i965: Emit SNB FF unit state after the unit's push constants.

There's a BUN for the WM unit that says WM_STATE must immediately
follow PS_CONSTANTS, which this addresses.  Presumably other units are
roughly the same, too.

diff --git a/src/mesa/drivers/dri/i965/gen6_gs_state.c b/src/mesa/drivers/dri/i965/gen6_gs_state.c
index 9a4a13810ad3b9e915b8d249f595bf7f7fe62a44..4032605a8558bd16cf2dd46b78ab8db1c22c08e8 100644 (file)
--- a/src/mesa/drivers/dri/i965/gen6_gs_state.c
+++ b/src/mesa/drivers/dri/i965/gen6_gs_state.c
@@ -38,6 +38,17 @@ upload_gs_state(struct brw_context *brw)
 {
    struct intel_context *intel = &brw->intel;
 
+   /* Disable all the constant buffers. */
+   BEGIN_BATCH(5);
+   OUT_BATCH(CMD_3D_CONSTANT_GS_STATE << 16 | (5 - 2));
+   OUT_BATCH(0);
+   OUT_BATCH(0);
+   OUT_BATCH(0);
+   OUT_BATCH(0);
+   ADVANCE_BATCH();
+
+   intel_batchbuffer_emit_mi_flush(intel->batch);
+
    if (brw->gs.prog_bo) {
       BEGIN_BATCH(7);
       OUT_BATCH(CMD_3D_GS_STATE << 16 | (7 - 2));
@@ -69,17 +80,6 @@ upload_gs_state(struct brw_context *brw)
       OUT_BATCH(0);
       ADVANCE_BATCH();
    }
-
-   /* Disable all the constant buffers. */
-   BEGIN_BATCH(5);
-   OUT_BATCH(CMD_3D_CONSTANT_GS_STATE << 16 | (5 - 2));
-   OUT_BATCH(0);
-   OUT_BATCH(0);
-   OUT_BATCH(0);
-   OUT_BATCH(0);
-   ADVANCE_BATCH();
-
-   intel_batchbuffer_emit_mi_flush(intel->batch);
 }
 
 const struct brw_tracked_state gen6_gs_state = {

diff --git a/src/mesa/drivers/dri/i965/gen6_vs_state.c b/src/mesa/drivers/dri/i965/gen6_vs_state.c
index ba00e4e1fb6e00dc0a0566320500b97cab768469..04fc16e6792d7566a1c0fd9cbb98fc134cc026f4 100644 (file)
--- a/src/mesa/drivers/dri/i965/gen6_vs_state.c
+++ b/src/mesa/drivers/dri/i965/gen6_vs_state.c
@@ -46,21 +46,6 @@ upload_vs_state(struct brw_context *brw)
    drm_intel_bo *constant_bo;
    int i;
 
-   BEGIN_BATCH(6);
-   OUT_BATCH(CMD_3D_VS_STATE << 16 | (6 - 2));
-   OUT_RELOC(brw->vs.prog_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
-   OUT_BATCH((0 << GEN6_VS_SAMPLER_COUNT_SHIFT) |
-            (brw->vs.nr_surfaces << GEN6_VS_BINDING_TABLE_ENTRY_COUNT_SHIFT));
-   OUT_BATCH(0); /* scratch space base offset */
-   OUT_BATCH((1 << GEN6_VS_DISPATCH_START_GRF_SHIFT) |
-            (brw->vs.prog_data->urb_read_length << GEN6_VS_URB_READ_LENGTH_SHIFT) |
-            (0 << GEN6_VS_URB_ENTRY_READ_OFFSET_SHIFT));
-   OUT_BATCH((0 << GEN6_VS_MAX_THREADS_SHIFT) |
-            GEN6_VS_STATISTICS_ENABLE);
-   ADVANCE_BATCH();
-
-   intel_batchbuffer_emit_mi_flush(intel->batch);
-
    if (vp->use_const_buffer || nr_params == 0) {
       /* Disable the push constant buffers. */
       BEGIN_BATCH(5);
@@ -106,6 +91,21 @@ upload_vs_state(struct brw_context *brw)
    }
 
    intel_batchbuffer_emit_mi_flush(intel->batch);
+
+   BEGIN_BATCH(6);
+   OUT_BATCH(CMD_3D_VS_STATE << 16 | (6 - 2));
+   OUT_RELOC(brw->vs.prog_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
+   OUT_BATCH((0 << GEN6_VS_SAMPLER_COUNT_SHIFT) |
+            (brw->vs.nr_surfaces << GEN6_VS_BINDING_TABLE_ENTRY_COUNT_SHIFT));
+   OUT_BATCH(0); /* scratch space base offset */
+   OUT_BATCH((1 << GEN6_VS_DISPATCH_START_GRF_SHIFT) |
+            (brw->vs.prog_data->urb_read_length << GEN6_VS_URB_READ_LENGTH_SHIFT) |
+            (0 << GEN6_VS_URB_ENTRY_READ_OFFSET_SHIFT));
+   OUT_BATCH((0 << GEN6_VS_MAX_THREADS_SHIFT) |
+            GEN6_VS_STATISTICS_ENABLE);
+   ADVANCE_BATCH();
+
+   intel_batchbuffer_emit_mi_flush(intel->batch);
 }
 
 const struct brw_tracked_state gen6_vs_state = {

diff --git a/src/mesa/drivers/dri/i965/gen6_wm_state.c b/src/mesa/drivers/dri/i965/gen6_wm_state.c
index d896f7dcc0f058d6658729c55c860f0c8e3d9f0a..f4ae7ed75052a888433b5bb28ed195126020a6d7 100644 (file)
--- a/src/mesa/drivers/dri/i965/gen6_wm_state.c
+++ b/src/mesa/drivers/dri/i965/gen6_wm_state.c
@@ -47,6 +47,49 @@ upload_wm_state(struct brw_context *brw)
    int i;
    uint32_t dw2, dw4, dw5, dw6;
 
+   if (fp->use_const_buffer || nr_params == 0) {
+      /* Disable the push constant buffers. */
+      BEGIN_BATCH(5);
+      OUT_BATCH(CMD_3D_CONSTANT_PS_STATE << 16 | (5 - 2));
+      OUT_BATCH(0);
+      OUT_BATCH(0);
+      OUT_BATCH(0);
+      OUT_BATCH(0);
+      ADVANCE_BATCH();
+   } else {
+      /* Updates the ParamaterValues[i] pointers for all parameters of the
+       * basic type of PROGRAM_STATE_VAR.
+       */
+      _mesa_load_state_parameters(ctx, fp->program.Base.Parameters);
+
+      constant_bo = drm_intel_bo_alloc(intel->bufmgr, "WM constant_bo",
+                                      nr_params * 4 * sizeof(float),
+                                      4096);
+      intel_bo_map_gtt_preferred(intel, constant_bo, GL_TRUE);
+      for (i = 0; i < nr_params; i++) {
+        memcpy((char *)constant_bo->virtual + i * 4 * sizeof(float),
+               fp->program.Base.Parameters->ParameterValues[i],
+               4 * sizeof(float));
+      }
+      intel_bo_unmap_gtt_preferred(intel, constant_bo);
+
+      BEGIN_BATCH(5);
+      OUT_BATCH(CMD_3D_CONSTANT_PS_STATE << 16 |
+               GEN6_CONSTANT_BUFFER_0_ENABLE |
+               (5 - 2));
+      OUT_RELOC(constant_bo,
+               I915_GEM_DOMAIN_RENDER, 0, /* XXX: bad domain */
+               ALIGN(nr_params, 2) / 2 - 1);
+      OUT_BATCH(0);
+      OUT_BATCH(0);
+      OUT_BATCH(0);
+      ADVANCE_BATCH();
+
+      drm_intel_bo_unreference(constant_bo);
+   }
+
+   intel_batchbuffer_emit_mi_flush(intel->batch);
+
    dw2 = dw4 = dw5 = dw6 = 0;
    dw4 |= GEN6_WM_STATISTICS_ENABLE;
    dw5 |= GEN6_WM_LINE_AA_WIDTH_1_0;
@@ -103,49 +146,6 @@ upload_wm_state(struct brw_context *brw)
    ADVANCE_BATCH();
 
    intel_batchbuffer_emit_mi_flush(intel->batch);
-
-   if (fp->use_const_buffer || nr_params == 0) {
-      /* Disable the push constant buffers. */
-      BEGIN_BATCH(5);
-      OUT_BATCH(CMD_3D_CONSTANT_PS_STATE << 16 | (5 - 2));
-      OUT_BATCH(0);
-      OUT_BATCH(0);
-      OUT_BATCH(0);
-      OUT_BATCH(0);
-      ADVANCE_BATCH();
-   } else {
-      /* Updates the ParamaterValues[i] pointers for all parameters of the
-       * basic type of PROGRAM_STATE_VAR.
-       */
-      _mesa_load_state_parameters(ctx, fp->program.Base.Parameters);
-
-      constant_bo = drm_intel_bo_alloc(intel->bufmgr, "WM constant_bo",
-                                      nr_params * 4 * sizeof(float),
-                                      4096);
-      intel_bo_map_gtt_preferred(intel, constant_bo, GL_TRUE);
-      for (i = 0; i < nr_params; i++) {
-        memcpy((char *)constant_bo->virtual + i * 4 * sizeof(float),
-               fp->program.Base.Parameters->ParameterValues[i],
-               4 * sizeof(float));
-      }
-      intel_bo_unmap_gtt_preferred(intel, constant_bo);
-
-      BEGIN_BATCH(5);
-      OUT_BATCH(CMD_3D_CONSTANT_PS_STATE << 16 |
-               GEN6_CONSTANT_BUFFER_0_ENABLE |
-               (5 - 2));
-      OUT_RELOC(constant_bo,
-               I915_GEM_DOMAIN_RENDER, 0, /* XXX: bad domain */
-               ALIGN(nr_params, 2) / 2 - 1);
-      OUT_BATCH(0);
-      OUT_BATCH(0);
-      OUT_BATCH(0);
-      ADVANCE_BATCH();
-
-      drm_intel_bo_unreference(constant_bo);
-   }
-
-   intel_batchbuffer_emit_mi_flush(intel->batch);
 }
 
 const struct brw_tracked_state gen6_wm_state = {