gallium/util: replace pipe_condvar_init() with cnd_init()
[mesa.git] / src / gallium / state_trackers / nine / nine_state.c
index 928d82050c92a56807b0f11430452b25528af607..e1a2ff0cf979d8d6888140b04a9cf6b575ff21c6 100644 (file)
  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  * USE OR OTHER DEALINGS IN THE SOFTWARE. */
 
+#define NINE_STATE
+
 #include "device9.h"
+#include "swapchain9.h"
 #include "basetexture9.h"
 #include "buffer9.h"
 #include "indexbuffer9.h"
 #include "pipe/p_context.h"
 #include "pipe/p_state.h"
 #include "cso_cache/cso_context.h"
+#include "util/u_atomic.h"
 #include "util/u_upload_mgr.h"
 #include "util/u_math.h"
 #include "util/u_box.h"
 #include "util/u_simple_shaders.h"
+#include "util/u_gen_mipmap.h"
+
+/* CSMT headers */
+#include "nine_queue.h"
+#include "nine_csmt_helper.h"
+#include "os/os_thread.h"
 
 #define DBG_CHANNEL DBG_DEVICE
 
+/* Nine CSMT */
+
+struct csmt_instruction {
+    int (* func)(struct NineDevice9 *This, struct csmt_instruction *instr);
+};
+
+struct csmt_context {
+    pipe_thread worker;
+    struct nine_queue_pool* pool;
+    BOOL terminate;
+    pipe_condvar event_processed;
+    pipe_mutex mutex_processed;
+    struct NineDevice9 *device;
+    BOOL processed;
+    BOOL toPause;
+    BOOL hasPaused;
+    pipe_mutex thread_running;
+    pipe_mutex thread_resume;
+};
+
+/* Wait for instruction to be processed.
+ * Caller has to ensure that only one thread waits at time.
+ */
+static void
+nine_csmt_wait_processed(struct csmt_context *ctx)
+{
+    pipe_mutex_lock(ctx->mutex_processed);
+    while (!p_atomic_read(&ctx->processed)) {
+        pipe_condvar_wait(ctx->event_processed, ctx->mutex_processed);
+    }
+    pipe_mutex_unlock(ctx->mutex_processed);
+}
+
+/* CSMT worker thread */
+static
+PIPE_THREAD_ROUTINE(nine_csmt_worker, arg)
+{
+    struct csmt_context *ctx = arg;
+    struct csmt_instruction *instr;
+    DBG("CSMT worker spawned\n");
+
+    pipe_thread_setname("CSMT-Worker");
+
+    while (1) {
+        nine_queue_wait_flush(ctx->pool);
+        pipe_mutex_lock(ctx->thread_running);
+
+        /* Get instruction. NULL on empty cmdbuf. */
+        while (!p_atomic_read(&ctx->terminate) &&
+               (instr = (struct csmt_instruction *)nine_queue_get(ctx->pool))) {
+
+            /* decode */
+            if (instr->func(ctx->device, instr)) {
+                pipe_mutex_lock(ctx->mutex_processed);
+                p_atomic_set(&ctx->processed, TRUE);
+                pipe_condvar_signal(ctx->event_processed);
+                pipe_mutex_unlock(ctx->mutex_processed);
+            }
+            if (p_atomic_read(&ctx->toPause)) {
+                pipe_mutex_unlock(ctx->thread_running);
+                /* will wait here the thread can be resumed */
+                pipe_mutex_lock(ctx->thread_resume);
+                pipe_mutex_lock(ctx->thread_running);
+                pipe_mutex_unlock(ctx->thread_resume);
+            }
+        }
+
+        pipe_mutex_unlock(ctx->thread_running);
+        if (p_atomic_read(&ctx->terminate)) {
+            pipe_mutex_lock(ctx->mutex_processed);
+            p_atomic_set(&ctx->processed, TRUE);
+            pipe_condvar_signal(ctx->event_processed);
+            pipe_mutex_unlock(ctx->mutex_processed);
+            break;
+        }
+    }
+
+    DBG("CSMT worker destroyed\n");
+    return 0;
+}
+
+/* Create a CSMT context.
+ * Spawns a worker thread.
+ */
+struct csmt_context *
+nine_csmt_create( struct NineDevice9 *This )
+{
+    struct csmt_context *ctx;
+
+    ctx = CALLOC_STRUCT(csmt_context);
+    if (!ctx)
+        return NULL;
+
+    ctx->pool = nine_queue_create();
+    if (!ctx->pool) {
+        FREE(ctx);
+        return NULL;
+    }
+    cnd_init(&ctx->event_processed);
+    pipe_mutex_init(ctx->mutex_processed);
+    pipe_mutex_init(ctx->thread_running);
+    pipe_mutex_init(ctx->thread_resume);
+
+#if DEBUG
+    pipe_thread_setname("Main thread");
+#endif
+
+    ctx->device = This;
+
+    ctx->worker = pipe_thread_create(nine_csmt_worker, ctx);
+    if (!ctx->worker) {
+        nine_queue_delete(ctx->pool);
+        FREE(ctx);
+        return NULL;
+    }
+
+    DBG("Returning context %p\n", ctx);
+
+    return ctx;
+}
+
+static int
+nop_func( struct NineDevice9 *This, struct csmt_instruction *instr )
+{
+    (void) This;
+    (void) instr;
+
+    return 1;
+}
+
+/* Push nop instruction and flush the queue.
+ * Waits for the worker to complete. */
+void
+nine_csmt_process( struct NineDevice9 *device )
+{
+    struct csmt_instruction* instr;
+    struct csmt_context *ctx = device->csmt_ctx;
+
+    if (!device->csmt_active)
+        return;
+
+    if (nine_queue_isempty(ctx->pool))
+        return;
+
+    DBG("device=%p\n", device);
+
+    /* NOP */
+    instr = nine_queue_alloc(ctx->pool, sizeof(struct csmt_instruction));
+    assert(instr);
+    instr->func = nop_func;
+
+    p_atomic_set(&ctx->processed, FALSE);
+    nine_queue_flush(ctx->pool);
+
+    nine_csmt_wait_processed(ctx);
+}
+
+/* Destroys a CSMT context.
+ * Waits for the worker thread to terminate.
+ */
+void
+nine_csmt_destroy( struct NineDevice9 *device, struct csmt_context *ctx )
+{
+    struct csmt_instruction* instr;
+    pipe_thread render_thread = ctx->worker;
+
+    DBG("device=%p ctx=%p\n", device, ctx);
+
+    /* Push nop and flush the queue. */
+    instr = nine_queue_alloc(ctx->pool, sizeof(struct csmt_instruction));
+    assert(instr);
+    instr->func = nop_func;
+
+    p_atomic_set(&ctx->processed, FALSE);
+    /* Signal worker to terminate. */
+    p_atomic_set(&ctx->terminate, TRUE);
+    nine_queue_flush(ctx->pool);
+
+    nine_csmt_wait_processed(ctx);
+    nine_queue_delete(ctx->pool);
+    pipe_mutex_destroy(ctx->mutex_processed);
+
+    FREE(ctx);
+
+    pipe_thread_wait(render_thread);
+}
+
+static void
+nine_csmt_pause( struct NineDevice9 *device )
+{
+    struct csmt_context *ctx = device->csmt_ctx;
+
+    if (!device->csmt_active)
+        return;
+
+    /* No need to pause the thread */
+    if (nine_queue_no_flushed_work(ctx->pool))
+        return;
+
+    pipe_mutex_lock(ctx->thread_resume);
+    p_atomic_set(&ctx->toPause, TRUE);
+
+    /* Wait the thread is paused */
+    pipe_mutex_lock(ctx->thread_running);
+    ctx->hasPaused = TRUE;
+    p_atomic_set(&ctx->toPause, FALSE);
+}
+
+static void
+nine_csmt_resume( struct NineDevice9 *device )
+{
+    struct csmt_context *ctx = device->csmt_ctx;
+
+    if (!device->csmt_active)
+        return;
+
+    if (!ctx->hasPaused)
+        return;
+
+    ctx->hasPaused = FALSE;
+    pipe_mutex_unlock(ctx->thread_running);
+    pipe_mutex_unlock(ctx->thread_resume);
+}
+
+struct pipe_context *
+nine_context_get_pipe( struct NineDevice9 *device )
+{
+    nine_csmt_process(device);
+    return device->context.pipe;
+}
+
+struct pipe_context *
+nine_context_get_pipe_multithread( struct NineDevice9 *device )
+{
+    struct csmt_context *ctx = device->csmt_ctx;
+
+    if (!device->csmt_active)
+        return device->context.pipe;
+
+    if (!pipe_thread_is_self(ctx->worker))
+        nine_csmt_process(device);
+
+    return device->context.pipe;
+}
+
+struct pipe_context *
+nine_context_get_pipe_acquire( struct NineDevice9 *device )
+{
+    nine_csmt_pause(device);
+    return device->context.pipe;
+}
+
+void
+nine_context_get_pipe_release( struct NineDevice9 *device )
+{
+    nine_csmt_resume(device);
+}
+
+/* Nine state functions */
+
 /* Check if some states need to be set dirty */
 
 static inline DWORD
@@ -154,47 +424,47 @@ prepare_vs_constants_userbuf_swvp(struct NineDevice9 *device)
 
     if (!device->driver_caps.user_cbufs) {
         struct pipe_constant_buffer *cb = &(context->pipe_data.cb0_swvp);
-        u_upload_data(device->constbuf_uploader,
+        u_upload_data(device->context.pipe->const_uploader,
                       0,
                       cb->buffer_size,
                       device->constbuf_alignment,
                       cb->user_buffer,
                       &(cb->buffer_offset),
                       &(cb->buffer));
-        u_upload_unmap(device->constbuf_uploader);
+        u_upload_unmap(device->context.pipe->const_uploader);
         cb->user_buffer = NULL;
 
         cb = &(context->pipe_data.cb1_swvp);
-        u_upload_data(device->constbuf_uploader,
+        u_upload_data(device->context.pipe->const_uploader,
                       0,
                       cb->buffer_size,
                       device->constbuf_alignment,
                       cb->user_buffer,
                       &(cb->buffer_offset),
                       &(cb->buffer));
-        u_upload_unmap(device->constbuf_uploader);
+        u_upload_unmap(device->context.pipe->const_uploader);
         cb->user_buffer = NULL;
 
         cb = &(context->pipe_data.cb2_swvp);
-        u_upload_data(device->constbuf_uploader,
+        u_upload_data(device->context.pipe->const_uploader,
                       0,
                       cb->buffer_size,
                       device->constbuf_alignment,
                       cb->user_buffer,
                       &(cb->buffer_offset),
                       &(cb->buffer));
-        u_upload_unmap(device->constbuf_uploader);
+        u_upload_unmap(device->context.pipe->const_uploader);
         cb->user_buffer = NULL;
 
         cb = &(context->pipe_data.cb3_swvp);
-        u_upload_data(device->constbuf_uploader,
+        u_upload_data(device->context.pipe->const_uploader,
                       0,
                       cb->buffer_size,
                       device->constbuf_alignment,
                       cb->user_buffer,
                       &(cb->buffer_offset),
                       &(cb->buffer));
-        u_upload_unmap(device->constbuf_uploader);
+        u_upload_unmap(device->context.pipe->const_uploader);
         cb->user_buffer = NULL;
     }
 
@@ -212,7 +482,7 @@ prepare_vs_constants_userbuf(struct NineDevice9 *device)
     cb.buffer_size = context->vs->const_used_size;
     cb.user_buffer = context->vs_const_f;
 
-    if (device->swvp) {
+    if (context->swvp) {
         prepare_vs_constants_userbuf_swvp(device);
         return;
     }
@@ -253,14 +523,14 @@ prepare_vs_constants_userbuf(struct NineDevice9 *device)
 
     if (!device->driver_caps.user_cbufs) {
         context->pipe_data.cb_vs.buffer_size = cb.buffer_size;
-        u_upload_data(device->constbuf_uploader,
+        u_upload_data(device->context.pipe->const_uploader,
                       0,
                       cb.buffer_size,
                       device->constbuf_alignment,
                       cb.user_buffer,
                       &context->pipe_data.cb_vs.buffer_offset,
                       &context->pipe_data.cb_vs.buffer);
-        u_upload_unmap(device->constbuf_uploader);
+        u_upload_unmap(device->context.pipe->const_uploader);
         context->pipe_data.cb_vs.user_buffer = NULL;
     } else
         context->pipe_data.cb_vs = cb;
@@ -324,14 +594,14 @@ prepare_ps_constants_userbuf(struct NineDevice9 *device)
 
     if (!device->driver_caps.user_cbufs) {
         context->pipe_data.cb_ps.buffer_size = cb.buffer_size;
-        u_upload_data(device->constbuf_uploader,
+        u_upload_data(device->context.pipe->const_uploader,
                       0,
                       cb.buffer_size,
                       device->constbuf_alignment,
                       cb.user_buffer,
                       &context->pipe_data.cb_ps.buffer_offset,
                       &context->pipe_data.cb_ps.buffer);
-        u_upload_unmap(device->constbuf_uploader);
+        u_upload_unmap(device->context.pipe->const_uploader);
         context->pipe_data.cb_ps.user_buffer = NULL;
     } else
         context->pipe_data.cb_ps = cb;
@@ -848,7 +1118,7 @@ commit_vs_constants(struct NineDevice9 *device)
     if (unlikely(!context->programmable_vs))
         pipe->set_constant_buffer(pipe, PIPE_SHADER_VERTEX, 0, &context->pipe_data.cb_vs_ff);
     else {
-        if (device->swvp) {
+        if (context->swvp) {
             pipe->set_constant_buffer(pipe, PIPE_SHADER_VERTEX, 0, &context->pipe_data.cb0_swvp);
             pipe->set_constant_buffer(pipe, PIPE_SHADER_VERTEX, 1, &context->pipe_data.cb1_swvp);
             pipe->set_constant_buffer(pipe, PIPE_SHADER_VERTEX, 2, &context->pipe_data.cb2_swvp);
@@ -1094,11 +1364,44 @@ NineDevice9_ResolveZ( struct NineDevice9 *device )
 #define ALPHA_TO_COVERAGE_ENABLE   MAKEFOURCC('A', '2', 'M', '1')
 #define ALPHA_TO_COVERAGE_DISABLE  MAKEFOURCC('A', '2', 'M', '0')
 
+/* Nine_context functions.
+ * Serialized through CSMT macros.
+ */
+
+static void
+nine_context_set_texture_apply(struct NineDevice9 *device,
+                               DWORD stage,
+                               BOOL enabled,
+                               BOOL shadow,
+                               DWORD lod,
+                               D3DRESOURCETYPE type,
+                               uint8_t pstype,
+                               struct pipe_resource *res,
+                               struct pipe_sampler_view *view0,
+                               struct pipe_sampler_view *view1);
+static void
+nine_context_set_stream_source_apply(struct NineDevice9 *device,
+                                    UINT StreamNumber,
+                                    struct pipe_resource *res,
+                                    UINT OffsetInBytes,
+                                    UINT Stride);
 
-void
-nine_context_set_render_state(struct NineDevice9 *device,
-                              D3DRENDERSTATETYPE State,
-                              DWORD Value)
+static void
+nine_context_set_indices_apply(struct NineDevice9 *device,
+                               struct pipe_resource *res,
+                               UINT IndexSize,
+                               UINT OffsetInBytes);
+
+static void
+nine_context_set_pixel_shader_constant_i_transformed(struct NineDevice9 *device,
+                                                     UINT StartRegister,
+                                                     const int *pConstantData,
+                                                     unsigned pConstantData_size,
+                                                     UINT Vector4iCount);
+
+CSMT_ITEM_NO_WAIT(nine_context_set_render_state,
+                  ARG_VAL(D3DRENDERSTATETYPE, State),
+                  ARG_VAL(DWORD, Value))
 {
     struct nine_context *context = &device->context;
 
@@ -1137,48 +1440,69 @@ nine_context_set_render_state(struct NineDevice9 *device,
     context->changed.group |= nine_render_state_group[State];
 }
 
+CSMT_ITEM_NO_WAIT(nine_context_set_texture_apply,
+                  ARG_VAL(DWORD, stage),
+                  ARG_VAL(BOOL, enabled),
+                  ARG_VAL(BOOL, shadow),
+                  ARG_VAL(DWORD, lod),
+                  ARG_VAL(D3DRESOURCETYPE, type),
+                  ARG_VAL(uint8_t, pstype),
+                  ARG_BIND_RES(struct pipe_resource, res),
+                  ARG_BIND_VIEW(struct pipe_sampler_view, view0),
+                  ARG_BIND_VIEW(struct pipe_sampler_view, view1))
+{
+    struct nine_context *context = &device->context;
+
+    context->texture[stage].enabled = enabled;
+    context->samplers_shadow &= ~(1 << stage);
+    context->samplers_shadow |= shadow << stage;
+    context->texture[stage].shadow = shadow;
+    context->texture[stage].lod = lod;
+    context->texture[stage].type = type;
+    context->texture[stage].pstype = pstype;
+    pipe_resource_reference(&context->texture[stage].resource, res);
+    pipe_sampler_view_reference(&context->texture[stage].view[0], view0);
+    pipe_sampler_view_reference(&context->texture[stage].view[1], view1);
+
+    context->changed.group |= NINE_STATE_TEXTURE;
+}
+
 void
 nine_context_set_texture(struct NineDevice9 *device,
                          DWORD Stage,
                          struct NineBaseTexture9 *tex)
 {
-    struct nine_context *context = &device->context;
+    BOOL enabled = FALSE;
+    BOOL shadow = FALSE;
+    DWORD lod = 0;
+    D3DRESOURCETYPE type = D3DRTYPE_TEXTURE;
+    uint8_t pstype = 0;
+    struct pipe_resource *res = NULL;
+    struct pipe_sampler_view *view0 = NULL, *view1 = NULL;
 
-    context->samplers_shadow &= ~(1 << Stage);
     /* For managed pool, the data can be initially incomplete.
      * In that case, the texture is rebound later
      * (in NineBaseTexture9_Validate/NineBaseTexture9_UploadSelf). */
     if (tex && tex->base.resource) {
-        context->samplers_shadow |= tex->shadow << Stage;
-        context->texture[Stage].enabled = TRUE;
-        context->texture[Stage].shadow = tex->shadow;
-        context->texture[Stage].lod = tex->managed.lod;
-        context->texture[Stage].type = tex->base.type;
-        context->texture[Stage].pstype = tex->pstype;
-        pipe_resource_reference(&context->texture[Stage].resource,
-                                tex->base.resource);
-        pipe_sampler_view_reference(&context->texture[Stage].view[0],
-                                    NineBaseTexture9_GetSamplerView(tex, 0));
-        pipe_sampler_view_reference(&context->texture[Stage].view[1],
-                                    NineBaseTexture9_GetSamplerView(tex, 1));
-    } else {
-        context->texture[Stage].enabled = FALSE;
-        pipe_resource_reference(&context->texture[Stage].resource,
-                                NULL);
-        pipe_sampler_view_reference(&context->texture[Stage].view[0],
-                                    NULL);
-        pipe_sampler_view_reference(&context->texture[Stage].view[1],
-                                    NULL);
+        enabled = TRUE;
+        shadow = tex->shadow;
+        lod = tex->managed.lod;
+        type = tex->base.type;
+        pstype = tex->pstype;
+        res = tex->base.resource;
+        view0 = NineBaseTexture9_GetSamplerView(tex, 0);
+        view1 = NineBaseTexture9_GetSamplerView(tex, 1);
     }
 
-    context->changed.group |= NINE_STATE_TEXTURE;
+    nine_context_set_texture_apply(device, Stage, enabled,
+                                   shadow, lod, type, pstype,
+                                   res, view0, view1);
 }
 
-void
-nine_context_set_sampler_state(struct NineDevice9 *device,
-                               DWORD Sampler,
-                               D3DSAMPLERSTATETYPE Type,
-                               DWORD Value)
+CSMT_ITEM_NO_WAIT(nine_context_set_sampler_state,
+                  ARG_VAL(DWORD, Sampler),
+                  ARG_VAL(D3DSAMPLERSTATETYPE, Type),
+                  ARG_VAL(DWORD, Value))
 {
     struct nine_context *context = &device->context;
 
@@ -1190,6 +1514,22 @@ nine_context_set_sampler_state(struct NineDevice9 *device,
     context->changed.sampler[Sampler] |= 1 << Type;
 }
 
+CSMT_ITEM_NO_WAIT(nine_context_set_stream_source_apply,
+                  ARG_VAL(UINT, StreamNumber),
+                  ARG_BIND_RES(struct pipe_resource, res),
+                  ARG_VAL(UINT, OffsetInBytes),
+                  ARG_VAL(UINT, Stride))
+{
+    struct nine_context *context = &device->context;
+    const unsigned i = StreamNumber;
+
+    context->vtxbuf[i].stride = Stride;
+    context->vtxbuf[i].buffer_offset = OffsetInBytes;
+    pipe_resource_reference(&context->vtxbuf[i].buffer, res);
+
+    context->changed.vtxbuf |= 1 << StreamNumber;
+}
+
 void
 nine_context_set_stream_source(struct NineDevice9 *device,
                                UINT StreamNumber,
@@ -1197,23 +1537,22 @@ nine_context_set_stream_source(struct NineDevice9 *device,
                                UINT OffsetInBytes,
                                UINT Stride)
 {
-    struct nine_context *context = &device->context;
-    const unsigned i = StreamNumber;
+    struct pipe_resource *res = NULL;
+    unsigned offset = 0;
 
-    context->changed.vtxbuf |= 1 << StreamNumber;
+    if (pVBuf9)
+        res = NineVertexBuffer9_GetResource(pVBuf9, &offset);
+    /* in the future when there is internal offset, add it
+     * to OffsetInBytes */
 
-    if (pVBuf9) {
-        context->vtxbuf[i].stride = Stride;
-        context->vtxbuf[i].buffer_offset = OffsetInBytes;
-    }
-    pipe_resource_reference(&context->vtxbuf[i].buffer,
-                            pVBuf9 ? NineVertexBuffer9_GetResource(pVBuf9) : NULL);
+    nine_context_set_stream_source_apply(device, StreamNumber,
+                                         res, offset + OffsetInBytes,
+                                         Stride);
 }
 
-void
-nine_context_set_stream_source_freq(struct NineDevice9 *device,
-                                    UINT StreamNumber,
-                                    UINT Setting)
+CSMT_ITEM_NO_WAIT(nine_context_set_stream_source_freq,
+                  ARG_VAL(UINT, StreamNumber),
+                  ARG_VAL(UINT, Setting))
 {
     struct nine_context *context = &device->context;
 
@@ -1228,28 +1567,42 @@ nine_context_set_stream_source_freq(struct NineDevice9 *device,
         context->changed.group |= NINE_STATE_STREAMFREQ;
 }
 
+CSMT_ITEM_NO_WAIT(nine_context_set_indices_apply,
+                  ARG_BIND_RES(struct pipe_resource, res),
+                  ARG_VAL(UINT, IndexSize),
+                  ARG_VAL(UINT, OffsetInBytes))
+{
+    struct nine_context *context = &device->context;
+
+    context->idxbuf.index_size = IndexSize;
+    context->idxbuf.offset = OffsetInBytes;
+    pipe_resource_reference(&context->idxbuf.buffer, res);
+    context->idxbuf.user_buffer = NULL;
+
+    context->changed.group |= NINE_STATE_IDXBUF;
+}
+
 void
 nine_context_set_indices(struct NineDevice9 *device,
                          struct NineIndexBuffer9 *idxbuf)
 {
-    struct nine_context *context = &device->context;
     const struct pipe_index_buffer *pipe_idxbuf;
+    struct pipe_resource *res = NULL;
+    UINT IndexSize = 0;
+    UINT OffsetInBytes = 0;
 
     if (idxbuf) {
         pipe_idxbuf = NineIndexBuffer9_GetBuffer(idxbuf);
-        context->idxbuf.index_size = pipe_idxbuf->index_size;
-        pipe_resource_reference(&context->idxbuf.buffer, pipe_idxbuf->buffer);
-        context->idxbuf.offset = pipe_idxbuf->offset;
-        context->idxbuf.user_buffer = NULL;
-    } else
-        pipe_resource_reference(&context->idxbuf.buffer, NULL);
+        IndexSize = pipe_idxbuf->index_size;
+        res = pipe_idxbuf->buffer;
+        OffsetInBytes = pipe_idxbuf->offset;
+    }
 
-    context->changed.group |= NINE_STATE_IDXBUF;
+    nine_context_set_indices_apply(device, res, IndexSize, OffsetInBytes);
 }
 
-void
-nine_context_set_vertex_declaration(struct NineDevice9 *device,
-                                    struct NineVertexDeclaration9 *vdecl)
+CSMT_ITEM_NO_WAIT(nine_context_set_vertex_declaration,
+                  ARG_BIND_REF(struct NineVertexDeclaration9, vdecl))
 {
     struct nine_context *context = &device->context;
     BOOL was_programmable_vs = context->programmable_vs;
@@ -1265,9 +1618,8 @@ nine_context_set_vertex_declaration(struct NineDevice9 *device,
     context->changed.group |= NINE_STATE_VDECL;
 }
 
-void
-nine_context_set_vertex_shader(struct NineDevice9 *device,
-                               struct NineVertexShader9 *pShader)
+CSMT_ITEM_NO_WAIT(nine_context_set_vertex_shader,
+                  ARG_BIND_REF(struct NineVertexShader9, pShader))
 {
     struct nine_context *context = &device->context;
     BOOL was_programmable_vs = context->programmable_vs;
@@ -1283,18 +1635,18 @@ nine_context_set_vertex_shader(struct NineDevice9 *device,
     context->changed.group |= NINE_STATE_VS;
 }
 
-void
-nine_context_set_vertex_shader_constant_f(struct NineDevice9 *device,
-                                          UINT StartRegister,
-                                          const float *pConstantData,
-                                          UINT Vector4fCount)
+CSMT_ITEM_NO_WAIT(nine_context_set_vertex_shader_constant_f,
+                  ARG_VAL(UINT, StartRegister),
+                  ARG_MEM(float, pConstantData),
+                  ARG_MEM_SIZE(unsigned, pConstantData_size),
+                  ARG_VAL(UINT, Vector4fCount))
 {
     struct nine_context *context = &device->context;
     float *vs_const_f = device->may_swvp ? context->vs_const_f_swvp : context->vs_const_f;
 
     memcpy(&vs_const_f[StartRegister * 4],
            pConstantData,
-           Vector4fCount * 4 * sizeof(context->vs_const_f[0]));
+           pConstantData_size);
 
     if (device->may_swvp) {
         Vector4fCount = MIN2(StartRegister + Vector4fCount, NINE_MAX_CONST_F) - StartRegister;
@@ -1308,12 +1660,11 @@ nine_context_set_vertex_shader_constant_f(struct NineDevice9 *device,
     context->changed.group |= NINE_STATE_VS_CONST;
 }
 
-
-void
-nine_context_set_vertex_shader_constant_i(struct NineDevice9 *device,
-                                          UINT StartRegister,
-                                          const int *pConstantData,
-                                          UINT Vector4iCount)
+CSMT_ITEM_NO_WAIT(nine_context_set_vertex_shader_constant_i,
+                  ARG_VAL(UINT, StartRegister),
+                  ARG_MEM(int, pConstantData),
+                  ARG_MEM_SIZE(unsigned, pConstantData_size),
+                  ARG_VAL(UINT, Vector4iCount))
 {
     struct nine_context *context = &device->context;
     int i;
@@ -1321,7 +1672,7 @@ nine_context_set_vertex_shader_constant_i(struct NineDevice9 *device,
     if (device->driver_caps.vs_integer) {
         memcpy(&context->vs_const_i[4 * StartRegister],
                pConstantData,
-               Vector4iCount * sizeof(int[4]));
+               pConstantData_size);
     } else {
         for (i = 0; i < Vector4iCount; i++) {
             context->vs_const_i[4 * (StartRegister + i)] = fui((float)(pConstantData[4 * i]));
@@ -1335,16 +1686,18 @@ nine_context_set_vertex_shader_constant_i(struct NineDevice9 *device,
     context->changed.group |= NINE_STATE_VS_CONST;
 }
 
-void
-nine_context_set_vertex_shader_constant_b(struct NineDevice9 *device,
-                                          UINT StartRegister,
-                                          const BOOL *pConstantData,
-                                          UINT BoolCount)
+CSMT_ITEM_NO_WAIT(nine_context_set_vertex_shader_constant_b,
+                  ARG_VAL(UINT, StartRegister),
+                  ARG_MEM(BOOL, pConstantData),
+                  ARG_MEM_SIZE(unsigned, pConstantData_size),
+                  ARG_VAL(UINT, BoolCount))
 {
     struct nine_context *context = &device->context;
     int i;
     uint32_t bool_true = device->driver_caps.vs_integer ? 0xFFFFFFFF : fui(1.0f);
 
+    (void) pConstantData_size;
+
     for (i = 0; i < BoolCount; i++)
         context->vs_const_b[StartRegister + i] = pConstantData[i] ? bool_true : 0;
 
@@ -1352,9 +1705,8 @@ nine_context_set_vertex_shader_constant_b(struct NineDevice9 *device,
     context->changed.group |= NINE_STATE_VS_CONST;
 }
 
-void
-nine_context_set_pixel_shader(struct NineDevice9 *device,
-                              struct NinePixelShader9* ps)
+CSMT_ITEM_NO_WAIT(nine_context_set_pixel_shader,
+                  ARG_BIND_REF(struct NinePixelShader9, ps))
 {
     struct nine_context *context = &device->context;
     unsigned old_mask = context->ps ? context->ps->rt_mask : 1;
@@ -1375,27 +1727,44 @@ nine_context_set_pixel_shader(struct NineDevice9 *device,
         context->changed.group |= NINE_STATE_FB;
 }
 
-void
-nine_context_set_pixel_shader_constant_f(struct NineDevice9 *device,
-                                        UINT StartRegister,
-                                        const float *pConstantData,
-                                        UINT Vector4fCount)
+CSMT_ITEM_NO_WAIT(nine_context_set_pixel_shader_constant_f,
+                  ARG_VAL(UINT, StartRegister),
+                  ARG_MEM(float, pConstantData),
+                  ARG_MEM_SIZE(unsigned, pConstantData_size),
+                  ARG_VAL(UINT, Vector4fCount))
 {
     struct nine_context *context = &device->context;
 
     memcpy(&context->ps_const_f[StartRegister * 4],
            pConstantData,
-           Vector4fCount * 4 * sizeof(context->ps_const_f[0]));
+           pConstantData_size);
 
     context->changed.ps_const_f = TRUE;
     context->changed.group |= NINE_STATE_PS_CONST;
 }
 
-void
-nine_context_set_pixel_shader_constant_i(struct NineDevice9 *device,
-                                         UINT StartRegister,
-                                         const int *pConstantData,
-                                         UINT Vector4iCount)
+/* For stateblocks */
+CSMT_ITEM_NO_WAIT(nine_context_set_pixel_shader_constant_i_transformed,
+                  ARG_VAL(UINT, StartRegister),
+                  ARG_MEM(int, pConstantData),
+                  ARG_MEM_SIZE(unsigned, pConstantData_size),
+                  ARG_VAL(UINT, Vector4iCount))
+{
+    struct nine_context *context = &device->context;
+
+    memcpy(&context->ps_const_i[StartRegister][0],
+           pConstantData,
+           Vector4iCount * sizeof(context->ps_const_i[0]));
+
+    context->changed.ps_const_i = TRUE;
+    context->changed.group |= NINE_STATE_PS_CONST;
+}
+
+CSMT_ITEM_NO_WAIT(nine_context_set_pixel_shader_constant_i,
+                  ARG_VAL(UINT, StartRegister),
+                  ARG_MEM(int, pConstantData),
+                  ARG_MEM_SIZE(unsigned, pConstantData_size),
+                  ARG_VAL(UINT, Vector4iCount))
 {
     struct nine_context *context = &device->context;
     int i;
@@ -1403,7 +1772,7 @@ nine_context_set_pixel_shader_constant_i(struct NineDevice9 *device,
     if (device->driver_caps.ps_integer) {
         memcpy(&context->ps_const_i[StartRegister][0],
                pConstantData,
-               Vector4iCount * sizeof(context->ps_const_i[0]));
+               pConstantData_size);
     } else {
         for (i = 0; i < Vector4iCount; i++) {
             context->ps_const_i[StartRegister+i][0] = fui((float)(pConstantData[4*i]));
@@ -1416,16 +1785,18 @@ nine_context_set_pixel_shader_constant_i(struct NineDevice9 *device,
     context->changed.group |= NINE_STATE_PS_CONST;
 }
 
-void
-nine_context_set_pixel_shader_constant_b(struct NineDevice9 *device,
-                                         UINT StartRegister,
-                                         const BOOL *pConstantData,
-                                         UINT BoolCount)
+CSMT_ITEM_NO_WAIT(nine_context_set_pixel_shader_constant_b,
+                  ARG_VAL(UINT, StartRegister),
+                  ARG_MEM(BOOL, pConstantData),
+                  ARG_MEM_SIZE(unsigned, pConstantData_size),
+                  ARG_VAL(UINT, BoolCount))
 {
     struct nine_context *context = &device->context;
     int i;
     uint32_t bool_true = device->driver_caps.ps_integer ? 0xFFFFFFFF : fui(1.0f);
 
+    (void) pConstantData_size;
+
     for (i = 0; i < BoolCount; i++)
         context->ps_const_b[StartRegister + i] = pConstantData[i] ? bool_true : 0;
 
@@ -1433,10 +1804,10 @@ nine_context_set_pixel_shader_constant_b(struct NineDevice9 *device,
     context->changed.group |= NINE_STATE_PS_CONST;
 }
 
-void
-nine_context_set_render_target(struct NineDevice9 *device,
-                               DWORD RenderTargetIndex,
-                               struct NineSurface9 *rt)
+/* XXX: use resource, as resource might change */
+CSMT_ITEM_NO_WAIT(nine_context_set_render_target,
+                  ARG_VAL(DWORD, RenderTargetIndex),
+                  ARG_BIND_REF(struct NineSurface9, rt))
 {
     struct nine_context *context = &device->context;
     const unsigned i = RenderTargetIndex;
@@ -1468,9 +1839,9 @@ nine_context_set_render_target(struct NineDevice9 *device,
     }
 }
 
-void
-nine_context_set_depth_stencil(struct NineDevice9 *device,
-                               struct NineSurface9 *ds)
+/* XXX: use resource instead of ds, as resource might change */
+CSMT_ITEM_NO_WAIT(nine_context_set_depth_stencil,
+                  ARG_BIND_REF(struct NineSurface9, ds))
 {
     struct nine_context *context = &device->context;
 
@@ -1478,9 +1849,8 @@ nine_context_set_depth_stencil(struct NineDevice9 *device,
     context->changed.group |= NINE_STATE_FB;
 }
 
-void
-nine_context_set_viewport(struct NineDevice9 *device,
-                          const D3DVIEWPORT9 *viewport)
+CSMT_ITEM_NO_WAIT(nine_context_set_viewport,
+                  ARG_COPY_REF(D3DVIEWPORT9, viewport))
 {
     struct nine_context *context = &device->context;
 
@@ -1488,9 +1858,8 @@ nine_context_set_viewport(struct NineDevice9 *device,
     context->changed.group |= NINE_STATE_VIEWPORT;
 }
 
-void
-nine_context_set_scissor(struct NineDevice9 *device,
-                         const struct pipe_scissor_state *scissor)
+CSMT_ITEM_NO_WAIT(nine_context_set_scissor,
+                  ARG_COPY_REF(struct pipe_scissor_state, scissor))
 {
     struct nine_context *context = &device->context;
 
@@ -1498,10 +1867,9 @@ nine_context_set_scissor(struct NineDevice9 *device,
     context->changed.group |= NINE_STATE_SCISSOR;
 }
 
-void
-nine_context_set_transform(struct NineDevice9 *device,
-                           D3DTRANSFORMSTATETYPE State,
-                           const D3DMATRIX *pMatrix)
+CSMT_ITEM_NO_WAIT(nine_context_set_transform,
+                  ARG_VAL(D3DTRANSFORMSTATETYPE, State),
+                  ARG_COPY_REF(D3DMATRIX, pMatrix))
 {
     struct nine_context *context = &device->context;
     D3DMATRIX *M = nine_state_access_transform(&context->ff, State, TRUE);
@@ -1511,9 +1879,8 @@ nine_context_set_transform(struct NineDevice9 *device,
     context->changed.group |= NINE_STATE_FF;
 }
 
-void
-nine_context_set_material(struct NineDevice9 *device,
-                          const D3DMATERIAL9 *pMaterial)
+CSMT_ITEM_NO_WAIT(nine_context_set_material,
+                  ARG_COPY_REF(D3DMATERIAL9, pMaterial))
 {
     struct nine_context *context = &device->context;
 
@@ -1521,10 +1888,9 @@ nine_context_set_material(struct NineDevice9 *device,
     context->changed.group |= NINE_STATE_FF_MATERIAL;
 }
 
-void
-nine_context_set_light(struct NineDevice9 *device,
-                       DWORD Index,
-                       const D3DLIGHT9 *pLight)
+CSMT_ITEM_NO_WAIT(nine_context_set_light,
+                  ARG_VAL(DWORD, Index),
+                  ARG_COPY_REF(D3DLIGHT9, pLight))
 {
     struct nine_context *context = &device->context;
 
@@ -1532,21 +1898,35 @@ nine_context_set_light(struct NineDevice9 *device,
     context->changed.group |= NINE_STATE_FF_LIGHTING;
 }
 
-void
-nine_context_light_enable(struct NineDevice9 *device,
-                          DWORD Index,
-                          BOOL Enable)
+
+/* For stateblocks */
+static void
+nine_context_light_enable_stateblock(struct NineDevice9 *device,
+                                     const uint16_t active_light[NINE_MAX_LIGHTS_ACTIVE], /* TODO: use pointer that convey size for csmt */
+                                     unsigned int num_lights_active)
+{
+    struct nine_context *context = &device->context;
+
+    /* TODO: Use CSMT_* to avoid calling nine_csmt_process */
+    nine_csmt_process(device);
+    memcpy(context->ff.active_light, active_light, NINE_MAX_LIGHTS_ACTIVE * sizeof(context->ff.active_light[0]));
+    context->ff.num_lights_active = num_lights_active;
+    context->changed.group |= NINE_STATE_FF_LIGHTING;
+}
+
+CSMT_ITEM_NO_WAIT(nine_context_light_enable,
+                  ARG_VAL(DWORD, Index),
+                  ARG_VAL(BOOL, Enable))
 {
     struct nine_context *context = &device->context;
 
     nine_state_light_enable(&context->ff, &context->changed.group, Index, Enable);
 }
 
-void
-nine_context_set_texture_stage_state(struct NineDevice9 *device,
-                                     DWORD Stage,
-                                     D3DTEXTURESTAGESTATETYPE Type,
-                                     DWORD Value)
+CSMT_ITEM_NO_WAIT(nine_context_set_texture_stage_state,
+                  ARG_VAL(DWORD, Stage),
+                  ARG_VAL(D3DTEXTURESTAGESTATETYPE, Type),
+                  ARG_VAL(DWORD, Value))
 {
     struct nine_context *context = &device->context;
     int bumpmap_index = -1;
@@ -1587,10 +1967,9 @@ nine_context_set_texture_stage_state(struct NineDevice9 *device,
     context->ff.changed.tex_stage[Stage][Type / 32] |= 1 << (Type % 32);
 }
 
-void
-nine_context_set_clip_plane(struct NineDevice9 *device,
-                            DWORD Index,
-                            const float *pPlane)
+CSMT_ITEM_NO_WAIT(nine_context_set_clip_plane,
+                  ARG_VAL(DWORD, Index),
+                  ARG_COPY_REF(struct nine_clipplane, pPlane))
 {
     struct nine_context *context = &device->context;
 
@@ -1598,6 +1977,17 @@ nine_context_set_clip_plane(struct NineDevice9 *device,
     context->changed.ucp = TRUE;
 }
 
+CSMT_ITEM_NO_WAIT(nine_context_set_swvp,
+                  ARG_VAL(boolean, swvp))
+{
+    struct nine_context *context = &device->context;
+
+    context->swvp = swvp;
+    context->changed.group |= NINE_STATE_SWVP;
+}
+
+#if 0
+
 void
 nine_context_apply_stateblock(struct NineDevice9 *device,
                               const struct nine_state *src)
@@ -1651,9 +2041,10 @@ nine_context_apply_stateblock(struct NineDevice9 *device,
         for (i = 0; m; ++i, m >>= 1) {
             if (src->changed.vtxbuf & (1 << i)) {
                 if (src->stream[i]) {
-                    context->vtxbuf[i].buffer_offset = src->vtxbuf[i].buffer_offset;
+                    unsigned offset = 0;
                     pipe_resource_reference(&context->vtxbuf[i].buffer,
-                        src->stream[i] ? NineVertexBuffer9_GetResource(src->stream[i]) : NULL);
+                        src->stream[i] ? NineVertexBuffer9_GetResource(src->stream[i], &offset) : NULL);
+                    context->vtxbuf[i].buffer_offset = src->vtxbuf[i].buffer_offset + offset;
                     context->vtxbuf[i].stride = src->vtxbuf[i].stride;
                 }
             }
@@ -1823,6 +2214,183 @@ nine_context_apply_stateblock(struct NineDevice9 *device,
     }
 }
 
+#endif
+
+/* Do not write to nine_context directly. Slower,
+ * but works with csmt. TODO: write a special csmt version that
+ * would record the list of commands as much as possible,
+ * and use the version above else.
+ */
+void
+nine_context_apply_stateblock(struct NineDevice9 *device,
+                              const struct nine_state *src)
+{
+    int i;
+
+    /* No need to apply src->changed.group, since all calls do
+    * set context->changed.group */
+
+    for (i = 0; i < ARRAY_SIZE(src->changed.rs); ++i) {
+        uint32_t m = src->changed.rs[i];
+        while (m) {
+            const int r = ffs(m) - 1;
+            m &= ~(1 << r);
+            nine_context_set_render_state(device, i * 32 + r, src->rs_advertised[i * 32 + r]);
+        }
+    }
+
+    /* Textures */
+    if (src->changed.texture) {
+        uint32_t m = src->changed.texture;
+        unsigned s;
+
+        for (s = 0; m; ++s, m >>= 1) {
+            struct NineBaseTexture9 *tex = src->texture[s];
+            if (!(m & 1))
+                continue;
+            nine_context_set_texture(device, s, tex);
+        }
+    }
+
+    /* Sampler state */
+    if (src->changed.group & NINE_STATE_SAMPLER) {
+        unsigned s;
+
+        for (s = 0; s < NINE_MAX_SAMPLERS; ++s) {
+            uint32_t m = src->changed.sampler[s];
+            while (m) {
+                const int i = ffs(m) - 1;
+                m &= ~(1 << i);
+                nine_context_set_sampler_state(device, s, i, src->samp_advertised[s][i]);
+            }
+        }
+    }
+
+    /* Vertex buffers */
+    if (src->changed.vtxbuf | src->changed.stream_freq) {
+        uint32_t m = src->changed.vtxbuf | src->changed.stream_freq;
+        for (i = 0; m; ++i, m >>= 1) {
+            if (src->changed.vtxbuf & (1 << i))
+                nine_context_set_stream_source(device, i, src->stream[i], src->vtxbuf[i].buffer_offset, src->vtxbuf[i].stride);
+            if (src->changed.stream_freq & (1 << i))
+                nine_context_set_stream_source_freq(device, i, src->stream_freq[i]);
+        }
+    }
+
+    /* Index buffer */
+    if (src->changed.group & NINE_STATE_IDXBUF)
+        nine_context_set_indices(device, src->idxbuf);
+
+    /* Vertex declaration */
+    if ((src->changed.group & NINE_STATE_VDECL) && src->vdecl)
+        nine_context_set_vertex_declaration(device, src->vdecl);
+
+    /* Vertex shader */
+    if (src->changed.group & NINE_STATE_VS)
+        nine_context_set_vertex_shader(device, src->vs);
+
+    /* Pixel shader */
+    if (src->changed.group & NINE_STATE_PS)
+        nine_context_set_pixel_shader(device, src->ps);
+
+    /* Vertex constants */
+    if (src->changed.group & NINE_STATE_VS_CONST) {
+        struct nine_range *r;
+        for (r = src->changed.vs_const_f; r; r = r->next)
+            nine_context_set_vertex_shader_constant_f(device, r->bgn,
+                                                      &src->vs_const_f[r->bgn * 4],
+                                                      sizeof(float[4]) * (r->end - r->bgn),
+                                                      r->end - r->bgn);
+        for (r = src->changed.vs_const_i; r; r = r->next)
+            nine_context_set_vertex_shader_constant_i(device, r->bgn,
+                                                      &src->vs_const_i[r->bgn * 4],
+                                                      sizeof(int[4]) * (r->end - r->bgn),
+                                                      r->end - r->bgn);
+        for (r = src->changed.vs_const_b; r; r = r->next)
+            nine_context_set_vertex_shader_constant_b(device, r->bgn,
+                                                      &src->vs_const_b[r->bgn * 4],
+                                                      sizeof(BOOL) * (r->end - r->bgn),
+                                                      r->end - r->bgn);
+    }
+
+    /* Pixel constants */
+    if (src->changed.group & NINE_STATE_PS_CONST) {
+        struct nine_range *r;
+        for (r = src->changed.ps_const_f; r; r = r->next)
+            nine_context_set_pixel_shader_constant_f(device, r->bgn,
+                                                     &src->ps_const_f[r->bgn * 4],
+                                                     sizeof(float[4]) * (r->end - r->bgn),
+                                                     r->end - r->bgn);
+        if (src->changed.ps_const_i) {
+            uint16_t m = src->changed.ps_const_i;
+            for (i = ffs(m) - 1, m >>= i; m; ++i, m >>= 1)
+                if (m & 1)
+                    nine_context_set_pixel_shader_constant_i_transformed(device, i,
+                                                                         src->ps_const_i[i], sizeof(int[4]), 1);
+        }
+        if (src->changed.ps_const_b) {
+            uint16_t m = src->changed.ps_const_b;
+            for (i = ffs(m) - 1, m >>= i; m; ++i, m >>= 1)
+                if (m & 1)
+                    nine_context_set_pixel_shader_constant_b(device, i,
+                                                             &src->ps_const_b[i], sizeof(BOOL), 1);
+        }
+    }
+
+    /* Viewport */
+    if (src->changed.group & NINE_STATE_VIEWPORT)
+        nine_context_set_viewport(device, &src->viewport);
+
+    /* Scissor */
+    if (src->changed.group & NINE_STATE_SCISSOR)
+        nine_context_set_scissor(device, &src->scissor);
+
+    /* User Clip Planes */
+    if (src->changed.ucp)
+        for (i = 0; i < PIPE_MAX_CLIP_PLANES; ++i)
+            if (src->changed.ucp & (1 << i))
+                nine_context_set_clip_plane(device, i, (struct nine_clipplane*)&src->clip.ucp[i][0]);
+
+    if (!(src->changed.group & NINE_STATE_FF))
+        return;
+
+    /* Fixed function state. */
+
+    if (src->changed.group & NINE_STATE_FF_MATERIAL)
+        nine_context_set_material(device, &src->ff.material);
+
+    if (src->changed.group & NINE_STATE_FF_PSSTAGES) {
+        unsigned s;
+        for (s = 0; s < NINE_MAX_TEXTURE_STAGES; ++s) {
+            for (i = 0; i < NINED3DTSS_COUNT; ++i)
+                if (src->ff.changed.tex_stage[s][i / 32] & (1 << (i % 32)))
+                   nine_context_set_texture_stage_state(device, s, i, src->ff.tex_stage[s][i]);
+        }
+    }
+    if (src->changed.group & NINE_STATE_FF_LIGHTING) {
+        for (i = 0; i < src->ff.num_lights; ++i)
+            if (src->ff.light[i].Type != NINED3DLIGHT_INVALID)
+                nine_context_set_light(device, i, &src->ff.light[i]);
+
+        nine_context_light_enable_stateblock(device, src->ff.active_light, src->ff.num_lights_active);
+    }
+    if (src->changed.group & NINE_STATE_FF_VSTRANSF) {
+        for (i = 0; i < ARRAY_SIZE(src->ff.changed.transform); ++i) {
+            unsigned s;
+            if (!src->ff.changed.transform[i])
+                continue;
+            for (s = i * 32; s < (i * 32 + 32); ++s) {
+                if (!(src->ff.changed.transform[i] & (1 << (s % 32))))
+                    continue;
+                nine_context_set_transform(device, s,
+                                           nine_state_access_transform(
+                                               (struct nine_ff_state *)&src->ff,
+                                                                       s, FALSE));
+            }
+        }
+    }
+}
+
 static void
 nine_update_state_framebuffer_clear(struct NineDevice9 *device)
 {
@@ -1832,15 +2400,13 @@ nine_update_state_framebuffer_clear(struct NineDevice9 *device)
         update_framebuffer(device, TRUE);
 }
 
-/* Checks were already done before the call */
-void
-nine_context_clear_fb(struct NineDevice9 *device,
-              DWORD Count,
-              const D3DRECT *pRects,
-              DWORD Flags,
-              D3DCOLOR Color,
-              float Z,
-              DWORD Stencil)
+CSMT_ITEM_NO_WAIT(nine_context_clear_fb,
+                  ARG_VAL(DWORD, Count),
+                  ARG_COPY_REF(D3DRECT, pRects),
+                  ARG_VAL(DWORD, Flags),
+                  ARG_VAL(D3DCOLOR, Color),
+                  ARG_VAL(float, Z),
+                  ARG_VAL(DWORD, Stencil))
 {
     struct nine_context *context = &device->context;
     const int sRGB = context->rs[D3DRS_SRGBWRITEENABLE] ? 1 : 0;
@@ -1995,11 +2561,10 @@ init_draw_info(struct pipe_draw_info *info,
     info->indirect_params = NULL;
 }
 
-void
-nine_context_draw_primitive(struct NineDevice9 *device,
-                            D3DPRIMITIVETYPE PrimitiveType,
-                            UINT StartVertex,
-                            UINT PrimitiveCount)
+CSMT_ITEM_NO_WAIT(nine_context_draw_primitive,
+                  ARG_VAL(D3DPRIMITIVETYPE, PrimitiveType),
+                  ARG_VAL(UINT, StartVertex),
+                  ARG_VAL(UINT, PrimitiveCount))
 {
     struct nine_context *context = &device->context;
     struct pipe_draw_info info;
@@ -2016,14 +2581,13 @@ nine_context_draw_primitive(struct NineDevice9 *device,
     context->pipe->draw_vbo(context->pipe, &info);
 }
 
-void
-nine_context_draw_indexed_primitive(struct NineDevice9 *device,
-                                    D3DPRIMITIVETYPE PrimitiveType,
-                                    INT BaseVertexIndex,
-                                    UINT MinVertexIndex,
-                                    UINT NumVertices,
-                                    UINT StartIndex,
-                                    UINT PrimitiveCount)
+CSMT_ITEM_NO_WAIT(nine_context_draw_indexed_primitive,
+                  ARG_VAL(D3DPRIMITIVETYPE, PrimitiveType),
+                   ARG_VAL(INT, BaseVertexIndex),
+                   ARG_VAL(UINT, MinVertexIndex),
+                   ARG_VAL(UINT, NumVertices),
+                   ARG_VAL(UINT, StartIndex),
+                   ARG_VAL(UINT, PrimitiveCount))
 {
     struct nine_context *context = &device->context;
     struct pipe_draw_info info;
@@ -2041,11 +2605,10 @@ nine_context_draw_indexed_primitive(struct NineDevice9 *device,
     context->pipe->draw_vbo(context->pipe, &info);
 }
 
-void
-nine_context_draw_primitive_from_vtxbuf(struct NineDevice9 *device,
-                                        D3DPRIMITIVETYPE PrimitiveType,
-                                        UINT PrimitiveCount,
-                                        struct pipe_vertex_buffer *vtxbuf)
+CSMT_ITEM_NO_WAIT(nine_context_draw_primitive_from_vtxbuf,
+                  ARG_VAL(D3DPRIMITIVETYPE, PrimitiveType),
+                  ARG_VAL(UINT, PrimitiveCount),
+                  ARG_BIND_BUF(struct pipe_vertex_buffer, vtxbuf))
 {
     struct nine_context *context = &device->context;
     struct pipe_draw_info info;
@@ -2064,14 +2627,13 @@ nine_context_draw_primitive_from_vtxbuf(struct NineDevice9 *device,
     context->pipe->draw_vbo(context->pipe, &info);
 }
 
-void
-nine_context_draw_indexed_primitive_from_vtxbuf_idxbuf(struct NineDevice9 *device,
-                                                       D3DPRIMITIVETYPE PrimitiveType,
-                                                       UINT MinVertexIndex,
-                                                       UINT NumVertices,
-                                                       UINT PrimitiveCount,
-                                                       struct pipe_vertex_buffer *vbuf,
-                                                       struct pipe_index_buffer *ibuf)
+CSMT_ITEM_NO_WAIT(nine_context_draw_indexed_primitive_from_vtxbuf_idxbuf,
+                  ARG_VAL(D3DPRIMITIVETYPE, PrimitiveType),
+                  ARG_VAL(UINT, MinVertexIndex),
+                  ARG_VAL(UINT, NumVertices),
+                  ARG_VAL(UINT, PrimitiveCount),
+                  ARG_BIND_BUF(struct pipe_vertex_buffer, vbuf),
+                  ARG_BIND_BUF(struct pipe_index_buffer, ibuf))
 {
     struct nine_context *context = &device->context;
     struct pipe_draw_info info;
@@ -2090,32 +2652,160 @@ nine_context_draw_indexed_primitive_from_vtxbuf_idxbuf(struct NineDevice9 *devic
     context->pipe->draw_vbo(context->pipe, &info);
 }
 
+CSMT_ITEM_NO_WAIT(nine_context_resource_copy_region,
+                  ARG_BIND_REF(struct NineUnknown, dst),
+                  ARG_BIND_REF(struct NineUnknown, src),
+                  ARG_BIND_RES(struct pipe_resource, dst_res),
+                  ARG_VAL(unsigned, dst_level),
+                  ARG_COPY_REF(struct pipe_box, dst_box),
+                  ARG_BIND_RES(struct pipe_resource, src_res),
+                  ARG_VAL(unsigned, src_level),
+                  ARG_COPY_REF(struct pipe_box, src_box))
+{
+    struct nine_context *context = &device->context;
+
+    (void) dst;
+    (void) src;
+
+    context->pipe->resource_copy_region(context->pipe,
+            dst_res, dst_level,
+            dst_box->x, dst_box->y, dst_box->z,
+            src_res, src_level,
+            src_box);
+}
+
+CSMT_ITEM_NO_WAIT(nine_context_blit,
+                  ARG_BIND_REF(struct NineUnknown, dst),
+                  ARG_BIND_REF(struct NineUnknown, src),
+                  ARG_BIND_BLIT(struct pipe_blit_info, blit))
+{
+    struct nine_context *context = &device->context;
+
+    (void) dst;
+    (void) src;
+
+    context->pipe->blit(context->pipe, blit);
+}
+
+CSMT_ITEM_NO_WAIT(nine_context_clear_render_target,
+                  ARG_BIND_REF(struct NineSurface9, surface),
+                  ARG_VAL(D3DCOLOR, color),
+                  ARG_VAL(UINT, x),
+                  ARG_VAL(UINT, y),
+                  ARG_VAL(UINT, width),
+                  ARG_VAL(UINT, height))
+{
+    struct nine_context *context = &device->context;
+    struct pipe_surface *surf;
+    union pipe_color_union rgba;
+
+    d3dcolor_to_pipe_color_union(&rgba, color);
+    surf = NineSurface9_GetSurface(surface, 0);
+    context->pipe->clear_render_target(context->pipe, surf, &rgba, x, y, width, height, false);
+}
+
+CSMT_ITEM_NO_WAIT(nine_context_gen_mipmap,
+                  ARG_BIND_REF(struct NineUnknown, dst),
+                  ARG_BIND_RES(struct pipe_resource, res),
+                  ARG_VAL(UINT, base_level),
+                  ARG_VAL(UINT, last_level),
+                  ARG_VAL(UINT, first_layer),
+                  ARG_VAL(UINT, last_layer),
+                  ARG_VAL(UINT, filter))
+{
+    struct nine_context *context = &device->context;
+
+    /* We just bind dst for the bind count */
+    (void)dst;
+
+    util_gen_mipmap(context->pipe, res, res->format, base_level,
+                    last_level, first_layer, last_layer, filter);
+}
+
+CSMT_ITEM_NO_WAIT_WITH_COUNTER(nine_context_range_upload,
+                               ARG_BIND_RES(struct pipe_resource, res),
+                               ARG_VAL(unsigned, offset),
+                               ARG_VAL(unsigned, size),
+                               ARG_VAL(const void *, data))
+{
+    struct nine_context *context = &device->context;
+
+    context->pipe->buffer_subdata(context->pipe, res, 0, offset, size, data);
+}
+
+CSMT_ITEM_NO_WAIT_WITH_COUNTER(nine_context_box_upload,
+                               ARG_BIND_REF(struct NineUnknown, dst),
+                               ARG_BIND_RES(struct pipe_resource, res),
+                               ARG_VAL(unsigned, level),
+                               ARG_COPY_REF(struct pipe_box, dst_box),
+                               ARG_VAL(enum pipe_format, src_format),
+                               ARG_VAL(const void *, src),
+                               ARG_VAL(unsigned, src_stride),
+                               ARG_VAL(unsigned, src_layer_stride),
+                               ARG_COPY_REF(struct pipe_box, src_box))
+{
+    struct nine_context *context = &device->context;
+    struct pipe_context *pipe = context->pipe;
+    struct pipe_transfer *transfer = NULL;
+    uint8_t *map;
+
+    /* We just bind dst for the bind count */
+    (void)dst;
+
+    map = pipe->transfer_map(pipe,
+                             res,
+                             level,
+                             PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD_RANGE,
+                             dst_box, &transfer);
+    if (!map)
+        return;
+
+    /* Note: if formats are the sames, it will revert
+     * to normal memcpy */
+    (void) util_format_translate_3d(res->format,
+                                    map, transfer->stride,
+                                    transfer->layer_stride,
+                                    0, 0, 0,
+                                    src_format,
+                                    src, src_stride,
+                                    src_layer_stride,
+                                    src_box->x, src_box->y, src_box->z,
+                                    dst_box->width, dst_box->height,
+                                    dst_box->depth);
+
+    pipe_transfer_unmap(pipe, transfer);
+}
+
 struct pipe_query *
 nine_context_create_query(struct NineDevice9 *device, unsigned query_type)
 {
-    struct nine_context *context = &device->context;
+    struct pipe_context *pipe;
+    struct pipe_query *res;
 
-    return context->pipe->create_query(context->pipe, query_type, 0);
+    pipe = nine_context_get_pipe_acquire(device);
+    res = pipe->create_query(pipe, query_type, 0);
+    nine_context_get_pipe_release(device);
+    return res;
 }
 
-void
-nine_context_destroy_query(struct NineDevice9 *device, struct pipe_query *query)
+CSMT_ITEM_DO_WAIT(nine_context_destroy_query,
+                  ARG_REF(struct pipe_query, query))
 {
     struct nine_context *context = &device->context;
 
     context->pipe->destroy_query(context->pipe, query);
 }
 
-void
-nine_context_begin_query(struct NineDevice9 *device, struct pipe_query *query)
+CSMT_ITEM_NO_WAIT_WITH_COUNTER(nine_context_begin_query,
+                               ARG_REF(struct pipe_query, query))
 {
     struct nine_context *context = &device->context;
 
     (void) context->pipe->begin_query(context->pipe, query);
 }
 
-void
-nine_context_end_query(struct NineDevice9 *device, struct pipe_query *query)
+CSMT_ITEM_NO_WAIT_WITH_COUNTER(nine_context_end_query,
+                               ARG_REF(struct pipe_query, query))
 {
     struct nine_context *context = &device->context;
 
@@ -2124,13 +2814,27 @@ nine_context_end_query(struct NineDevice9 *device, struct pipe_query *query)
 
 boolean
 nine_context_get_query_result(struct NineDevice9 *device, struct pipe_query *query,
-                              boolean flush, boolean wait,
+                              unsigned *counter, boolean flush, boolean wait,
                               union pipe_query_result *result)
 {
-    struct nine_context *context = &device->context;
+    struct pipe_context *pipe;
+    boolean ret;
+
+    if (wait)
+        nine_csmt_process(device);
+    else if (p_atomic_read(counter) > 0) {
+        if (flush && device->csmt_active)
+            nine_queue_flush(device->csmt_ctx->pool);
+        DBG("Pending begin/end. Returning\n");
+        return false;
+    }
 
-    (void) flush;
-    return context->pipe->get_query_result(context->pipe, query, wait, result);
+    pipe = nine_context_get_pipe_acquire(device);
+    ret = pipe->get_query_result(pipe, query, wait, result);
+    nine_context_get_pipe_release(device);
+
+    DBG("Query result %s\n", ret ? "found" : "not yet available");
+    return ret;
 }
 
 /* State defaults */
@@ -2291,6 +2995,9 @@ static const DWORD nine_samp_state_defaults[NINED3DSAMP_LAST + 1] =
     [NINED3DSAMP_CUBETEX] = 0
 };
 
+/* Note: The following 4 functions assume there is no
+ * pending commands */
+
 void nine_state_restore_non_cso(struct NineDevice9 *device)
 {
     struct nine_context *context = &device->context;
@@ -2557,7 +3264,7 @@ update_vertex_elements_sw(struct NineDevice9 *device)
 static void
 update_vertex_buffers_sw(struct NineDevice9 *device, int start_vertice, int num_vertices)
 {
-    struct pipe_context *pipe = NineDevice9_GetPipe(device);
+    struct pipe_context *pipe = nine_context_get_pipe_acquire(device);
     struct pipe_context *pipe_sw = device->pipe_sw;
     struct nine_state *state = &device->state;
     struct nine_state_sw_internal *sw_internal = &device->state_sw_internal;
@@ -2572,30 +3279,31 @@ update_vertex_buffers_sw(struct NineDevice9 *device, int start_vertice, int num_
     for (i = 0; mask; mask >>= 1, ++i) {
         if (mask & 1) {
             if (state->stream[i]) {
+                unsigned offset;
                 struct pipe_resource *buf;
                 struct pipe_box box;
 
                 vtxbuf = state->vtxbuf[i];
-                vtxbuf.buffer = NineVertexBuffer9_GetResource(state->stream[i]);
+                vtxbuf.buffer = NineVertexBuffer9_GetResource(state->stream[i], &offset);
 
                 DBG("Locking %p (offset %d, length %d)\n", vtxbuf.buffer,
                     vtxbuf.buffer_offset, num_vertices * vtxbuf.stride);
 
-                u_box_1d(vtxbuf.buffer_offset + start_vertice * vtxbuf.stride,
+                u_box_1d(vtxbuf.buffer_offset + offset + start_vertice * vtxbuf.stride,
                          num_vertices * vtxbuf.stride, &box);
                 buf = vtxbuf.buffer;
                 vtxbuf.user_buffer = pipe->transfer_map(pipe, buf, 0, PIPE_TRANSFER_READ, &box,
                                                         &(sw_internal->transfers_so[i]));
                 vtxbuf.buffer = NULL;
                 if (!device->driver_caps.user_sw_vbufs) {
-                    u_upload_data(device->vertex_sw_uploader,
+                    u_upload_data(device->pipe_sw->stream_uploader,
                                   0,
                                   box.width,
                                   16,
                                   vtxbuf.user_buffer,
                                   &(vtxbuf.buffer_offset),
                                   &(vtxbuf.buffer));
-                    u_upload_unmap(device->vertex_sw_uploader);
+                    u_upload_unmap(device->pipe_sw->stream_uploader);
                     vtxbuf.user_buffer = NULL;
                 }
                 pipe_sw->set_vertex_buffers(pipe_sw, i, 1, &vtxbuf);
@@ -2605,6 +3313,7 @@ update_vertex_buffers_sw(struct NineDevice9 *device, int start_vertice, int num_
                 pipe_sw->set_vertex_buffers(pipe_sw, i, 1, NULL);
         }
     }
+    nine_context_get_pipe_release(device);
 }
 
 static void
@@ -2643,14 +3352,14 @@ update_vs_constants_sw(struct NineDevice9 *device)
 
         buf = cb.user_buffer;
         if (!device->driver_caps.user_sw_cbufs) {
-            u_upload_data(device->constbuf_sw_uploader,
+            u_upload_data(device->pipe_sw->const_uploader,
                           0,
                           cb.buffer_size,
                           16,
                           cb.user_buffer,
                           &(cb.buffer_offset),
                           &(cb.buffer));
-            u_upload_unmap(device->constbuf_sw_uploader);
+            u_upload_unmap(device->pipe_sw->const_uploader);
             cb.user_buffer = NULL;
         }
 
@@ -2660,14 +3369,14 @@ update_vs_constants_sw(struct NineDevice9 *device)
 
         cb.user_buffer = (char *)buf + 4096 * sizeof(float[4]);
         if (!device->driver_caps.user_sw_cbufs) {
-            u_upload_data(device->constbuf_sw_uploader,
+            u_upload_data(device->pipe_sw->const_uploader,
                           0,
                           cb.buffer_size,
                           16,
                           cb.user_buffer,
                           &(cb.buffer_offset),
                           &(cb.buffer));
-            u_upload_unmap(device->constbuf_sw_uploader);
+            u_upload_unmap(device->pipe_sw->const_uploader);
             cb.user_buffer = NULL;
         }
 
@@ -2685,14 +3394,14 @@ update_vs_constants_sw(struct NineDevice9 *device)
         cb.user_buffer = state->vs_const_i;
 
         if (!device->driver_caps.user_sw_cbufs) {
-            u_upload_data(device->constbuf_sw_uploader,
+            u_upload_data(device->pipe_sw->const_uploader,
                           0,
                           cb.buffer_size,
                           16,
                           cb.user_buffer,
                           &(cb.buffer_offset),
                           &(cb.buffer));
-            u_upload_unmap(device->constbuf_sw_uploader);
+            u_upload_unmap(device->pipe_sw->const_uploader);
             cb.user_buffer = NULL;
         }
 
@@ -2710,14 +3419,14 @@ update_vs_constants_sw(struct NineDevice9 *device)
         cb.user_buffer = state->vs_const_b;
 
         if (!device->driver_caps.user_sw_cbufs) {
-            u_upload_data(device->constbuf_sw_uploader,
+            u_upload_data(device->pipe_sw->const_uploader,
                           0,
                           cb.buffer_size,
                           16,
                           cb.user_buffer,
                           &(cb.buffer_offset),
                           &(cb.buffer));
-            u_upload_unmap(device->constbuf_sw_uploader);
+            u_upload_unmap(device->pipe_sw->const_uploader);
             cb.user_buffer = NULL;
         }
 
@@ -2741,14 +3450,14 @@ update_vs_constants_sw(struct NineDevice9 *device)
         cb.user_buffer = viewport_data;
 
         {
-            u_upload_data(device->constbuf_sw_uploader,
+            u_upload_data(device->pipe_sw->const_uploader,
                           0,
                           cb.buffer_size,
                           16,
                           cb.user_buffer,
                           &(cb.buffer_offset),
                           &(cb.buffer));
-            u_upload_unmap(device->constbuf_sw_uploader);
+            u_upload_unmap(device->pipe_sw->const_uploader);
             cb.user_buffer = NULL;
         }
 
@@ -2782,7 +3491,7 @@ void
 nine_state_after_draw_sw(struct NineDevice9 *device)
 {
     struct nine_state_sw_internal *sw_internal = &device->state_sw_internal;
-    struct pipe_context *pipe = NineDevice9_GetPipe(device);
+    struct pipe_context *pipe = nine_context_get_pipe_acquire(device);
     struct pipe_context *pipe_sw = device->pipe_sw;
     int i;
 
@@ -2792,6 +3501,7 @@ nine_state_after_draw_sw(struct NineDevice9 *device)
             pipe->transfer_unmap(pipe, sw_internal->transfers_so[i]);
         sw_internal->transfers_so[i] = NULL;
     }
+    nine_context_get_pipe_release(device);
 }
 
 void