i965/gen7: Emit workaround flush when changing GS enable state.

v2: Don't go to extra work to avoid extraneous flushes.  (Previous
experiments in the kernel have suggested that flushing the pipeline
when it is already empty is extremely cheap).

Cc: "10.0" <mesa-stable@lists.freedesktop.org>

Reviewed-by: Eric Anholt <eric@anholt.net>

7 files changed, 72 insertions, 22 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_context.c b/src/mesa/drivers/dri/i965/brw_context.c
index bee98e3..0399ec0 100644
--- a/src/mesa/drivers/dri/i965/brw_context.c
+++ b/src/mesa/drivers/dri/i965/brw_context.c
@@ -764,6 +764,7 @@ brwCreateContext(gl_api api,
 
    brw->prim_restart.in_progress = false;
    brw->prim_restart.enable_cut_index = false;
+   brw->gs.enabled = false;
 
    if (brw->gen < 6) {
       brw->curbe.last_buf = calloc(1, 4096);
diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h
index 8b1cbb3..4a08986 100644
--- a/src/mesa/drivers/dri/i965/brw_context.h
+++ b/src/mesa/drivers/dri/i965/brw_context.h
@@ -1300,6 +1300,12 @@ struct brw_context
    struct {
       struct brw_stage_state base;
       struct brw_gs_prog_data *prog_data;
+
+      /**
+       * True if the 3DSTATE_GS command most recently emitted to the 3D
+       * pipeline enabled the GS; false otherwise.
+       */
+      bool enabled;
    } gs;
 
    struct {
diff --git a/src/mesa/drivers/dri/i965/gen7_blorp.cpp b/src/mesa/drivers/dri/i965/gen7_blorp.cpp
index 540c46d..d481538 100644
--- a/src/mesa/drivers/dri/i965/gen7_blorp.cpp
+++ b/src/mesa/drivers/dri/i965/gen7_blorp.cpp
@@ -402,6 +402,21 @@ gen7_blorp_emit_gs_disable(struct brw_context *brw,
    OUT_BATCH(0);
    ADVANCE_BATCH();
 
+   /**
+    * From Graphics BSpec: 3D-Media-GPGPU Engine > 3D Pipeline Stages >
+    * Geometry > Geometry Shader > State:
+    *
+    *     "Note: Because of corruption in IVB:GT2, software needs to flush the
+    *     whole fixed function pipeline when the GS enable changes value in
+    *     the 3DSTATE_GS."
+    *
+    * The hardware architects have clarified that in this context "flush the
+    * whole fixed function pipeline" means to emit a PIPE_CONTROL with the "CS
+    * Stall" bit set.
+    */
+   if (!brw->is_haswell && brw->gt == 2 && brw->gs.enabled)
+      gen7_emit_cs_stall_flush(brw);
+
    BEGIN_BATCH(7);
    OUT_BATCH(_3DSTATE_GS << 16 | (7 - 2));
    OUT_BATCH(0);
@@ -411,6 +426,7 @@ gen7_blorp_emit_gs_disable(struct brw_context *brw,
    OUT_BATCH(0);
    OUT_BATCH(0);
    ADVANCE_BATCH();
+   brw->gs.enabled = false;
 }
 
 /* 3DSTATE_STREAMOUT
diff --git a/src/mesa/drivers/dri/i965/gen7_gs_state.c b/src/mesa/drivers/dri/i965/gen7_gs_state.c
index 584f2db..d2ba354 100644
--- a/src/mesa/drivers/dri/i965/gen7_gs_state.c
+++ b/src/mesa/drivers/dri/i965/gen7_gs_state.c
@@ -80,6 +80,21 @@ upload_gs_state(struct brw_context *brw)
 
    gen7_upload_constant_state(brw, stage_state, active, _3DSTATE_CONSTANT_GS);
 
+   /**
+    * From Graphics BSpec: 3D-Media-GPGPU Engine > 3D Pipeline Stages >
+    * Geometry > Geometry Shader > State:
+    *
+    *     "Note: Because of corruption in IVB:GT2, software needs to flush the
+    *     whole fixed function pipeline when the GS enable changes value in
+    *     the 3DSTATE_GS."
+    *
+    * The hardware architects have clarified that in this context "flush the
+    * whole fixed function pipeline" means to emit a PIPE_CONTROL with the "CS
+    * Stall" bit set.
+    */
+   if (!brw->is_haswell && brw->gt == 2 && brw->gs.enabled != active)
+      gen7_emit_cs_stall_flush(brw);
+
    if (active) {
       BEGIN_BATCH(7);
       OUT_BATCH(_3DSTATE_GS << 16 | (7 - 2));
@@ -176,6 +191,7 @@ upload_gs_state(struct brw_context *brw)
       OUT_BATCH(0);
       ADVANCE_BATCH();
    }
+   brw->gs.enabled = active;
 }
 
 const struct brw_tracked_state gen7_gs_state = {
diff --git a/src/mesa/drivers/dri/i965/gen7_urb.c b/src/mesa/drivers/dri/i965/gen7_urb.c
index 6dcdfe4..c638586 100644
--- a/src/mesa/drivers/dri/i965/gen7_urb.c
+++ b/src/mesa/drivers/dri/i965/gen7_urb.c
@@ -122,28 +122,8 @@ gen7_emit_push_constant_state(struct brw_context *brw, unsigned vs_size,
     *
     * No such restriction exists for Haswell.
     */
-   if (!brw->is_haswell) {
-      BEGIN_BATCH(4);
-      OUT_BATCH(_3DSTATE_PIPE_CONTROL | (4 - 2));
-      /* From p61 of the Ivy Bridge PRM (1.10.4 PIPE_CONTROL Command: DW1[20]
-       * CS Stall):
-       *
-       *     One of the following must also be set:
-       *     - Render Target Cache Flush Enable ([12] of DW1)
-       *     - Depth Cache Flush Enable ([0] of DW1)
-       *     - Stall at Pixel Scoreboard ([1] of DW1)
-       *     - Depth Stall ([13] of DW1)
-       *     - Post-Sync Operation ([13] of DW1)
-       *
-       * We choose to do a Post-Sync Operation (Write Immediate Data), since
-       * it seems like it will incur the least additional performance penalty.
-       */
-      OUT_BATCH(PIPE_CONTROL_CS_STALL | PIPE_CONTROL_WRITE_IMMEDIATE);
-      OUT_RELOC(brw->batch.workaround_bo,
-                I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION, 0);
-      OUT_BATCH(0);
-      ADVANCE_BATCH();
-   }
+   if (!brw->is_haswell)
+      gen7_emit_cs_stall_flush(brw);
 }
 
 const struct brw_tracked_state gen7_push_constant_space = {
diff --git a/src/mesa/drivers/dri/i965/intel_batchbuffer.c b/src/mesa/drivers/dri/i965/intel_batchbuffer.c
index fb0b45b..babe9ea 100644
--- a/src/mesa/drivers/dri/i965/intel_batchbuffer.c
+++ b/src/mesa/drivers/dri/i965/intel_batchbuffer.c
@@ -511,6 +511,36 @@ gen7_emit_vs_workaround_flush(struct brw_context *brw)
    ADVANCE_BATCH();
 }
 
+
+/**
+ * Emit a PIPE_CONTROL command for gen7 with the CS Stall bit set.
+ */
+void
+gen7_emit_cs_stall_flush(struct brw_context *brw)
+{
+   BEGIN_BATCH(4);
+   OUT_BATCH(_3DSTATE_PIPE_CONTROL | (4 - 2));
+   /* From p61 of the Ivy Bridge PRM (1.10.4 PIPE_CONTROL Command: DW1[20]
+    * CS Stall):
+    *
+    *     One of the following must also be set:
+    *     - Render Target Cache Flush Enable ([12] of DW1)
+    *     - Depth Cache Flush Enable ([0] of DW1)
+    *     - Stall at Pixel Scoreboard ([1] of DW1)
+    *     - Depth Stall ([13] of DW1)
+    *     - Post-Sync Operation ([13] of DW1)
+    *
+    * We choose to do a Post-Sync Operation (Write Immediate Data), since
+    * it seems like it will incur the least additional performance penalty.
+    */
+   OUT_BATCH(PIPE_CONTROL_CS_STALL | PIPE_CONTROL_WRITE_IMMEDIATE);
+   OUT_RELOC(brw->batch.workaround_bo,
+             I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION, 0);
+   OUT_BATCH(0);
+   ADVANCE_BATCH();
+}
+
+
 /**
  * Emits a PIPE_CONTROL with a non-zero post-sync operation, for
  * implementing two workarounds on gen6.  From section 1.4.7.1
diff --git a/src/mesa/drivers/dri/i965/intel_batchbuffer.h b/src/mesa/drivers/dri/i965/intel_batchbuffer.h
index d46f48e..cabbb69 100644
--- a/src/mesa/drivers/dri/i965/intel_batchbuffer.h
+++ b/src/mesa/drivers/dri/i965/intel_batchbuffer.h
@@ -59,6 +59,7 @@ void intel_batchbuffer_emit_mi_flush(struct brw_context *brw);
 void intel_emit_post_sync_nonzero_flush(struct brw_context *brw);
 void intel_emit_depth_stall_flushes(struct brw_context *brw);
 void gen7_emit_vs_workaround_flush(struct brw_context *brw);
+void gen7_emit_cs_stall_flush(struct brw_context *brw);
 
 static INLINE uint32_t float_as_int(float f)
 {