emit(&job->bcl, VCM_CACHE_SIZE, vcm) {
+                vcm.number_of_16_vertex_batches_for_binning =
+                        v3d->prog.cs->prog_data.vs->vcm_cache_size;
+                vcm.number_of_16_vertex_batches_for_rendering =
+                        v3d->prog.vs->prog_data.vs->vcm_cache_size;
+        }
+
         cl_emit(&job->bcl, GL_SHADER_STATE, state) {
                 state.address = cl_address(job->indirect.bo, shader_rec_offset);
                 state.number_of_attribute_arrays = num_elements_to_emit;