i965: Refactor SIMD16-to-2xSIMD8 checks.
[mesa.git] / src / mesa / drivers / dri / i965 / brw_misc_state.c
index a6b108b69832693679c46878fbbf02980972a6dc..bc810769793f8c816aaea41eaa6fb883abc6bb2b 100644 (file)
@@ -34,7 +34,6 @@
 #include "intel_batchbuffer.h"
 #include "intel_fbo.h"
 #include "intel_mipmap_tree.h"
-#include "intel_regions.h"
 
 #include "brw_context.h"
 #include "brw_state.h"
@@ -48,10 +47,6 @@ static void upload_drawing_rect(struct brw_context *brw)
 {
    struct gl_context *ctx = &brw->ctx;
 
-   /* 3DSTATE_DRAWING_RECTANGLE is non-pipelined. */
-   if (brw->gen == 6)
-      intel_emit_post_sync_nonzero_flush(brw);
-
    BEGIN_BATCH(4);
    OUT_BATCH(_3DSTATE_DRAWING_RECTANGLE << 16 | (4 - 2));
    OUT_BATCH(0); /* xmin, ymin */
@@ -65,7 +60,6 @@ const struct brw_tracked_state brw_drawing_rect = {
    .dirty = {
       .mesa = _NEW_BUFFERS,
       .brw = BRW_NEW_CONTEXT,
-      .cache = 0
    },
    .emit = upload_drawing_rect
 };
@@ -117,16 +111,11 @@ static void upload_psp_urb_cbs(struct brw_context *brw )
 const struct brw_tracked_state brw_psp_urb_cbs = {
    .dirty = {
       .mesa = 0,
-      .brw = (BRW_NEW_URB_FENCE |
-             BRW_NEW_BATCH |
-             BRW_NEW_STATE_BASE_ADDRESS),
-      .cache = (CACHE_NEW_VS_UNIT |
-               CACHE_NEW_FF_GS_UNIT |
-               CACHE_NEW_FF_GS_PROG |
-               CACHE_NEW_CLIP_UNIT |
-               CACHE_NEW_SF_UNIT |
-               CACHE_NEW_WM_UNIT |
-               CACHE_NEW_CC_UNIT)
+      .brw = BRW_NEW_BATCH |
+             BRW_NEW_FF_GS_PROG_DATA |
+             BRW_NEW_GEN4_UNIT_STATE |
+             BRW_NEW_STATE_BASE_ADDRESS |
+             BRW_NEW_URB_FENCE,
    },
    .emit = upload_psp_urb_cbs,
 };
@@ -179,13 +168,13 @@ brw_get_depthstencil_tile_masks(struct intel_mipmap_tree *depth_mt,
    uint32_t tile_mask_x = 0, tile_mask_y = 0;
 
    if (depth_mt) {
-      intel_region_get_tile_masks(depth_mt->region,
-                                  &tile_mask_x, &tile_mask_y, false);
+      intel_miptree_get_tile_masks(depth_mt, &tile_mask_x, &tile_mask_y, false);
 
-      if (intel_miptree_slice_has_hiz(depth_mt, depth_level, depth_layer)) {
+      if (intel_miptree_level_has_hiz(depth_mt, depth_level)) {
          uint32_t hiz_tile_mask_x, hiz_tile_mask_y;
-         intel_region_get_tile_masks(depth_mt->hiz_mt->region,
-                                     &hiz_tile_mask_x, &hiz_tile_mask_y, false);
+         intel_miptree_get_tile_masks(depth_mt->hiz_buf->mt,
+                                      &hiz_tile_mask_x, &hiz_tile_mask_y,
+                                      false);
 
          /* Each HiZ row represents 2 rows of pixels */
          hiz_tile_mask_y = hiz_tile_mask_y << 1 | 1;
@@ -205,9 +194,9 @@ brw_get_depthstencil_tile_masks(struct intel_mipmap_tree *depth_mt,
          tile_mask_y |= 63;
       } else {
          uint32_t stencil_tile_mask_x, stencil_tile_mask_y;
-         intel_region_get_tile_masks(stencil_mt->region,
-                                     &stencil_tile_mask_x,
-                                     &stencil_tile_mask_y, false);
+         intel_miptree_get_tile_masks(stencil_mt,
+                                      &stencil_tile_mask_x,
+                                      &stencil_tile_mask_y, false);
 
          tile_mask_x |= stencil_tile_mask_x;
          tile_mask_y |= stencil_tile_mask_y;
@@ -262,10 +251,10 @@ brw_workaround_depthstencil_alignment(struct brw_context *brw,
    if (stencil_irb)
       brw->depthstencil.stencil_mt = get_stencil_miptree(stencil_irb);
 
-   /* Gen7+ doesn't require the workarounds, since we always program the
+   /* Gen6+ doesn't require the workarounds, since we always program the
     * surface state at the start of the whole surface.
     */
-   if (brw->gen >= 7)
+   if (brw->gen >= 6)
       return;
 
    /* Check if depth buffer is in depth/stencil format.  If so, then it's only
@@ -446,17 +435,16 @@ brw_workaround_depthstencil_alignment(struct brw_context *brw,
       depth_mt = depth_irb->mt;
       brw->depthstencil.depth_mt = depth_mt;
       brw->depthstencil.depth_offset =
-         intel_region_get_aligned_offset(depth_mt->region,
-                                         depth_irb->draw_x & ~tile_mask_x,
-                                         depth_irb->draw_y & ~tile_mask_y,
-                                         false);
+         intel_miptree_get_aligned_offset(depth_mt,
+                                          depth_irb->draw_x & ~tile_mask_x,
+                                          depth_irb->draw_y & ~tile_mask_y,
+                                          false);
       if (intel_renderbuffer_has_hiz(depth_irb)) {
          brw->depthstencil.hiz_offset =
-            intel_region_get_aligned_offset(depth_mt->region,
-                                            depth_irb->draw_x & ~tile_mask_x,
-                                            (depth_irb->draw_y & ~tile_mask_y) /
-                                            2,
-                                            false);
+            intel_miptree_get_aligned_offset(depth_mt,
+                                             depth_irb->draw_x & ~tile_mask_x,
+                                             (depth_irb->draw_y & ~tile_mask_y) / 2,
+                                             false);
       }
    }
    if (stencil_irb) {
@@ -469,7 +457,7 @@ brw_workaround_depthstencil_alignment(struct brw_context *brw,
           * that the region is untiled even though it's W tiled.
           */
          brw->depthstencil.stencil_offset =
-            (stencil_draw_y & ~tile_mask_y) * stencil_mt->region->pitch +
+            (stencil_draw_y & ~tile_mask_y) * stencil_mt->pitch +
             (stencil_draw_x & ~tile_mask_x) * 64;
       }
    }
@@ -525,8 +513,8 @@ brw_emit_depthbuffer(struct brw_context *brw)
       /* Prior to Gen7, if using separate stencil, hiz must be enabled. */
       assert(brw->gen >= 7 || !separate_stencil || hiz);
 
-      assert(brw->gen < 6 || depth_mt->region->tiling == I915_TILING_Y);
-      assert(!hiz || depth_mt->region->tiling == I915_TILING_Y);
+      assert(brw->gen < 6 || depth_mt->tiling == I915_TILING_Y);
+      assert(!hiz || depth_mt->tiling == I915_TILING_Y);
 
       depthbuffer_format = brw_depthbuffer_format(brw);
       depth_surface_type = BRW_SURFACE_2D;
@@ -553,9 +541,9 @@ brw_emit_depthbuffer(struct brw_context *brw)
    }
 
    if (depth_mt)
-      brw_render_cache_set_check_flush(brw, depth_mt->region->bo);
+      brw_render_cache_set_check_flush(brw, depth_mt->bo);
    if (stencil_mt)
-      brw_render_cache_set_check_flush(brw, stencil_mt->region->bo);
+      brw_render_cache_set_check_flush(brw, stencil_mt->bo);
 
    brw->vtbl.emit_depth_stencil_hiz(brw, depth_mt, depth_offset,
                                     depthbuffer_format, depth_surface_type,
@@ -589,7 +577,6 @@ brw_emit_depth_stencil_hiz(struct brw_context *brw,
     * non-pipelined state that will need the PIPE_CONTROL workaround.
     */
    if (brw->gen == 6) {
-      intel_emit_post_sync_nonzero_flush(brw);
       intel_emit_depth_stall_flushes(brw);
    }
 
@@ -603,17 +590,17 @@ brw_emit_depth_stencil_hiz(struct brw_context *brw,
 
    BEGIN_BATCH(len);
    OUT_BATCH(_3DSTATE_DEPTH_BUFFER << 16 | (len - 2));
-   OUT_BATCH((depth_mt ? depth_mt->region->pitch - 1 : 0) |
+   OUT_BATCH((depth_mt ? depth_mt->pitch - 1 : 0) |
              (depthbuffer_format << 18) |
              ((enable_hiz_ss ? 1 : 0) << 21) | /* separate stencil enable */
              ((enable_hiz_ss ? 1 : 0) << 22) | /* hiz enable */
              (BRW_TILEWALK_YMAJOR << 26) |
-             ((depth_mt ? depth_mt->region->tiling != I915_TILING_NONE : 1)
+             ((depth_mt ? depth_mt->tiling != I915_TILING_NONE : 1)
               << 27) |
              (depth_surface_type << 29));
 
    if (depth_mt) {
-      OUT_RELOC(depth_mt->region->bo,
+      OUT_RELOC(depth_mt->bo,
                I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
                depth_offset);
    } else {
@@ -645,11 +632,11 @@ brw_emit_depth_stencil_hiz(struct brw_context *brw,
 
       /* Emit hiz buffer. */
       if (hiz) {
-         struct intel_mipmap_tree *hiz_mt = depth_mt->hiz_mt;
+         struct intel_mipmap_tree *hiz_mt = depth_mt->hiz_buf->mt;
         BEGIN_BATCH(3);
         OUT_BATCH((_3DSTATE_HIER_DEPTH_BUFFER << 16) | (3 - 2));
-        OUT_BATCH(hiz_mt->region->pitch - 1);
-        OUT_RELOC(hiz_mt->region->bo,
+        OUT_BATCH(hiz_mt->pitch - 1);
+        OUT_RELOC(hiz_mt->bo,
                   I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
                   brw->depthstencil.hiz_offset);
         ADVANCE_BATCH();
@@ -663,8 +650,6 @@ brw_emit_depth_stencil_hiz(struct brw_context *brw,
 
       /* Emit stencil buffer. */
       if (separate_stencil) {
-        struct intel_region *region = stencil_mt->region;
-
         BEGIN_BATCH(3);
         OUT_BATCH((_3DSTATE_STENCIL_BUFFER << 16) | (3 - 2));
          /* The stencil buffer has quirky pitch requirements.  From Vol 2a,
@@ -672,8 +657,8 @@ brw_emit_depth_stencil_hiz(struct brw_context *brw,
           *    The pitch must be set to 2x the value computed based on width, as
           *    the stencil buffer is stored with two rows interleaved.
           */
-        OUT_BATCH(2 * region->pitch - 1);
-        OUT_RELOC(region->bo,
+        OUT_BATCH(2 * stencil_mt->pitch - 1);
+        OUT_RELOC(stencil_mt->bo,
                   I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
                   brw->depthstencil.stencil_offset);
         ADVANCE_BATCH();
@@ -695,9 +680,6 @@ brw_emit_depth_stencil_hiz(struct brw_context *brw,
     *     when HiZ is enabled and the DEPTH_BUFFER_STATE changes.
     */
    if (brw->gen >= 6 || hiz) {
-      if (brw->gen == 6)
-        intel_emit_post_sync_nonzero_flush(brw);
-
       BEGIN_BATCH(2);
       OUT_BATCH(_3DSTATE_CLEAR_PARAMS << 16 |
                GEN5_DEPTH_CLEAR_VALID |
@@ -711,7 +693,6 @@ const struct brw_tracked_state brw_depthbuffer = {
    .dirty = {
       .mesa = _NEW_BUFFERS,
       .brw = BRW_NEW_BATCH,
-      .cache = 0,
    },
    .emit = brw_emit_depthbuffer,
 };
@@ -731,9 +712,6 @@ static void upload_polygon_stipple(struct brw_context *brw)
    if (!ctx->Polygon.StippleFlag)
       return;
 
-   if (brw->gen == 6)
-      intel_emit_post_sync_nonzero_flush(brw);
-
    BEGIN_BATCH(33);
    OUT_BATCH(_3DSTATE_POLY_STIPPLE_PATTERN << 16 | (33 - 2));
 
@@ -757,10 +735,9 @@ static void upload_polygon_stipple(struct brw_context *brw)
 
 const struct brw_tracked_state brw_polygon_stipple = {
    .dirty = {
-      .mesa = (_NEW_POLYGONSTIPPLE |
-              _NEW_POLYGON),
+      .mesa = _NEW_POLYGON |
+              _NEW_POLYGONSTIPPLE,
       .brw = BRW_NEW_CONTEXT,
-      .cache = 0
    },
    .emit = upload_polygon_stipple
 };
@@ -778,9 +755,6 @@ static void upload_polygon_stipple_offset(struct brw_context *brw)
    if (!ctx->Polygon.StippleFlag)
       return;
 
-   if (brw->gen == 6)
-      intel_emit_post_sync_nonzero_flush(brw);
-
    BEGIN_BATCH(2);
    OUT_BATCH(_3DSTATE_POLY_STIPPLE_OFFSET << 16 | (2-2));
 
@@ -801,10 +775,9 @@ static void upload_polygon_stipple_offset(struct brw_context *brw)
 
 const struct brw_tracked_state brw_polygon_stipple_offset = {
    .dirty = {
-      .mesa = (_NEW_BUFFERS |
-              _NEW_POLYGON),
+      .mesa = _NEW_BUFFERS |
+              _NEW_POLYGON,
       .brw = BRW_NEW_CONTEXT,
-      .cache = 0
    },
    .emit = upload_polygon_stipple_offset
 };
@@ -823,9 +796,6 @@ static void upload_aa_line_parameters(struct brw_context *brw)
    if (brw->gen == 4 && !brw->is_g4x)
       return;
 
-   if (brw->gen == 6)
-      intel_emit_post_sync_nonzero_flush(brw);
-
    BEGIN_BATCH(3);
    OUT_BATCH(_3DSTATE_AA_LINE_PARAMETERS << 16 | (3 - 2));
    /* use legacy aa line coverage computation */
@@ -838,7 +808,6 @@ const struct brw_tracked_state brw_aa_line_parameters = {
    .dirty = {
       .mesa = _NEW_LINE,
       .brw = BRW_NEW_CONTEXT,
-      .cache = 0
    },
    .emit = upload_aa_line_parameters
 };
@@ -856,9 +825,6 @@ static void upload_line_stipple(struct brw_context *brw)
    if (!ctx->Line.StippleFlag)
       return;
 
-   if (brw->gen == 6)
-      intel_emit_post_sync_nonzero_flush(brw);
-
    BEGIN_BATCH(3);
    OUT_BATCH(_3DSTATE_LINE_STIPPLE_PATTERN << 16 | (3 - 2));
    OUT_BATCH(ctx->Line.StipplePattern);
@@ -883,7 +849,6 @@ const struct brw_tracked_state brw_line_stipple = {
    .dirty = {
       .mesa = _NEW_LINE,
       .brw = BRW_NEW_CONTEXT,
-      .cache = 0
    },
    .emit = upload_line_stipple
 };
@@ -898,15 +863,11 @@ brw_upload_invariant_state(struct brw_context *brw)
 {
    const bool is_965 = brw->gen == 4 && !brw->is_g4x;
 
-   /* 3DSTATE_SIP, 3DSTATE_MULTISAMPLE, etc. are nonpipelined. */
-   if (brw->gen == 6)
-      intel_emit_post_sync_nonzero_flush(brw);
-
    /* Select the 3D pipeline (as opposed to media) */
    const uint32_t _3DSTATE_PIPELINE_SELECT =
       is_965 ? CMD_PIPELINE_SELECT_965 : CMD_PIPELINE_SELECT_GM45;
    BEGIN_BATCH(1);
-   OUT_BATCH(_3DSTATE_PIPELINE_SELECT << 16 | 0);
+   OUT_BATCH(_3DSTATE_PIPELINE_SELECT << 16 | (brw->gen >= 9 ? (3 << 8) : 0));
    ADVANCE_BATCH();
 
    if (brw->gen < 6) {
@@ -933,8 +894,7 @@ brw_upload_invariant_state(struct brw_context *brw)
    const uint32_t _3DSTATE_VF_STATISTICS =
       is_965 ? GEN4_3DSTATE_VF_STATISTICS : GM45_3DSTATE_VF_STATISTICS;
    BEGIN_BATCH(1);
-   OUT_BATCH(_3DSTATE_VF_STATISTICS << 16 |
-            (unlikely(INTEL_DEBUG & DEBUG_STATS) ? 1 : 0));
+   OUT_BATCH(_3DSTATE_VF_STATISTICS << 16 | 1);
    ADVANCE_BATCH();
 }
 
@@ -942,7 +902,6 @@ const struct brw_tracked_state brw_invariant_state = {
    .dirty = {
       .mesa = 0,
       .brw = BRW_NEW_CONTEXT,
-      .cache = 0
    },
    .emit = brw_upload_invariant_state
 };
@@ -971,9 +930,6 @@ static void upload_state_base_address( struct brw_context *brw )
    if (brw->gen >= 6) {
       uint8_t mocs = brw->gen == 7 ? GEN7_MOCS_L3 : 0;
 
-      if (brw->gen == 6)
-        intel_emit_post_sync_nonzero_flush(brw);
-
        BEGIN_BATCH(10);
        OUT_BATCH(CMD_STATE_BASE_ADDRESS << 16 | (10 - 2));
        OUT_BATCH(mocs << 8 | /* General State Memory Object Control State */
@@ -1064,9 +1020,8 @@ static void upload_state_base_address( struct brw_context *brw )
 const struct brw_tracked_state brw_state_base_address = {
    .dirty = {
       .mesa = 0,
-      .brw = (BRW_NEW_BATCH |
-             BRW_NEW_PROGRAM_CACHE),
-      .cache = 0,
+      .brw = BRW_NEW_BATCH |
+             BRW_NEW_PROGRAM_CACHE,
    },
    .emit = upload_state_base_address
 };