i965/fs: Add support for translating ir_triop_fma into MAD.

[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 bdcf1648939155db6c3baff0301da0434d158c12..350f644dfcdcfa2dfc703c070703fc458fb83551 100644 (file)
--- a/src/mesa/drivers/dri/i965/gen7_urb.c
+++ b/src/mesa/drivers/dri/i965/gen7_urb.c
@@ -39,46 +39,41 @@
  * +-------------------------------------------------------------+
  *
  * Notably, push constants must be stored at the beginning of the URB
- * space, while entries can be stored anywhere.  Ivybridge has a maximum
- * constant buffer size of 16kB.
+ * space, while entries can be stored anywhere.  Ivybridge and Haswell
+ * GT1/GT2 have a maximum constant buffer size of 16kB, while Haswell GT3
+ * doubles this (32kB).
  *
  * Currently we split the constant buffer space evenly between VS and FS.
  * This is probably not ideal, but simple.
  *
- * Ivybridge GT1 has 128kB of URB space.
- * Ivybridge GT2 has 256kB of URB space.
+ * Ivybridge GT1 and Haswell GT1 have 128kB of URB space.
+ * Ivybridge GT2 and Haswell GT2 have 256kB of URB space.
+ * Haswell GT3 has 512kB of URB space.
  *
- * See "Volume 2a: 3D Pipeline," section 1.8.
+ * See "Volume 2a: 3D Pipeline," section 1.8, "Volume 1b: Configurations",
+ * and the documentation for 3DSTATE_PUSH_CONSTANT_ALLOC_xS.
  */
 void
 gen7_allocate_push_constants(struct brw_context *brw)
 {
-   struct intel_context *intel = &brw->intel;
-   BEGIN_BATCH(2);
+   unsigned size = 8;
+   if (brw->is_haswell && brw->gt == 3)
+      size = 16;
+
+   BEGIN_BATCH(4);
    OUT_BATCH(_3DSTATE_PUSH_CONSTANT_ALLOC_VS << 16 | (2 - 2));
-   OUT_BATCH(8);
-   ADVANCE_BATCH();
+   OUT_BATCH(size);
 
-   BEGIN_BATCH(2);
    OUT_BATCH(_3DSTATE_PUSH_CONSTANT_ALLOC_PS << 16 | (2 - 2));
-   OUT_BATCH(8 | 8 << GEN7_PUSH_CONSTANT_BUFFER_OFFSET_SHIFT);
+   OUT_BATCH(size | size << GEN7_PUSH_CONSTANT_BUFFER_OFFSET_SHIFT);
    ADVANCE_BATCH();
 }
 
-const struct brw_tracked_state gen7_push_constant_alloc = {
-   .dirty = {
-      .mesa = 0,
-      .brw = BRW_NEW_CONTEXT,
-      .cache = 0,
-   },
-   .emit = gen7_allocate_push_constants,
-};
-
 static void
 gen7_upload_urb(struct brw_context *brw)
 {
-   struct intel_context *intel = &brw->intel;
-   const int push_size_kB = 16;
+   const int push_size_kB = brw->is_haswell && brw->gt == 3 ? 32 : 16;
+
    /* Total space for entries is URB size - 16kB for push constants */
    int handle_region_size = (brw->urb.size - push_size_kB) * 1024; /* bytes */
 
@@ -100,7 +95,7 @@ gen7_upload_urb(struct brw_context *brw)
    /* GS requirement */
    assert(!brw->gs.prog_active);
 
-   gen7_emit_vs_workaround_flush(intel);
+   gen7_emit_vs_workaround_flush(brw);
    gen7_emit_urb_state(brw, brw->urb.nr_vs_entries, vs_size, brw->urb.vs_start);
 }
 
@@ -108,29 +103,21 @@ void
 gen7_emit_urb_state(struct brw_context *brw, GLuint nr_vs_entries,
                     GLuint vs_size, GLuint vs_start)
 {
-   struct intel_context *intel = &brw->intel;
-
-   BEGIN_BATCH(2);
+   BEGIN_BATCH(8);
    OUT_BATCH(_3DSTATE_URB_VS << 16 | (2 - 2));
    OUT_BATCH(nr_vs_entries |
              ((vs_size - 1) << GEN7_URB_ENTRY_SIZE_SHIFT) |
              (vs_start << GEN7_URB_STARTING_ADDRESS_SHIFT));
-   ADVANCE_BATCH();
 
    /* Allocate the GS, HS, and DS zero space - we don't use them. */
-   BEGIN_BATCH(2);
    OUT_BATCH(_3DSTATE_URB_GS << 16 | (2 - 2));
    OUT_BATCH((0 << GEN7_URB_ENTRY_SIZE_SHIFT) |
              (vs_start << GEN7_URB_STARTING_ADDRESS_SHIFT));
-   ADVANCE_BATCH();
 
-   BEGIN_BATCH(2);
    OUT_BATCH(_3DSTATE_URB_HS << 16 | (2 - 2));
    OUT_BATCH((0 << GEN7_URB_ENTRY_SIZE_SHIFT) |
              (vs_start << GEN7_URB_STARTING_ADDRESS_SHIFT));
-   ADVANCE_BATCH();
 
-   BEGIN_BATCH(2);
    OUT_BATCH(_3DSTATE_URB_DS << 16 | (2 - 2));
    OUT_BATCH((0 << GEN7_URB_ENTRY_SIZE_SHIFT) |
              (vs_start << GEN7_URB_STARTING_ADDRESS_SHIFT));