i965: Split brw_emit_reloc into brw_batch_reloc and brw_state_reloc.

brw_batch_reloc emits a relocation from the batchbuffer to elsewhere.
brw_state_reloc emits a relocation from the statebuffer to elsewhere.

For now, they do the same thing, but when we actually split the two
buffers, we'll change brw_state_reloc to use the state buffer.

Reviewed-by: Matt Turner <[email protected]>
Reviewed-by: Chris Wilson <[email protected]>

5 files changed, 73 insertions, 48 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
index 1f89b723544..d110482cc8e 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
@@ -176,9 +176,9 @@ brw_emit_surface_state(struct brw_context *brw,
                                  surf_offset);
 
    isl_surf_fill_state(&brw->isl_dev, state, .surf = &mt->surf, .view = &view,
-                       .address = brw_emit_reloc(&brw->batch,
-                                                 *surf_offset + brw->isl_dev.ss.addr_offset,
-                                                 mt->bo, offset, reloc_flags),
+                       .address = brw_state_reloc(&brw->batch,
+                                                  *surf_offset + brw->isl_dev.ss.addr_offset,
+                                                  mt->bo, offset, reloc_flags),
                        .aux_surf = aux_surf, .aux_usage = aux_usage,
                        .aux_address = aux_offset,
                        .mocs = mocs, .clear_color = clear_color,
@@ -194,11 +194,11 @@ brw_emit_surface_state(struct brw_context *brw,
        */
       assert((aux_offset & 0xfff) == 0);
       uint32_t *aux_addr = state + brw->isl_dev.ss.aux_addr_offset;
-      *aux_addr = brw_emit_reloc(&brw->batch,
-                                 *surf_offset +
-                                 brw->isl_dev.ss.aux_addr_offset,
-                                 aux_bo, *aux_addr,
-                                 reloc_flags);
+      *aux_addr = brw_state_reloc(&brw->batch,
+                                  *surf_offset +
+                                  brw->isl_dev.ss.aux_addr_offset,
+                                  aux_bo, *aux_addr,
+                                  reloc_flags);
    }
 }
 
@@ -607,10 +607,10 @@ brw_emit_buffer_surface_state(struct brw_context *brw,
 
    isl_buffer_fill_state(&brw->isl_dev, dw,
                          .address = !bo ? buffer_offset :
-                                    brw_emit_reloc(&brw->batch,
-                                                   *out_offset + brw->isl_dev.ss.addr_offset,
-                                                   bo, buffer_offset,
-                                                   reloc_flags),
+                                    brw_state_reloc(&brw->batch,
+                                                    *out_offset + brw->isl_dev.ss.addr_offset,
+                                                    bo, buffer_offset,
+                                                    reloc_flags),
                          .size = buffer_size,
                          .format = surface_format,
                          .stride = pitch,
@@ -777,8 +777,8 @@ brw_update_sol_surface(struct brw_context *brw,
       BRW_SURFACE_MIPMAPLAYOUT_BELOW << BRW_SURFACE_MIPLAYOUT_SHIFT |
       surface_format << BRW_SURFACE_FORMAT_SHIFT |
       BRW_SURFACE_RC_READ_WRITE;
-   surf[1] = brw_emit_reloc(&brw->batch,
-                            *out_offset + 4, bo, offset_bytes, RELOC_WRITE);
+   surf[1] = brw_state_reloc(&brw->batch,
+                             *out_offset + 4, bo, offset_bytes, RELOC_WRITE);
    surf[2] = (width << BRW_SURFACE_WIDTH_SHIFT |
 	      height << BRW_SURFACE_HEIGHT_SHIFT);
    surf[3] = (depth << BRW_SURFACE_DEPTH_SHIFT |
@@ -870,9 +870,9 @@ emit_null_surface_state(struct brw_context *brw,
 
    surf[0] = (BRW_SURFACE_2D << BRW_SURFACE_TYPE_SHIFT |
 	      ISL_FORMAT_B8G8R8A8_UNORM << BRW_SURFACE_FORMAT_SHIFT);
-   surf[1] = brw_emit_reloc(&brw->batch, *out_offset + 4,
-                            brw->wm.multisampled_null_render_target_bo,
-                            0, RELOC_WRITE);
+   surf[1] = brw_state_reloc(&brw->batch, *out_offset + 4,
+                             brw->wm.multisampled_null_render_target_bo,
+                             0, RELOC_WRITE);
 
    surf[2] = ((width - 1) << BRW_SURFACE_WIDTH_SHIFT |
               (height - 1) << BRW_SURFACE_HEIGHT_SHIFT);
@@ -940,12 +940,12 @@ gen4_update_renderbuffer_surface(struct brw_context *brw,
 
    /* reloc */
    assert(mt->offset % mt->cpp == 0);
-   surf[1] = brw_emit_reloc(&brw->batch, offset + 4, mt->bo,
-                            mt->offset +
-                            intel_renderbuffer_get_tile_offsets(irb,
-                                                                &tile_x,
-                                                                &tile_y),
-                            RELOC_WRITE);
+   surf[1] = brw_state_reloc(&brw->batch, offset + 4, mt->bo,
+                             mt->offset +
+                             intel_renderbuffer_get_tile_offsets(irb,
+                                                                 &tile_x,
+                                                                 &tile_y),
+                             RELOC_WRITE);
 
    surf[2] = ((rb->Width - 1) << BRW_SURFACE_WIDTH_SHIFT |
 	      (rb->Height - 1) << BRW_SURFACE_HEIGHT_SHIFT);
diff --git a/src/mesa/drivers/dri/i965/genX_blorp_exec.c b/src/mesa/drivers/dri/i965/genX_blorp_exec.c
index 2b531cbcd12..d1ff8ab4b73 100644
--- a/src/mesa/drivers/dri/i965/genX_blorp_exec.c
+++ b/src/mesa/drivers/dri/i965/genX_blorp_exec.c
@@ -59,9 +59,9 @@ blorp_emit_reloc(struct blorp_batch *batch,
    struct brw_context *brw = batch->driver_batch;
 
    uint32_t offset = (char *)location - (char *)brw->batch.map;
-   return brw_emit_reloc(&brw->batch, offset,
-                         address.buffer, address.offset + delta,
-                         address.reloc_flags);
+   return brw_batch_reloc(&brw->batch, offset,
+                          address.buffer, address.offset + delta,
+                          address.reloc_flags);
 }
 
 static void
@@ -73,8 +73,8 @@ blorp_surface_reloc(struct blorp_batch *batch, uint32_t ss_offset,
    struct brw_bo *bo = address.buffer;
 
    uint64_t reloc_val =
-      brw_emit_reloc(&brw->batch, ss_offset, bo, address.offset + delta,
-                     address.reloc_flags);
+      brw_state_reloc(&brw->batch, ss_offset, bo, address.offset + delta,
+                      address.reloc_flags);
 
    void *reloc_ptr = (void *)brw->batch.map + ss_offset;
 #if GEN_GEN >= 8
diff --git a/src/mesa/drivers/dri/i965/genX_state_upload.c b/src/mesa/drivers/dri/i965/genX_state_upload.c
index 7083e7b2d6b..ea488d72356 100644
--- a/src/mesa/drivers/dri/i965/genX_state_upload.c
+++ b/src/mesa/drivers/dri/i965/genX_state_upload.c
@@ -87,9 +87,10 @@ __gen_combine_address(struct brw_context *brw, void *location,
    } else {
       uint32_t offset = (char *) location - (char *) brw->batch.map;
 
-      return brw_emit_reloc(&brw->batch, offset, address.bo,
-                            address.offset + delta,
-                            address.reloc_flags);
+      /* TODO: Use brw_state_reloc for some things on Gen4-5 */
+      return brw_batch_reloc(&brw->batch, offset, address.bo,
+                             address.offset + delta,
+                             address.reloc_flags);
    }
 }
 
diff --git a/src/mesa/drivers/dri/i965/intel_batchbuffer.c b/src/mesa/drivers/dri/i965/intel_batchbuffer.c
index ae27ec5445b..0e8ca184392 100644
--- a/src/mesa/drivers/dri/i965/intel_batchbuffer.c
+++ b/src/mesa/drivers/dri/i965/intel_batchbuffer.c
@@ -794,13 +794,12 @@ brw_batch_references(struct intel_batchbuffer *batch, struct brw_bo *bo)
 
 /*  This is the only way buffers get added to the validate list.
  */
-uint64_t
-brw_emit_reloc(struct intel_batchbuffer *batch, uint32_t batch_offset,
-               struct brw_bo *target, uint32_t target_offset,
-               unsigned int reloc_flags)
+static uint64_t
+emit_reloc(struct intel_batchbuffer *batch,
+           struct brw_reloc_list *rlist, uint32_t offset,
+           struct brw_bo *target, uint32_t target_offset,
+           unsigned int reloc_flags)
 {
-   struct brw_reloc_list *rlist = &batch->batch_relocs;
-
    assert(target != NULL);
 
    if (rlist->reloc_count == rlist->reloc_array_size) {
@@ -810,9 +809,6 @@ brw_emit_reloc(struct intel_batchbuffer *batch, uint32_t batch_offset,
                               sizeof(struct drm_i915_gem_relocation_entry));
    }
 
-   /* Check args */
-   assert(batch_offset <= batch->bo->size - sizeof(uint32_t));
-
    unsigned int index = add_exec_bo(batch, target);
    struct drm_i915_gem_exec_object2 *entry = &batch->validation_list[index];
 
@@ -821,7 +817,7 @@ brw_emit_reloc(struct intel_batchbuffer *batch, uint32_t batch_offset,
 
    rlist->relocs[rlist->reloc_count++] =
       (struct drm_i915_gem_relocation_entry) {
-         .offset = batch_offset,
+         .offset = offset,
          .delta = target_offset,
          .target_handle = batch->use_batch_first ? index : target->gem_handle,
          .presumed_offset = entry->offset,
@@ -834,6 +830,29 @@ brw_emit_reloc(struct intel_batchbuffer *batch, uint32_t batch_offset,
    return entry->offset + target_offset;
 }
 
+uint64_t
+brw_batch_reloc(struct intel_batchbuffer *batch, uint32_t batch_offset,
+                struct brw_bo *target, uint32_t target_offset,
+                unsigned int reloc_flags)
+{
+   assert(batch_offset <= batch->bo->size - sizeof(uint32_t));
+
+   return emit_reloc(batch, &batch->batch_relocs, batch_offset,
+                     target, target_offset, reloc_flags);
+}
+
+uint64_t
+brw_state_reloc(struct intel_batchbuffer *batch, uint32_t state_offset,
+                struct brw_bo *target, uint32_t target_offset,
+                unsigned int reloc_flags)
+{
+   assert(state_offset <= batch->bo->size - sizeof(uint32_t));
+
+   return emit_reloc(batch, &batch->batch_relocs, state_offset,
+                     target, target_offset, reloc_flags);
+}
+
+
 uint32_t
 brw_state_batch_size(struct brw_context *brw, uint32_t offset)
 {
diff --git a/src/mesa/drivers/dri/i965/intel_batchbuffer.h b/src/mesa/drivers/dri/i965/intel_batchbuffer.h
index 99d2747f282..6eb1014c1cb 100644
--- a/src/mesa/drivers/dri/i965/intel_batchbuffer.h
+++ b/src/mesa/drivers/dri/i965/intel_batchbuffer.h
@@ -71,11 +71,16 @@ bool brw_batch_references(struct intel_batchbuffer *batch, struct brw_bo *bo);
 
 #define RELOC_WRITE EXEC_OBJECT_WRITE
 #define RELOC_NEEDS_GGTT EXEC_OBJECT_NEEDS_GTT
-uint64_t brw_emit_reloc(struct intel_batchbuffer *batch,
-                        uint32_t batch_offset,
-                        struct brw_bo *target,
-                        uint32_t target_offset,
-                        unsigned flags);
+uint64_t brw_batch_reloc(struct intel_batchbuffer *batch,
+                         uint32_t batch_offset,
+                         struct brw_bo *target,
+                         uint32_t target_offset,
+                         unsigned flags);
+uint64_t brw_state_reloc(struct intel_batchbuffer *batch,
+                         uint32_t batch_offset,
+                         struct brw_bo *target,
+                         uint32_t target_offset,
+                         unsigned flags);
 
 #define USED_BATCH(batch) ((uintptr_t)((batch).map_next - (batch).map))
 
@@ -164,7 +169,7 @@ intel_batchbuffer_advance(struct brw_context *brw)
 #define OUT_RELOC(buf, flags, delta) do {          \
    uint32_t __offset = (__map - brw->batch.map) * 4;                    \
    uint32_t reloc =                                                     \
-      brw_emit_reloc(&brw->batch, __offset, (buf), (delta), (flags));   \
+      brw_batch_reloc(&brw->batch, __offset, (buf), (delta), (flags));  \
    OUT_BATCH(reloc);                                                    \
 } while (0)
 
@@ -172,7 +177,7 @@ intel_batchbuffer_advance(struct brw_context *brw)
 #define OUT_RELOC64(buf, flags, delta) do {        \
    uint32_t __offset = (__map - brw->batch.map) * 4;                    \
    uint64_t reloc64 =                                                   \
-      brw_emit_reloc(&brw->batch, __offset, (buf), (delta), (flags));   \
+      brw_batch_reloc(&brw->batch, __offset, (buf), (delta), (flags));  \
    OUT_BATCH(reloc64);                                                  \
    OUT_BATCH(reloc64 >> 32);                                            \
 } while (0)