Move compiler.h and imports.h/c from src/mesa/main into src/util

[mesa.git] / src / mesa / drivers / dri / i965 / gen7_urb.c

diff --git a/src/mesa/drivers/dri/i965/gen7_urb.c b/src/mesa/drivers/dri/i965/gen7_urb.c
index d5b03ef44fcb314848f40656bed9038de8e06ed5..abc922a86a356a400c77d2922d706467c517ee20 100644 (file)
--- a/src/mesa/drivers/dri/i965/gen7_urb.c
+++ b/src/mesa/drivers/dri/i965/gen7_urb.c
@@ -62,15 +62,17 @@
 static void
 gen7_allocate_push_constants(struct brw_context *brw)
 {
+   const struct gen_device_info *devinfo = &brw->screen->devinfo;
+
    /* BRW_NEW_GEOMETRY_PROGRAM */
-   bool gs_present = brw->geometry_program;
+   bool gs_present = brw->programs[MESA_SHADER_GEOMETRY];
 
    /* BRW_NEW_TESS_PROGRAMS */
-   bool tess_present = brw->tess_eval_program;
+   bool tess_present = brw->programs[MESA_SHADER_TESS_EVAL];
 
    unsigned avail_size = 16;
    unsigned multiplier =
-      (brw->gen >= 8 || (brw->is_haswell && brw->gt == 3)) ? 2 : 1;
+      (devinfo->gen >= 8 || (devinfo->is_haswell && devinfo->gt == 3)) ? 2 : 1;
 
    int stages = 2 + gs_present + 2 * tess_present;
 
@@ -113,8 +115,36 @@ gen7_emit_push_constant_state(struct brw_context *brw, unsigned vs_size,
                               unsigned hs_size, unsigned ds_size,
                               unsigned gs_size, unsigned fs_size)
 {
+   const struct gen_device_info *devinfo = &brw->screen->devinfo;
    unsigned offset = 0;
 
+   /* From the SKL PRM, Workarounds section (#878):
+    *
+    *    Push constant buffer corruption possible. WA: Insert 2 zero-length
+    *    PushConst_PS before every intended PushConst_PS update, issue a
+    *    NULLPRIM after each of the zero len PC update to make sure CS commits
+    *    them.
+    *
+    * This workaround is attempting to solve a pixel shader push constant
+    * synchronization issue.
+    *
+    * There's an unpublished WA that involves re-emitting
+    * 3DSTATE_PUSH_CONSTANT_ALLOC_PS for every 500-ish 3DSTATE_CONSTANT_PS
+    * packets. Since our counting methods may not be reliable due to
+    * context-switching and pre-emption, we instead choose to approximate this
+    * behavior by re-emitting the packet at the top of the batch.
+    */
+   if (brw->ctx.NewDriverState == BRW_NEW_BATCH) {
+       /* SKL GT2 and GLK 2x6 have reliably demonstrated this issue thus far.
+        * We've also seen some intermittent failures from SKL GT4 and BXT in
+        * the past.
+        */
+      if (!devinfo->is_skylake &&
+          !devinfo->is_broxton &&
+          !devinfo->is_geminilake)
+         return;
+   }
+
    BEGIN_BATCH(10);
    OUT_BATCH(_3DSTATE_PUSH_CONSTANT_ALLOC_VS << 16 | (2 - 2));
    OUT_BATCH(vs_size | offset << GEN7_PUSH_CONSTANT_BUFFER_OFFSET_SHIFT);
@@ -143,7 +173,7 @@ gen7_emit_push_constant_state(struct brw_context *brw, unsigned vs_size,
     *
     * No such restriction exists for Haswell or Baytrail.
     */
-   if (brw->gen < 8 && !brw->is_haswell && !brw->is_baytrail)
+   if (devinfo->gen < 8 && !devinfo->is_haswell && !devinfo->is_baytrail)
       gen7_emit_cs_stall_flush(brw);
 }
 
@@ -151,6 +181,7 @@ const struct brw_tracked_state gen7_push_constant_space = {
    .dirty = {
       .mesa = 0,
       .brw = BRW_NEW_CONTEXT |
+             BRW_NEW_BATCH | /* Push constant workaround */
              BRW_NEW_GEOMETRY_PROGRAM |
              BRW_NEW_TESS_PROGRAMS,
    },
@@ -177,8 +208,6 @@ gen7_upload_urb(struct brw_context *brw, unsigned vs_size,
                 bool gs_present, bool tess_present)
 {
    const struct gen_device_info *devinfo = &brw->screen->devinfo;
-   const int push_size_kB =
-      (brw->gen >= 8 || (brw->is_haswell && brw->gt == 3)) ? 32 : 16;
 
    /* BRW_NEW_{VS,TCS,TES,GS}_PROG_DATA */
    struct brw_vue_prog_data *prog_data[4] = {
@@ -201,9 +230,7 @@ gen7_upload_urb(struct brw_context *brw, unsigned vs_size,
    /* If we're just switching between programs with the same URB requirements,
     * skip the rest of the logic.
     */
-   if (!(brw->ctx.NewDriverState & BRW_NEW_CONTEXT) &&
-       !(brw->ctx.NewDriverState & BRW_NEW_URB_SIZE) &&
-       brw->urb.vsize == entry_size[MESA_SHADER_VERTEX] &&
+   if (brw->urb.vsize == entry_size[MESA_SHADER_VERTEX] &&
        brw->urb.gs_present == gs_present &&
        brw->urb.gsize == entry_size[MESA_SHADER_GEOMETRY] &&
        brw->urb.tess_present == tess_present &&
@@ -220,15 +247,16 @@ gen7_upload_urb(struct brw_context *brw, unsigned vs_size,
 
    unsigned entries[4];
    unsigned start[4];
-   gen_get_urb_config(devinfo, 1024 * push_size_kB, 1024 * brw->urb.size,
-                      tess_present, gs_present, entry_size, entries, start);
+   gen_get_urb_config(devinfo, brw->l3.config,
+                      tess_present, gs_present, entry_size,
+                      entries, start, NULL);
 
-   if (brw->gen == 7 && !brw->is_haswell && !brw->is_baytrail)
+   if (devinfo->gen == 7 && !devinfo->is_haswell && !devinfo->is_baytrail)
       gen7_emit_vs_workaround_flush(brw);
 
    BEGIN_BATCH(8);
    for (int i = MESA_SHADER_VERTEX; i <= MESA_SHADER_GEOMETRY; i++) {
-      assert(brw->gen != 10 || entry_size[i] % 3);
+      assert(devinfo->gen != 10 || entry_size[i] % 3);
       OUT_BATCH((_3DSTATE_URB_VS + i) << 16 | (2 - 2));
       OUT_BATCH(entries[i] |
                 ((entry_size[i] - 1) << GEN7_URB_ENTRY_SIZE_SHIFT) |