X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;ds=sidebyside;f=src%2Fmesa%2Fdrivers%2Fdri%2Fi965%2Fgen6_queryobj.c;h=d508c4c9278f78a7e66bcce301287eb6668833b4;hb=a7e9b31d5bf98bdaabbb8b5c2459eb2c3a0af579;hp=56e9d5db9374f6c4f8cc2c4e709b75923a840158;hpb=32a3f5f6d768e5828be1d1f46b1b3f819f55cba8;p=mesa.git

diff --git a/src/mesa/drivers/dri/i965/gen6_queryobj.c b/src/mesa/drivers/dri/i965/gen6_queryobj.c
index 56e9d5db937..d508c4c9278 100644
--- a/src/mesa/drivers/dri/i965/gen6_queryobj.c
+++ b/src/mesa/drivers/dri/i965/gen6_queryobj.c
@@ -39,61 +39,6 @@
 #include "intel_batchbuffer.h"
 #include "intel_reg.h"
 
-/**
- * Emit PIPE_CONTROLs to write the current GPU timestamp into a buffer.
- */
-static void
-write_timestamp(struct brw_context *brw, drm_intel_bo *query_bo, int idx)
-{
-   /* Emit workaround flushes: */
-   if (brw->gen == 6) {
-      /* The timestamp write below is a non-zero post-sync op, which on
-       * Gen6 necessitates a CS stall.  CS stalls need stall at scoreboard
-       * set.  See the comments for intel_emit_post_sync_nonzero_flush().
-       */
-      BEGIN_BATCH(4);
-      OUT_BATCH(_3DSTATE_PIPE_CONTROL | (4 - 2));
-      OUT_BATCH(PIPE_CONTROL_CS_STALL | PIPE_CONTROL_STALL_AT_SCOREBOARD);
-      OUT_BATCH(0);
-      OUT_BATCH(0);
-      ADVANCE_BATCH();
-   }
-
-   BEGIN_BATCH(5);
-   OUT_BATCH(_3DSTATE_PIPE_CONTROL | (5 - 2));
-   OUT_BATCH(PIPE_CONTROL_WRITE_TIMESTAMP);
-   OUT_RELOC(query_bo,
-             I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
-             PIPE_CONTROL_GLOBAL_GTT_WRITE |
-             idx * sizeof(uint64_t));
-   OUT_BATCH(0);
-   OUT_BATCH(0);
-   ADVANCE_BATCH();
-}
-
-/**
- * Emit PIPE_CONTROLs to write the PS_DEPTH_COUNT register into a buffer.
- */
-static void
-write_depth_count(struct brw_context *brw, drm_intel_bo *query_bo, int idx)
-{
-   /* Emit Sandybridge workaround flush: */
-   if (brw->gen == 6)
-      intel_emit_post_sync_nonzero_flush(brw);
-
-   BEGIN_BATCH(5);
-   OUT_BATCH(_3DSTATE_PIPE_CONTROL | (5 - 2));
-   OUT_BATCH(PIPE_CONTROL_DEPTH_STALL |
-             PIPE_CONTROL_WRITE_DEPTH_COUNT);
-   OUT_RELOC(query_bo,
-             I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
-             PIPE_CONTROL_GLOBAL_GTT_WRITE |
-             (idx * sizeof(uint64_t)));
-   OUT_BATCH(0);
-   OUT_BATCH(0);
-   ADVANCE_BATCH();
-}
-
 /*
  * Write an arbitrary 64-bit register to a buffer via MI_STORE_REGISTER_MEM.
  *
@@ -112,43 +57,112 @@ brw_store_register_mem64(struct brw_context *brw,
    /* MI_STORE_REGISTER_MEM only stores a single 32-bit value, so to
     * read a full 64-bit register, we need to do two of them.
     */
-   BEGIN_BATCH(3);
-   OUT_BATCH(MI_STORE_REGISTER_MEM | (3 - 2));
-   OUT_BATCH(reg);
-   OUT_RELOC(bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
-             idx * sizeof(uint64_t));
-   ADVANCE_BATCH();
-
-   BEGIN_BATCH(3);
-   OUT_BATCH(MI_STORE_REGISTER_MEM | (3 - 2));
-   OUT_BATCH(reg + sizeof(uint32_t));
-   OUT_RELOC(bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
-             sizeof(uint32_t) + idx * sizeof(uint64_t));
-   ADVANCE_BATCH();
+   if (brw->gen >= 8) {
+      BEGIN_BATCH(8);
+      OUT_BATCH(MI_STORE_REGISTER_MEM | (4 - 2));
+      OUT_BATCH(reg);
+      OUT_RELOC64(bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+                  idx * sizeof(uint64_t));
+      OUT_BATCH(MI_STORE_REGISTER_MEM | (4 - 2));
+      OUT_BATCH(reg + sizeof(uint32_t));
+      OUT_RELOC64(bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+                  sizeof(uint32_t) + idx * sizeof(uint64_t));
+      ADVANCE_BATCH();
+   } else {
+      BEGIN_BATCH(6);
+      OUT_BATCH(MI_STORE_REGISTER_MEM | (3 - 2));
+      OUT_BATCH(reg);
+      OUT_RELOC(bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+                idx * sizeof(uint64_t));
+      OUT_BATCH(MI_STORE_REGISTER_MEM | (3 - 2));
+      OUT_BATCH(reg + sizeof(uint32_t));
+      OUT_RELOC(bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+                sizeof(uint32_t) + idx * sizeof(uint64_t));
+      ADVANCE_BATCH();
+   }
 }
 
 static void
 write_primitives_generated(struct brw_context *brw,
-                           drm_intel_bo *query_bo, int idx)
+                           drm_intel_bo *query_bo, int stream, int idx)
 {
-   intel_batchbuffer_emit_mi_flush(brw);
+   brw_emit_mi_flush(brw);
 
-   brw_store_register_mem64(brw, query_bo, CL_INVOCATION_COUNT, idx);
+   if (brw->gen >= 7 && stream > 0) {
+      brw_store_register_mem64(brw, query_bo,
+                               GEN7_SO_PRIM_STORAGE_NEEDED(stream), idx);
+   } else {
+      brw_store_register_mem64(brw, query_bo, CL_INVOCATION_COUNT, idx);
+   }
 }
 
 static void
 write_xfb_primitives_written(struct brw_context *brw,
-                             drm_intel_bo *bo, int idx)
+                             drm_intel_bo *bo, int stream, int idx)
 {
-   intel_batchbuffer_emit_mi_flush(brw);
+   brw_emit_mi_flush(brw);
 
    if (brw->gen >= 7) {
-      brw_store_register_mem64(brw, bo, GEN7_SO_NUM_PRIMS_WRITTEN(0), idx);
+      brw_store_register_mem64(brw, bo, GEN7_SO_NUM_PRIMS_WRITTEN(stream), idx);
    } else {
       brw_store_register_mem64(brw, bo, GEN6_SO_NUM_PRIMS_WRITTEN, idx);
    }
 }
 
+static inline const int
+pipeline_target_to_index(int target)
+{
+   if (target == GL_GEOMETRY_SHADER_INVOCATIONS)
+      return MAX_PIPELINE_STATISTICS - 1;
+   else
+      return target - GL_VERTICES_SUBMITTED_ARB;
+}
+
+static void
+emit_pipeline_stat(struct brw_context *brw, drm_intel_bo *bo,
+                   int stream, int target, int idx)
+{
+   /* One source of confusion is the tessellation shader statistics. The
+    * hardware has no statistics specific to the TE unit. Ideally we could have
+    * the HS primitives for TESS_CONTROL_SHADER_PATCHES_ARB, and the DS
+    * invocations as the register for TESS_CONTROL_SHADER_PATCHES_ARB.
+    * Unfortunately we don't have HS primitives, we only have HS invocations.
+    */
+
+   /* Everything except GEOMETRY_SHADER_INVOCATIONS can be kept in a simple
+    * lookup table
+    */
+   static const uint32_t target_to_register[] = {
+      IA_VERTICES_COUNT,   /* VERTICES_SUBMITTED */
+      IA_PRIMITIVES_COUNT, /* PRIMITIVES_SUBMITTED */
+      VS_INVOCATION_COUNT, /* VERTEX_SHADER_INVOCATIONS */
+      HS_INVOCATION_COUNT, /* TESS_CONTROL_SHADER_PATCHES */
+      DS_INVOCATION_COUNT, /* TESS_EVALUATION_SHADER_INVOCATIONS */
+      GS_PRIMITIVES_COUNT, /* GEOMETRY_SHADER_PRIMITIVES_EMITTED */
+      PS_INVOCATION_COUNT, /* FRAGMENT_SHADER_INVOCATIONS */
+      CS_INVOCATION_COUNT, /* COMPUTE_SHADER_INVOCATIONS */
+      CL_INVOCATION_COUNT, /* CLIPPING_INPUT_PRIMITIVES */
+      CL_PRIMITIVES_COUNT, /* CLIPPING_OUTPUT_PRIMITIVES */
+      GS_INVOCATION_COUNT /* This one is special... */
+   };
+   STATIC_ASSERT(ARRAY_SIZE(target_to_register) == MAX_PIPELINE_STATISTICS);
+   uint32_t reg = target_to_register[pipeline_target_to_index(target)];
+   /* Gen6 GS code counts full primitives, that is, it won't count individual
+    * triangles in a triangle strip. Use CL_INVOCATION_COUNT for that.
+    */
+   if (brw->gen == 6 && target == GL_GEOMETRY_SHADER_PRIMITIVES_EMITTED_ARB)
+      reg = CL_INVOCATION_COUNT;
+   assert(reg != 0);
+
+   /* Emit a flush to make sure various parts of the pipeline are complete and
+    * we get an accurate value
+    */
+   brw_emit_mi_flush(brw);
+
+   brw_store_register_mem64(brw, bo, reg, idx);
+}
+
+
 /**
  * Wait on the query object's BO and calculate the final result.
  */
@@ -161,20 +175,7 @@ gen6_queryobj_get_results(struct gl_context *ctx,
    if (query->bo == NULL)
       return;
 
-   /* If the application has requested the query result, but this batch is
-    * still contributing to it, flush it now so the results will be present
-    * when mapped.
-    */
-   if (drm_intel_bo_references(brw->batch.bo, query->bo))
-      intel_batchbuffer_flush(brw);
-
-   if (unlikely(brw->perf_debug)) {
-      if (drm_intel_bo_busy(query->bo)) {
-         perf_debug("Stalling on the GPU waiting for a query object.\n");
-      }
-   }
-
-   drm_intel_bo_map(query->bo, false);
+   brw_bo_map(brw, query->bo, false, "query object");
    uint64_t *results = query->bo->virtual;
    switch (query->Base.Target) {
    case GL_TIME_ELAPSED:
@@ -222,12 +223,37 @@ gen6_queryobj_get_results(struct gl_context *ctx,
 
    case GL_PRIMITIVES_GENERATED:
    case GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN:
+   case GL_VERTICES_SUBMITTED_ARB:
+   case GL_PRIMITIVES_SUBMITTED_ARB:
+   case GL_VERTEX_SHADER_INVOCATIONS_ARB:
+   case GL_GEOMETRY_SHADER_INVOCATIONS:
+   case GL_GEOMETRY_SHADER_PRIMITIVES_EMITTED_ARB:
+   case GL_CLIPPING_INPUT_PRIMITIVES_ARB:
+   case GL_CLIPPING_OUTPUT_PRIMITIVES_ARB:
+   case GL_COMPUTE_SHADER_INVOCATIONS_ARB:
+   case GL_TESS_CONTROL_SHADER_PATCHES_ARB:
+   case GL_TESS_EVALUATION_SHADER_INVOCATIONS_ARB:
       query->Base.Result = results[1] - results[0];
       break;
 
-   default:
-      assert(!"Unrecognized query target in brw_queryobj_get_results()");
+   case GL_FRAGMENT_SHADER_INVOCATIONS_ARB:
+      query->Base.Result = (results[1] - results[0]);
+      /* Implement the "WaDividePSInvocationCountBy4:HSW,BDW" workaround:
+       * "Invocation counter is 4 times actual.  WA: SW to divide HW reported
+       *  PS Invocations value by 4."
+       *
+       * Prior to Haswell, invocation count was counted by the WM, and it
+       * buggily counted invocations in units of subspans (2x2 unit). To get the
+       * correct value, the CS multiplied this by 4. With HSW the logic moved,
+       * and correctly emitted the number of pixel shader invocations, but,
+       * whomever forgot to undo the multiply by 4.
+       */
+      if (brw->gen == 8 || brw->is_haswell)
+         query->Base.Result /= 4;
       break;
+
+   default:
+      unreachable("Unrecognized query target in brw_queryobj_get_results()");
    }
    drm_intel_bo_unmap(query->bo);
 
@@ -236,6 +262,8 @@ gen6_queryobj_get_results(struct gl_context *ctx,
     */
    drm_intel_bo_unreference(query->bo);
    query->bo = NULL;
+
+   query->Base.Ready = true;
 }
 
 /**
@@ -275,26 +303,39 @@ gen6_begin_query(struct gl_context *ctx, struct gl_query_object *q)
        * obtain the time elapsed.  Notably, this includes time elapsed while
        * the system was doing other work, such as running other applications.
        */
-      write_timestamp(brw, query->bo, 0);
+      brw_write_timestamp(brw, query->bo, 0);
       break;
 
    case GL_ANY_SAMPLES_PASSED:
    case GL_ANY_SAMPLES_PASSED_CONSERVATIVE:
    case GL_SAMPLES_PASSED_ARB:
-      write_depth_count(brw, query->bo, 0);
+      brw_write_depth_count(brw, query->bo, 0);
       break;
 
    case GL_PRIMITIVES_GENERATED:
-      write_primitives_generated(brw, query->bo, 0);
+      write_primitives_generated(brw, query->bo, query->Base.Stream, 0);
       break;
 
    case GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN:
-      write_xfb_primitives_written(brw, query->bo, 0);
+      write_xfb_primitives_written(brw, query->bo, query->Base.Stream, 0);
       break;
 
-   default:
-      assert(!"Unrecognized query target in brw_begin_query()");
+   case GL_VERTICES_SUBMITTED_ARB:
+   case GL_PRIMITIVES_SUBMITTED_ARB:
+   case GL_VERTEX_SHADER_INVOCATIONS_ARB:
+   case GL_GEOMETRY_SHADER_INVOCATIONS:
+   case GL_GEOMETRY_SHADER_PRIMITIVES_EMITTED_ARB:
+   case GL_FRAGMENT_SHADER_INVOCATIONS_ARB:
+   case GL_CLIPPING_INPUT_PRIMITIVES_ARB:
+   case GL_CLIPPING_OUTPUT_PRIMITIVES_ARB:
+   case GL_COMPUTE_SHADER_INVOCATIONS_ARB:
+   case GL_TESS_CONTROL_SHADER_PATCHES_ARB:
+   case GL_TESS_EVALUATION_SHADER_INVOCATIONS_ARB:
+      emit_pipeline_stat(brw, query->bo, query->Base.Stream, query->Base.Target, 0);
       break;
+
+   default:
+      unreachable("Unrecognized query target in brw_begin_query()");
    }
 }
 
@@ -314,27 +355,62 @@ gen6_end_query(struct gl_context *ctx, struct gl_query_object *q)
 
    switch (query->Base.Target) {
    case GL_TIME_ELAPSED:
-      write_timestamp(brw, query->bo, 1);
+      brw_write_timestamp(brw, query->bo, 1);
       break;
 
    case GL_ANY_SAMPLES_PASSED:
    case GL_ANY_SAMPLES_PASSED_CONSERVATIVE:
    case GL_SAMPLES_PASSED_ARB:
-      write_depth_count(brw, query->bo, 1);
+      brw_write_depth_count(brw, query->bo, 1);
       break;
 
    case GL_PRIMITIVES_GENERATED:
-      write_primitives_generated(brw, query->bo, 1);
+      write_primitives_generated(brw, query->bo, query->Base.Stream, 1);
       break;
 
    case GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN:
-      write_xfb_primitives_written(brw, query->bo, 1);
+      write_xfb_primitives_written(brw, query->bo, query->Base.Stream, 1);
       break;
 
-   default:
-      assert(!"Unrecognized query target in brw_end_query()");
+   case GL_VERTICES_SUBMITTED_ARB:
+   case GL_PRIMITIVES_SUBMITTED_ARB:
+   case GL_VERTEX_SHADER_INVOCATIONS_ARB:
+   case GL_GEOMETRY_SHADER_PRIMITIVES_EMITTED_ARB:
+   case GL_FRAGMENT_SHADER_INVOCATIONS_ARB:
+   case GL_COMPUTE_SHADER_INVOCATIONS_ARB:
+   case GL_CLIPPING_INPUT_PRIMITIVES_ARB:
+   case GL_CLIPPING_OUTPUT_PRIMITIVES_ARB:
+   case GL_GEOMETRY_SHADER_INVOCATIONS:
+   case GL_TESS_CONTROL_SHADER_PATCHES_ARB:
+   case GL_TESS_EVALUATION_SHADER_INVOCATIONS_ARB:
+      emit_pipeline_stat(brw, query->bo,
+                         query->Base.Stream, query->Base.Target, 1);
       break;
+
+   default:
+      unreachable("Unrecognized query target in brw_end_query()");
    }
+
+   /* The current batch contains the commands to handle EndQuery(),
+    * but they won't actually execute until it is flushed.
+    */
+   query->flushed = false;
+}
+
+/**
+ * Flush the batch if it still references the query object BO.
+ */
+static void
+flush_batch_if_needed(struct brw_context *brw, struct brw_query_object *query)
+{
+   /* If the batch doesn't reference the BO, it must have been flushed
+    * (for example, due to being full).  Record that it's been flushed.
+    */
+   query->flushed = query->flushed ||
+      !drm_intel_bo_references(brw->batch.bo, query->bo);
+
+   if (!query->flushed)
+      intel_batchbuffer_flush(brw);
 }
 
 /**
@@ -345,10 +421,16 @@ gen6_end_query(struct gl_context *ctx, struct gl_query_object *q)
  */
 static void gen6_wait_query(struct gl_context *ctx, struct gl_query_object *q)
 {
+   struct brw_context *brw = brw_context(ctx);
    struct brw_query_object *query = (struct brw_query_object *)q;
 
+   /* If the application has requested the query result, but this batch is
+    * still contributing to it, flush it now to finish that work so the
+    * result will become available (eventually).
+    */
+   flush_batch_if_needed(brw, query);
+
    gen6_queryobj_get_results(ctx, query);
-   query->Base.Ready = true;
 }
 
 /**
@@ -362,6 +444,12 @@ static void gen6_check_query(struct gl_context *ctx, struct gl_query_object *q)
    struct brw_context *brw = brw_context(ctx);
    struct brw_query_object *query = (struct brw_query_object *)q;
 
+   /* If query->bo is NULL, we've already gathered the results - this is a
+    * redundant CheckQuery call.  Ignore it.
+    */
+   if (query->bo == NULL)
+      return;
+
    /* From the GL_ARB_occlusion_query spec:
     *
     *     "Instead of allowing for an infinite loop, performing a
@@ -369,12 +457,10 @@ static void gen6_check_query(struct gl_context *ctx, struct gl_query_object *q)
     *      not ready yet on the first time it is queried.  This ensures that
     *      the async query will return true in finite time.
     */
-   if (query->bo && drm_intel_bo_references(brw->batch.bo, query->bo))
-      intel_batchbuffer_flush(brw);
+   flush_batch_if_needed(brw, query);
 
-   if (query->bo == NULL || !drm_intel_bo_busy(query->bo)) {
+   if (!drm_intel_bo_busy(query->bo)) {
       gen6_queryobj_get_results(ctx, query);
-      query->Base.Ready = true;
    }
 }