nvc0: finish implementation of PIPE_QUERY_SO_OVERFLOW_PREDICATE

This also removes some useless code leftover from old changes.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

5 files changed, 32 insertions, 18 deletions

diff --git a/docs/features.txt b/docs/features.txt
index d579d245eb4..8f003ddd90b 100644
--- a/docs/features.txt
+++ b/docs/features.txt
@@ -232,7 +232,7 @@ GL 4.6, GLSL 4.60
   GL_ARB_shader_group_vote                              DONE (i965, nvc0, radeonsi)
   GL_ARB_spirv_extensions                               in progress (Nicolai Hähnle, Ian Romanick)
   GL_ARB_texture_filter_anisotropic                     DONE (freedreno, i965, nv50, nvc0, r600, radeonsi, softpipe (*), llvmpipe (*))
-  GL_ARB_transform_feedback_overflow_query              DONE (i965/gen6+, radeonsi, llvmpipe, softpipe)
+  GL_ARB_transform_feedback_overflow_query              DONE (i965/gen6+, nvc0, radeonsi, llvmpipe, softpipe)
   GL_KHR_no_error                                       DONE (all drivers)
 
 (*) softpipe and llvmpipe advertise 16x anisotropy but simply ignore the setting
diff --git a/docs/relnotes/18.1.0.html b/docs/relnotes/18.1.0.html
index a89861d2bda..deeb23db03d 100644
--- a/docs/relnotes/18.1.0.html
+++ b/docs/relnotes/18.1.0.html
@@ -46,6 +46,7 @@ Note: some of the new features are only available with certain drivers.
 <ul>
 <li>OpenGL 3.1 with ARB_compatibility on nv50, nvc0, r600, radeonsi, softpipe, llvmpipe, svga</li>
 <li>GL_ARB_bindless_texture on nvc0/maxwell+</li>
+<li>GL_ARB_transform_feedback_overflow_query on nvc0</li>
 <li>GL_EXT_semaphore on radeonsi</li>
 <li>GL_EXT_semaphore_fd on radeonsi</li>
 <li>GL_EXT_shader_framebuffer_fetch on i965 on desktop GL (GLES was already supported)</li>
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_query.c b/src/gallium/drivers/nouveau/nvc0/nvc0_query.c
index d0a9e0c2c04..1a3e4e794c0 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_query.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_query.c
@@ -113,8 +113,9 @@ nvc0_render_condition(struct pipe_context *pipe,
       /* NOTE: comparison of 2 queries only works if both have completed */
       switch (q->type) {
       case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
+      case PIPE_QUERY_SO_OVERFLOW_ANY_PREDICATE:
          cond = condition ? NVC0_3D_COND_MODE_EQUAL :
-                          NVC0_3D_COND_MODE_NOT_EQUAL;
+                            NVC0_3D_COND_MODE_NOT_EQUAL;
          wait = true;
          break;
       case PIPE_QUERY_OCCLUSION_COUNTER:
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_query_hw.c b/src/gallium/drivers/nouveau/nvc0/nvc0_query_hw.c
index 93cd64ae841..a420ed4ac0d 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_query_hw.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_query_hw.c
@@ -181,6 +181,10 @@ nvc0_hw_begin_query(struct nvc0_context *nvc0, struct nvc0_query *q)
    case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
       nvc0_hw_query_get(push, q, 0x10, 0x03005002 | (q->index << 5));
       break;
+   case PIPE_QUERY_SO_OVERFLOW_ANY_PREDICATE:
+      /* XXX: This get actually writes the number of overflowed streams */
+      nvc0_hw_query_get(push, q, 0x10, 0x0f005002);
+      break;
    case PIPE_QUERY_TIME_ELAPSED:
       nvc0_hw_query_get(push, q, 0x10, 0x00005002);
       break;
@@ -243,10 +247,11 @@ nvc0_hw_end_query(struct nvc0_context *nvc0, struct nvc0_query *q)
       nvc0_hw_query_get(push, q, 0x10, 0x06805002 | (q->index << 5));
       break;
    case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
-      /* TODO: How do we sum over all streams for render condition ? */
-      /* PRIMS_DROPPED doesn't write sequence, use a ZERO query to sync on */
       nvc0_hw_query_get(push, q, 0x00, 0x03005002 | (q->index << 5));
-      nvc0_hw_query_get(push, q, 0x20, 0x00005002);
+      break;
+   case PIPE_QUERY_SO_OVERFLOW_ANY_PREDICATE:
+      /* XXX: This get actually writes the number of overflowed streams */
+      nvc0_hw_query_get(push, q, 0x00, 0x0f005002);
       break;
    case PIPE_QUERY_TIMESTAMP:
    case PIPE_QUERY_TIME_ELAPSED:
@@ -334,6 +339,7 @@ nvc0_hw_get_query_result(struct nvc0_context *nvc0, struct nvc0_query *q,
       res64[1] = data64[2] - data64[6];
       break;
    case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
+   case PIPE_QUERY_SO_OVERFLOW_ANY_PREDICATE:
       res8[0] = data64[0] != data64[2];
       break;
    case PIPE_QUERY_TIMESTAMP:
@@ -417,15 +423,22 @@ nvc0_hw_get_query_result_resource(struct nvc0_context *nvc0,
    PUSH_REFN (push, hq->bo, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
    PUSH_REFN (push, buf->bo, buf->domain | NOUVEAU_BO_WR);
    BEGIN_1IC0(push, NVC0_3D(MACRO_QUERY_BUFFER_WRITE), 9);
-   if (q->type == PIPE_QUERY_OCCLUSION_PREDICATE ||
-       q->type == PIPE_QUERY_OCCLUSION_PREDICATE_CONSERVATIVE) /* XXX what if 64-bit? */
+   switch (q->type) {
+   case PIPE_QUERY_OCCLUSION_PREDICATE:
+   case PIPE_QUERY_OCCLUSION_PREDICATE_CONSERVATIVE: /* XXX what if 64-bit? */
+   case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
+   case PIPE_QUERY_SO_OVERFLOW_ANY_PREDICATE:
       PUSH_DATA(push, 0x00000001);
-   else if (result_type == PIPE_QUERY_TYPE_I32)
-      PUSH_DATA(push, 0x7fffffff);
-   else if (result_type == PIPE_QUERY_TYPE_U32)
-      PUSH_DATA(push, 0xffffffff);
-   else
-      PUSH_DATA(push, 0x00000000);
+      break;
+   default:
+      if (result_type == PIPE_QUERY_TYPE_I32)
+         PUSH_DATA(push, 0x7fffffff);
+      else if (result_type == PIPE_QUERY_TYPE_U32)
+         PUSH_DATA(push, 0xffffffff);
+      else
+         PUSH_DATA(push, 0x00000000);
+      break;
+   }
 
    switch (q->type) {
    case PIPE_QUERY_SO_STATISTICS:
@@ -519,6 +532,7 @@ nvc0_hw_create_query(struct nvc0_context *nvc0, unsigned type, unsigned index)
    q = &hq->base;
    q->funcs = &hw_query_funcs;
    q->type = type;
+   q->index = index;
 
    switch (q->type) {
    case PIPE_QUERY_OCCLUSION_COUNTER:
@@ -532,14 +546,14 @@ nvc0_hw_create_query(struct nvc0_context *nvc0, unsigned type, unsigned index)
       space = 512;
       break;
    case PIPE_QUERY_SO_STATISTICS:
-   case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
       hq->is64bit = true;
       space = 64;
       break;
+   case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
+   case PIPE_QUERY_SO_OVERFLOW_ANY_PREDICATE:
    case PIPE_QUERY_PRIMITIVES_GENERATED:
    case PIPE_QUERY_PRIMITIVES_EMITTED:
       hq->is64bit = true;
-      q->index = index;
       space = 32;
       break;
    case PIPE_QUERY_TIME_ELAPSED:
@@ -615,8 +629,6 @@ nvc0_hw_query_fifo_wait(struct nvc0_context *nvc0, struct nvc0_query *q)
    if (hq->is64bit && hq->fence->state < NOUVEAU_FENCE_STATE_EMITTED)
       nouveau_fence_emit(hq->fence);
 
-   if (q->type == PIPE_QUERY_SO_OVERFLOW_PREDICATE) offset += 0x20;
-
    PUSH_SPACE(push, 5);
    PUSH_REFN (push, hq->bo, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
    BEGIN_NVC0(push, SUBC_3D(NV84_SUBCHAN_SEMAPHORE_ADDRESS_HIGH), 4);
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
index b4a2c676f59..eb50149b647 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
@@ -254,6 +254,7 @@ nvc0_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
    case PIPE_CAP_COMPUTE:
    case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX:
    case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
+   case PIPE_CAP_QUERY_SO_OVERFLOW:
       return 1;
    case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
       return nouveau_screen(pscreen)->vram_domain & NOUVEAU_BO_VRAM ? 1 : 0;
@@ -298,7 +299,6 @@ nvc0_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
    case PIPE_CAP_INT64_DIVMOD:
    case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE:
    case PIPE_CAP_NIR_SAMPLERS_AS_DEREF:
-   case PIPE_CAP_QUERY_SO_OVERFLOW:
    case PIPE_CAP_MEMOBJ:
    case PIPE_CAP_LOAD_CONSTBUF:
    case PIPE_CAP_TGSI_ANY_REG_AS_ADDRESS: