st/nine: Back RT to nine_context
[mesa.git] / src / gallium / state_trackers / nine / nine_state.c
index 2faca121fe6f950b9c6f1974890787bf50174142..4b5b199e560a05ce3f3a6a977ed078c9698f9354 100644 (file)
 #include "buffer9.h"
 #include "indexbuffer9.h"
 #include "surface9.h"
+#include "vertexbuffer9.h"
 #include "vertexdeclaration9.h"
 #include "vertexshader9.h"
 #include "pixelshader9.h"
 #include "nine_pipe.h"
 #include "nine_ff.h"
+#include "nine_limits.h"
 #include "pipe/p_context.h"
 #include "pipe/p_state.h"
 #include "cso_cache/cso_context.h"
 #include "util/u_upload_mgr.h"
 #include "util/u_math.h"
+#include "util/u_box.h"
+#include "util/u_simple_shaders.h"
 
 #define DBG_CHANNEL DBG_DEVICE
 
@@ -44,9 +48,9 @@
 static inline DWORD
 check_multisample(struct NineDevice9 *device)
 {
-    DWORD *rs = device->state.rs;
+    DWORD *rs = device->context.rs;
     DWORD new_value = (rs[D3DRS_ZENABLE] || rs[D3DRS_STENCILENABLE]) &&
-                      device->state.rt[0]->desc.MultiSampleType >= 1 &&
+                      device->context.rt[0]->desc.MultiSampleType >= 1 &&
                       rs[D3DRS_MULTISAMPLEANTIALIAS];
     if (rs[NINED3DRS_MULTISAMPLE] != new_value) {
         rs[NINED3DRS_MULTISAMPLE] = new_value;
@@ -60,55 +64,183 @@ check_multisample(struct NineDevice9 *device)
 static inline void
 prepare_blend(struct NineDevice9 *device)
 {
-    nine_convert_blend_state(&device->state.pipe.blend, device->state.rs);
-    device->state.commit |= NINE_STATE_COMMIT_BLEND;
+    nine_convert_blend_state(&device->context.pipe.blend, device->context.rs);
+    device->context.commit |= NINE_STATE_COMMIT_BLEND;
 }
 
 static inline void
 prepare_dsa(struct NineDevice9 *device)
 {
-    nine_convert_dsa_state(&device->state.pipe.dsa, device->state.rs);
-    device->state.commit |= NINE_STATE_COMMIT_DSA;
+    nine_convert_dsa_state(&device->context.pipe.dsa, device->context.rs);
+    device->context.commit |= NINE_STATE_COMMIT_DSA;
 }
 
 static inline void
 prepare_rasterizer(struct NineDevice9 *device)
 {
-    nine_convert_rasterizer_state(device, &device->state.pipe.rast, device->state.rs);
-    device->state.commit |= NINE_STATE_COMMIT_RASTERIZER;
+    nine_convert_rasterizer_state(device, &device->context.pipe.rast, device->context.rs);
+    device->context.commit |= NINE_STATE_COMMIT_RASTERIZER;
+}
+
+static void
+prepare_vs_constants_userbuf_swvp(struct NineDevice9 *device)
+{
+    struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
+
+    if (context->changed.vs_const_f || state->changed.group & NINE_STATE_SWVP) {
+        struct pipe_constant_buffer cb;
+
+        cb.buffer_offset = 0;
+        cb.buffer_size = 4096 * sizeof(float[4]);
+        cb.user_buffer = context->vs_const_f_swvp;
+
+        if (context->vs->lconstf.ranges) {
+            const struct nine_lconstf *lconstf = &(context->vs->lconstf);
+            const struct nine_range *r = lconstf->ranges;
+            unsigned n = 0;
+            float *dst = context->vs_lconstf_temp;
+            float *src = (float *)cb.user_buffer;
+            memcpy(dst, src, cb.buffer_size);
+            while (r) {
+                unsigned p = r->bgn;
+                unsigned c = r->end - r->bgn;
+                memcpy(&dst[p * 4], &lconstf->data[n * 4], c * 4 * sizeof(float));
+                n += c;
+                r = r->next;
+            }
+            cb.user_buffer = dst;
+        }
+
+        /* Do not erase the buffer field.
+         * It is either NULL (user_cbufs), or a resource.
+         * u_upload_data will do the proper refcount */
+        context->pipe.cb0_swvp.buffer_offset = cb.buffer_offset;
+        context->pipe.cb0_swvp.buffer_size = cb.buffer_size;
+        context->pipe.cb0_swvp.user_buffer = cb.user_buffer;
+
+        cb.user_buffer = (char *)cb.user_buffer + 4096 * sizeof(float[4]);
+        context->pipe.cb1_swvp.buffer_offset = cb.buffer_offset;
+        context->pipe.cb1_swvp.buffer_size = cb.buffer_size;
+        context->pipe.cb1_swvp.user_buffer = cb.user_buffer;
+
+        context->changed.vs_const_f = 0;
+    }
+
+    if (context->changed.vs_const_i || state->changed.group & NINE_STATE_SWVP) {
+        struct pipe_constant_buffer cb;
+
+        cb.buffer_offset = 0;
+        cb.buffer_size = 2048 * sizeof(float[4]);
+        cb.user_buffer = context->vs_const_i;
+
+        context->pipe.cb2_swvp.buffer_offset = cb.buffer_offset;
+        context->pipe.cb2_swvp.buffer_size = cb.buffer_size;
+        context->pipe.cb2_swvp.user_buffer = cb.user_buffer;
+        context->changed.vs_const_i = 0;
+    }
+
+    if (context->changed.vs_const_b || state->changed.group & NINE_STATE_SWVP) {
+        struct pipe_constant_buffer cb;
+
+        cb.buffer_offset = 0;
+        cb.buffer_size = 512 * sizeof(float[4]);
+        cb.user_buffer = context->vs_const_b;
+
+        context->pipe.cb3_swvp.buffer_offset = cb.buffer_offset;
+        context->pipe.cb3_swvp.buffer_size = cb.buffer_size;
+        context->pipe.cb3_swvp.user_buffer = cb.user_buffer;
+        context->changed.vs_const_b = 0;
+    }
+
+    if (!device->driver_caps.user_cbufs) {
+        struct pipe_constant_buffer *cb = &(context->pipe.cb0_swvp);
+        u_upload_data(device->constbuf_uploader,
+                      0,
+                      cb->buffer_size,
+                      device->constbuf_alignment,
+                      cb->user_buffer,
+                      &(cb->buffer_offset),
+                      &(cb->buffer));
+        u_upload_unmap(device->constbuf_uploader);
+        cb->user_buffer = NULL;
+
+        cb = &(context->pipe.cb1_swvp);
+        u_upload_data(device->constbuf_uploader,
+                      0,
+                      cb->buffer_size,
+                      device->constbuf_alignment,
+                      cb->user_buffer,
+                      &(cb->buffer_offset),
+                      &(cb->buffer));
+        u_upload_unmap(device->constbuf_uploader);
+        cb->user_buffer = NULL;
+
+        cb = &(context->pipe.cb2_swvp);
+        u_upload_data(device->constbuf_uploader,
+                      0,
+                      cb->buffer_size,
+                      device->constbuf_alignment,
+                      cb->user_buffer,
+                      &(cb->buffer_offset),
+                      &(cb->buffer));
+        u_upload_unmap(device->constbuf_uploader);
+        cb->user_buffer = NULL;
+
+        cb = &(context->pipe.cb3_swvp);
+        u_upload_data(device->constbuf_uploader,
+                      0,
+                      cb->buffer_size,
+                      device->constbuf_alignment,
+                      cb->user_buffer,
+                      &(cb->buffer_offset),
+                      &(cb->buffer));
+        u_upload_unmap(device->constbuf_uploader);
+        cb->user_buffer = NULL;
+    }
+
+    state->changed.group &= ~NINE_STATE_VS_CONST;
+    context->commit |= NINE_STATE_COMMIT_CONST_VS;
 }
 
 static void
 prepare_vs_constants_userbuf(struct NineDevice9 *device)
 {
     struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
     struct pipe_constant_buffer cb;
     cb.buffer = NULL;
     cb.buffer_offset = 0;
-    cb.buffer_size = device->state.vs->const_used_size;
-    cb.user_buffer = device->state.vs_const_f;
+    cb.buffer_size = context->vs->const_used_size;
+    cb.user_buffer = context->vs_const_f;
 
-    if (!cb.buffer_size)
+    if (device->swvp) {
+        prepare_vs_constants_userbuf_swvp(device);
         return;
+    }
 
-    if (state->changed.vs_const_i) {
-        int *idst = (int *)&state->vs_const_f[4 * device->max_vs_const_f];
-        memcpy(idst, state->vs_const_i, NINE_MAX_CONST_I * sizeof(int[4]));
-        state->changed.vs_const_i = 0;
+    if (context->changed.vs_const_i || state->changed.group & NINE_STATE_SWVP) {
+        int *idst = (int *)&context->vs_const_f[4 * device->max_vs_const_f];
+        memcpy(idst, context->vs_const_i, NINE_MAX_CONST_I * sizeof(int[4]));
+        context->changed.vs_const_i = 0;
     }
-    if (state->changed.vs_const_b) {
-        int *idst = (int *)&state->vs_const_f[4 * device->max_vs_const_f];
+
+    if (context->changed.vs_const_b || state->changed.group & NINE_STATE_SWVP) {
+        int *idst = (int *)&context->vs_const_f[4 * device->max_vs_const_f];
         uint32_t *bdst = (uint32_t *)&idst[4 * NINE_MAX_CONST_I];
-        memcpy(bdst, state->vs_const_b, NINE_MAX_CONST_B * sizeof(BOOL));
-        state->changed.vs_const_b = 0;
+        memcpy(bdst, context->vs_const_b, NINE_MAX_CONST_B * sizeof(BOOL));
+        context->changed.vs_const_b = 0;
     }
 
-    if (device->state.vs->lconstf.ranges) {
+    if (!cb.buffer_size)
+        return;
+
+    if (context->vs->lconstf.ranges) {
         /* TODO: Can we make it so that we don't have to copy everything ? */
-        const struct nine_lconstf *lconstf =  &device->state.vs->lconstf;
+        const struct nine_lconstf *lconstf =  &(context->vs->lconstf);
         const struct nine_range *r = lconstf->ranges;
         unsigned n = 0;
-        float *dst = device->state.vs_lconstf_temp;
+        float *dst = context->vs_lconstf_temp;
         float *src = (float *)cb.user_buffer;
         memcpy(dst, src, cb.buffer_size);
         while (r) {
@@ -122,93 +254,70 @@ prepare_vs_constants_userbuf(struct NineDevice9 *device)
     }
 
     if (!device->driver_caps.user_cbufs) {
+        context->pipe.cb_vs.buffer_size = cb.buffer_size;
         u_upload_data(device->constbuf_uploader,
                       0,
                       cb.buffer_size,
                       device->constbuf_alignment,
                       cb.user_buffer,
-                      &cb.buffer_offset,
-                      &cb.buffer);
+                      &context->pipe.cb_vs.buffer_offset,
+                      &context->pipe.cb_vs.buffer);
         u_upload_unmap(device->constbuf_uploader);
-        cb.user_buffer = NULL;
-    }
-
-    state->pipe.cb_vs = cb;
+        context->pipe.cb_vs.user_buffer = NULL;
+    } else
+        context->pipe.cb_vs = cb;
 
-    if (device->state.changed.vs_const_f) {
-        struct nine_range *r = device->state.changed.vs_const_f;
-        struct nine_range *p = r;
-        while (p->next)
-            p = p->next;
-        nine_range_pool_put_chain(&device->range_pool, r, p);
-        device->state.changed.vs_const_f = NULL;
-    }
+    context->changed.vs_const_f = 0;
 
-    if (device->state.changed.vs_const_i) {
-        struct nine_range *r = device->state.changed.vs_const_i;
-        struct nine_range *p = r;
-        while (p->next)
-            p = p->next;
-        nine_range_pool_put_chain(&device->range_pool, r, p);
-        device->state.changed.vs_const_i = NULL;
-    }
-
-    if (device->state.changed.vs_const_b) {
-        struct nine_range *r = device->state.changed.vs_const_b;
-        struct nine_range *p = r;
-        while (p->next)
-            p = p->next;
-        nine_range_pool_put_chain(&device->range_pool, r, p);
-        device->state.changed.vs_const_b = NULL;
-    }
     state->changed.group &= ~NINE_STATE_VS_CONST;
-    state->commit |= NINE_STATE_COMMIT_CONST_VS;
+    context->commit |= NINE_STATE_COMMIT_CONST_VS;
 }
 
 static void
 prepare_ps_constants_userbuf(struct NineDevice9 *device)
 {
     struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
     struct pipe_constant_buffer cb;
     cb.buffer = NULL;
     cb.buffer_offset = 0;
     cb.buffer_size = device->state.ps->const_used_size;
-    cb.user_buffer = device->state.ps_const_f;
+    cb.user_buffer = context->ps_const_f;
 
-    if (state->changed.ps_const_i) {
-        int *idst = (int *)&state->ps_const_f[4 * device->max_ps_const_f];
-        memcpy(idst, state->ps_const_i, sizeof(state->ps_const_i));
-        state->changed.ps_const_i = 0;
+    if (context->changed.ps_const_i) {
+        int *idst = (int *)&context->ps_const_f[4 * device->max_ps_const_f];
+        memcpy(idst, context->ps_const_i, sizeof(context->ps_const_i));
+        context->changed.ps_const_i = 0;
     }
-    if (state->changed.ps_const_b) {
-        int *idst = (int *)&state->ps_const_f[4 * device->max_ps_const_f];
+    if (context->changed.ps_const_b) {
+        int *idst = (int *)&context->ps_const_f[4 * device->max_ps_const_f];
         uint32_t *bdst = (uint32_t *)&idst[4 * NINE_MAX_CONST_I];
-        memcpy(bdst, state->ps_const_b, sizeof(state->ps_const_b));
-        state->changed.ps_const_b = 0;
+        memcpy(bdst, context->ps_const_b, sizeof(context->ps_const_b));
+        context->changed.ps_const_b = 0;
     }
 
     /* Upload special constants needed to implement PS1.x instructions like TEXBEM,TEXBEML and BEM */
     if (device->state.ps->bumpenvmat_needed) {
-        memcpy(device->state.ps_lconstf_temp, cb.user_buffer, cb.buffer_size);
-        memcpy(&device->state.ps_lconstf_temp[4 * 8], &device->state.bumpmap_vars, sizeof(device->state.bumpmap_vars));
+        memcpy(context->ps_lconstf_temp, cb.user_buffer, cb.buffer_size);
+        memcpy(&context->ps_lconstf_temp[4 * 8], &device->context.bumpmap_vars, sizeof(device->context.bumpmap_vars));
 
-        cb.user_buffer = device->state.ps_lconstf_temp;
+        cb.user_buffer = context->ps_lconstf_temp;
     }
 
     if (state->ps->byte_code.version < 0x30 &&
-        state->rs[D3DRS_FOGENABLE]) {
-        float *dst = &state->ps_lconstf_temp[4 * 32];
-        if (cb.user_buffer != state->ps_lconstf_temp) {
-            memcpy(state->ps_lconstf_temp, cb.user_buffer, cb.buffer_size);
-            cb.user_buffer = state->ps_lconstf_temp;
+        context->rs[D3DRS_FOGENABLE]) {
+        float *dst = &context->ps_lconstf_temp[4 * 32];
+        if (cb.user_buffer != context->ps_lconstf_temp) {
+            memcpy(context->ps_lconstf_temp, cb.user_buffer, cb.buffer_size);
+            cb.user_buffer = context->ps_lconstf_temp;
         }
 
-        d3dcolor_to_rgba(dst, state->rs[D3DRS_FOGCOLOR]);
-        if (state->rs[D3DRS_FOGTABLEMODE] == D3DFOG_LINEAR) {
-            dst[4] = asfloat(state->rs[D3DRS_FOGEND]);
-            dst[5] = 1.0f / (asfloat(state->rs[D3DRS_FOGEND]) - asfloat(state->rs[D3DRS_FOGSTART]));
-        } else if (state->rs[D3DRS_FOGTABLEMODE] != D3DFOG_NONE) {
-            dst[4] = asfloat(state->rs[D3DRS_FOGDENSITY]);
+        d3dcolor_to_rgba(dst, context->rs[D3DRS_FOGCOLOR]);
+        if (context->rs[D3DRS_FOGTABLEMODE] == D3DFOG_LINEAR) {
+            dst[4] = asfloat(context->rs[D3DRS_FOGEND]);
+            dst[5] = 1.0f / (asfloat(context->rs[D3DRS_FOGEND]) - asfloat(context->rs[D3DRS_FOGSTART]));
+        } else if (context->rs[D3DRS_FOGTABLEMODE] != D3DFOG_NONE) {
+            dst[4] = asfloat(context->rs[D3DRS_FOGDENSITY]);
         }
         cb.buffer_size = 4 * 4 * 34;
     }
@@ -217,63 +326,57 @@ prepare_ps_constants_userbuf(struct NineDevice9 *device)
         return;
 
     if (!device->driver_caps.user_cbufs) {
+        context->pipe.cb_ps.buffer_size = cb.buffer_size;
         u_upload_data(device->constbuf_uploader,
                       0,
                       cb.buffer_size,
                       device->constbuf_alignment,
                       cb.user_buffer,
-                      &cb.buffer_offset,
-                      &cb.buffer);
+                      &context->pipe.cb_ps.buffer_offset,
+                      &context->pipe.cb_ps.buffer);
         u_upload_unmap(device->constbuf_uploader);
-        cb.user_buffer = NULL;
-    }
+        context->pipe.cb_ps.user_buffer = NULL;
+    } else
+        context->pipe.cb_ps = cb;
 
-    state->pipe.cb_ps = cb;
+    context->changed.ps_const_f = 0;
 
-    if (device->state.changed.ps_const_f) {
-        struct nine_range *r = device->state.changed.ps_const_f;
-        struct nine_range *p = r;
-        while (p->next)
-            p = p->next;
-        nine_range_pool_put_chain(&device->range_pool, r, p);
-        device->state.changed.ps_const_f = NULL;
-    }
     state->changed.group &= ~NINE_STATE_PS_CONST;
-    state->commit |= NINE_STATE_COMMIT_CONST_PS;
+    context->commit |= NINE_STATE_COMMIT_CONST_PS;
 }
 
 static inline uint32_t
 prepare_vs(struct NineDevice9 *device, uint8_t shader_changed)
 {
-    struct nine_state *state = &device->state;
-    struct NineVertexShader9 *vs = state->vs;
+    struct nine_context *context = &device->context;
+    struct NineVertexShader9 *vs = context->vs;
     uint32_t changed_group = 0;
     int has_key_changed = 0;
 
-    if (likely(state->programmable_vs))
-        has_key_changed = NineVertexShader9_UpdateKey(vs, state);
+    if (likely(context->programmable_vs))
+        has_key_changed = NineVertexShader9_UpdateKey(vs, device);
 
     if (!shader_changed && !has_key_changed)
         return 0;
 
     /* likely because we dislike FF */
-    if (likely(state->programmable_vs)) {
-        state->cso.vs = NineVertexShader9_GetVariant(vs);
+    if (likely(context->programmable_vs)) {
+        context->cso.vs = NineVertexShader9_GetVariant(vs);
     } else {
         vs = device->ff.vs;
-        state->cso.vs = vs->ff_cso;
+        context->cso.vs = vs->ff_cso;
     }
 
-    if (state->rs[NINED3DRS_VSPOINTSIZE] != vs->point_size) {
-        state->rs[NINED3DRS_VSPOINTSIZE] = vs->point_size;
+    if (context->rs[NINED3DRS_VSPOINTSIZE] != vs->point_size) {
+        context->rs[NINED3DRS_VSPOINTSIZE] = vs->point_size;
         changed_group |= NINE_STATE_RASTERIZER;
     }
 
-    if ((state->bound_samplers_mask_vs & vs->sampler_mask) != vs->sampler_mask)
+    if ((context->bound_samplers_mask_vs & vs->sampler_mask) != vs->sampler_mask)
         /* Bound dummy sampler. */
         changed_group |= NINE_STATE_SAMPLER;
 
-    state->commit |= NINE_STATE_COMMIT_VS;
+    context->commit |= NINE_STATE_COMMIT_VS;
     return changed_group;
 }
 
@@ -281,28 +384,29 @@ static inline uint32_t
 prepare_ps(struct NineDevice9 *device, uint8_t shader_changed)
 {
     struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
     struct NinePixelShader9 *ps = state->ps;
     uint32_t changed_group = 0;
     int has_key_changed = 0;
 
     if (likely(ps))
-        has_key_changed = NinePixelShader9_UpdateKey(ps, state);
+        has_key_changed = NinePixelShader9_UpdateKey(ps, state, context);
 
     if (!shader_changed && !has_key_changed)
         return 0;
 
     if (likely(ps)) {
-        state->cso.ps = NinePixelShader9_GetVariant(ps);
+        context->cso.ps = NinePixelShader9_GetVariant(ps);
     } else {
         ps = device->ff.ps;
-        state->cso.ps = ps->ff_cso;
+        context->cso.ps = ps->ff_cso;
     }
 
-    if ((state->bound_samplers_mask_ps & ps->sampler_mask) != ps->sampler_mask)
+    if ((context->bound_samplers_mask_ps & ps->sampler_mask) != ps->sampler_mask)
         /* Bound dummy sampler. */
         changed_group |= NINE_STATE_SAMPLER;
 
-    state->commit |= NINE_STATE_COMMIT_PS;
+    context->commit |= NINE_STATE_COMMIT_PS;
     return changed_group;
 }
 
@@ -315,19 +419,20 @@ update_framebuffer(struct NineDevice9 *device, bool is_clear)
 {
     struct pipe_context *pipe = device->pipe;
     struct nine_state *state = &device->state;
-    struct pipe_framebuffer_state *fb = &device->state.fb;
+    struct nine_context *context = &device->context;
+    struct pipe_framebuffer_state *fb = &context->pipe.fb;
     unsigned i;
-    struct NineSurface9 *rt0 = state->rt[0];
+    struct NineSurface9 *rt0 = context->rt[0];
     unsigned w = rt0->desc.Width;
     unsigned h = rt0->desc.Height;
-    D3DMULTISAMPLE_TYPE nr_samples = rt0->desc.MultiSampleType;
+    unsigned nr_samples = rt0->base.info.nr_samples;
     unsigned ps_mask = state->ps ? state->ps->rt_mask : 1;
     unsigned mask = is_clear ? 0xf : ps_mask;
-    const int sRGB = state->rs[D3DRS_SRGBWRITEENABLE] ? 1 : 0;
+    const int sRGB = context->rs[D3DRS_SRGBWRITEENABLE] ? 1 : 0;
 
     DBG("\n");
 
-    state->rt_mask = 0x0;
+    context->rt_mask = 0x0;
     fb->nr_cbufs = 0;
 
     /* all render targets must have the same size and the depth buffer must be
@@ -343,17 +448,17 @@ update_framebuffer(struct NineDevice9 *device, bool is_clear)
     if (rt0->desc.Format == D3DFMT_NULL && state->ds) {
         w = state->ds->desc.Width;
         h = state->ds->desc.Height;
-        nr_samples = state->ds->desc.MultiSampleType;
+        nr_samples = state->ds->base.info.nr_samples;
     }
 
     for (i = 0; i < device->caps.NumSimultaneousRTs; ++i) {
-        struct NineSurface9 *rt = state->rt[i];
+        struct NineSurface9 *rt = context->rt[i];
 
         if (rt && rt->desc.Format != D3DFMT_NULL && (mask & (1 << i)) &&
             rt->desc.Width == w && rt->desc.Height == h &&
-            rt->desc.MultiSampleType == nr_samples) {
+            rt->base.info.nr_samples == nr_samples) {
             fb->cbufs[i] = NineSurface9_GetSurface(rt, sRGB);
-            state->rt_mask |= 1 << i;
+            context->rt_mask |= 1 << i;
             fb->nr_cbufs = i + 1;
 
             if (unlikely(rt->desc.Usage & D3DUSAGE_AUTOGENMIPMAP)) {
@@ -371,7 +476,7 @@ update_framebuffer(struct NineDevice9 *device, bool is_clear)
 
     if (state->ds && state->ds->desc.Width >= w &&
         state->ds->desc.Height >= h &&
-        state->ds->desc.MultiSampleType == nr_samples) {
+        state->ds->base.info.nr_samples == nr_samples) {
         fb->zsbuf = NineSurface9_GetSurface(state->ds, 0);
     } else {
         fb->zsbuf = NULL;
@@ -382,7 +487,7 @@ update_framebuffer(struct NineDevice9 *device, bool is_clear)
 
     pipe->set_framebuffer_state(pipe, fb); /* XXX: cso ? */
 
-    if (is_clear && state->rt_mask == ps_mask)
+    if (is_clear && context->rt_mask == ps_mask)
         state->changed.group &= ~NINE_STATE_FB;
 }
 
@@ -438,7 +543,8 @@ static void
 update_vertex_elements(struct NineDevice9 *device)
 {
     struct nine_state *state = &device->state;
-    const struct NineVertexDeclaration9 *vdecl = device->state.vdecl;
+    struct nine_context *context = &device->context;
+    const struct NineVertexDeclaration9 *vdecl = device->context.vdecl;
     const struct NineVertexShader9 *vs;
     unsigned n, b, i;
     int index;
@@ -448,10 +554,10 @@ update_vertex_elements(struct NineDevice9 *device)
     BOOL need_dummy_vbo = FALSE;
     struct pipe_vertex_element ve[PIPE_MAX_ATTRIBS];
 
-    state->stream_usage_mask = 0;
+    context->stream_usage_mask = 0;
     memset(vdecl_index_map, -1, 16);
     memset(used_streams, 0, device->caps.MaxStreams);
-    vs = state->programmable_vs ? device->state.vs : device->ff.vs;
+    vs = context->programmable_vs ? context->vs : device->ff.vs;
 
     if (vdecl) {
         for (n = 0; n < vs->num_inputs; ++n) {
@@ -491,7 +597,7 @@ update_vertex_elements(struct NineDevice9 *device)
         if (index >= 0) {
             ve[n] = vdecl->elems[index];
             b = ve[n].vertex_buffer_index;
-            state->stream_usage_mask |= 1 << b;
+            context->stream_usage_mask |= 1 << b;
             /* XXX wine just uses 1 here: */
             if (state->stream_freq[b] & D3DSTREAMSOURCE_INSTANCEDATA)
                 ve[n].instance_divisor = state->stream_freq[b] & 0x7FFFFF;
@@ -507,77 +613,82 @@ update_vertex_elements(struct NineDevice9 *device)
         }
     }
 
-    if (state->dummy_vbo_bound_at != dummy_vbo_stream) {
-        if (state->dummy_vbo_bound_at >= 0)
-            state->changed.vtxbuf |= 1 << state->dummy_vbo_bound_at;
+    if (context->dummy_vbo_bound_at != dummy_vbo_stream) {
+        if (context->dummy_vbo_bound_at >= 0)
+            context->changed.vtxbuf |= 1 << context->dummy_vbo_bound_at;
         if (dummy_vbo_stream >= 0) {
-            state->changed.vtxbuf |= 1 << dummy_vbo_stream;
-            state->vbo_bound_done = FALSE;
+            context->changed.vtxbuf |= 1 << dummy_vbo_stream;
+            context->vbo_bound_done = FALSE;
         }
-        state->dummy_vbo_bound_at = dummy_vbo_stream;
+        context->dummy_vbo_bound_at = dummy_vbo_stream;
     }
 
     cso_set_vertex_elements(device->cso, vs->num_inputs, ve);
-
-    state->changed.stream_freq = 0;
 }
 
 static void
 update_vertex_buffers(struct NineDevice9 *device)
 {
     struct pipe_context *pipe = device->pipe;
-    struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
     struct pipe_vertex_buffer dummy_vtxbuf;
-    uint32_t mask = state->changed.vtxbuf;
+    uint32_t mask = context->changed.vtxbuf;
     unsigned i;
 
     DBG("mask=%x\n", mask);
 
-    if (state->dummy_vbo_bound_at >= 0) {
-        if (!state->vbo_bound_done) {
+    if (context->dummy_vbo_bound_at >= 0) {
+        if (!context->vbo_bound_done) {
             dummy_vtxbuf.buffer = device->dummy_vbo;
             dummy_vtxbuf.stride = 0;
             dummy_vtxbuf.user_buffer = NULL;
             dummy_vtxbuf.buffer_offset = 0;
-            pipe->set_vertex_buffers(pipe, state->dummy_vbo_bound_at,
+            pipe->set_vertex_buffers(pipe, context->dummy_vbo_bound_at,
                                      1, &dummy_vtxbuf);
-            state->vbo_bound_done = TRUE;
+            context->vbo_bound_done = TRUE;
         }
-        mask &= ~(1 << state->dummy_vbo_bound_at);
+        mask &= ~(1 << context->dummy_vbo_bound_at);
     }
 
     for (i = 0; mask; mask >>= 1, ++i) {
         if (mask & 1) {
-            if (state->vtxbuf[i].buffer)
-                pipe->set_vertex_buffers(pipe, i, 1, &state->vtxbuf[i]);
+            if (context->vtxbuf[i].buffer)
+                pipe->set_vertex_buffers(pipe, i, 1, &context->vtxbuf[i]);
             else
                 pipe->set_vertex_buffers(pipe, i, 1, NULL);
         }
     }
 
-    state->changed.vtxbuf = 0;
+    context->changed.vtxbuf = 0;
 }
 
 static inline boolean
-update_sampler_derived(struct nine_state *state, unsigned s)
+update_sampler_derived(struct nine_context *context, unsigned s)
 {
     boolean changed = FALSE;
 
-    if (state->samp[s][NINED3DSAMP_SHADOW] != state->texture[s]->shadow) {
+    if (context->samp[s][NINED3DSAMP_SHADOW] != context->texture[s]->shadow) {
+        changed = TRUE;
+        context->samp[s][NINED3DSAMP_SHADOW] = context->texture[s]->shadow;
+    }
+
+    if (context->samp[s][NINED3DSAMP_CUBETEX] !=
+        (NineResource9(context->texture[s])->type == D3DRTYPE_CUBETEXTURE)) {
         changed = TRUE;
-        state->samp[s][NINED3DSAMP_SHADOW] = state->texture[s]->shadow;
+        context->samp[s][NINED3DSAMP_CUBETEX] =
+                NineResource9(context->texture[s])->type == D3DRTYPE_CUBETEXTURE;
     }
 
-    if (state->samp[s][D3DSAMP_MIPFILTER] != D3DTEXF_NONE) {
-        int lod = state->samp[s][D3DSAMP_MAXMIPLEVEL] - state->texture[s]->managed.lod;
+    if (context->samp[s][D3DSAMP_MIPFILTER] != D3DTEXF_NONE) {
+        int lod = context->samp[s][D3DSAMP_MAXMIPLEVEL] - context->texture[s]->managed.lod;
         if (lod < 0)
             lod = 0;
-        if (state->samp[s][NINED3DSAMP_MINLOD] != lod) {
+        if (context->samp[s][NINED3DSAMP_MINLOD] != lod) {
             changed = TRUE;
-            state->samp[s][NINED3DSAMP_MINLOD] = lod;
+            context->samp[s][NINED3DSAMP_MINLOD] = lod;
         }
     } else {
-        state->changed.sampler[s] &= ~0x300; /* lod changes irrelevant */
+        context->changed.sampler[s] &= ~0x300; /* lod changes irrelevant */
     }
 
     return changed;
@@ -588,6 +699,7 @@ static void
 update_textures_and_samplers(struct NineDevice9 *device)
 {
     struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
     struct pipe_sampler_view *view[NINE_MAX_SAMPLERS];
     unsigned num_textures;
     unsigned i;
@@ -598,26 +710,26 @@ update_textures_and_samplers(struct NineDevice9 *device)
     /* TODO: Can we reduce iterations here ? */
 
     commit_samplers = FALSE;
-    state->bound_samplers_mask_ps = 0;
+    context->bound_samplers_mask_ps = 0;
     for (num_textures = 0, i = 0; i < NINE_MAX_SAMPLERS_PS; ++i) {
         const unsigned s = NINE_SAMPLER_PS(i);
         int sRGB;
 
-        if (!state->texture[s] && !(sampler_mask & (1 << i))) {
+        if (!context->texture[s] && !(sampler_mask & (1 << i))) {
             view[i] = NULL;
             continue;
         }
 
-        if (state->texture[s]) {
-            sRGB = state->samp[s][D3DSAMP_SRGBTEXTURE] ? 1 : 0;
+        if (context->texture[s]) {
+            sRGB = context->samp[s][D3DSAMP_SRGBTEXTURE] ? 1 : 0;
 
-            view[i] = NineBaseTexture9_GetSamplerView(state->texture[s], sRGB);
+            view[i] = NineBaseTexture9_GetSamplerView(context->texture[s], sRGB);
             num_textures = i + 1;
 
-            if (update_sampler_derived(state, s) || (state->changed.sampler[s] & 0x05fe)) {
-                state->changed.sampler[s] = 0;
+            if (update_sampler_derived(context, s) || (context->changed.sampler[s] & 0x05fe)) {
+                context->changed.sampler[s] = 0;
                 commit_samplers = TRUE;
-                nine_convert_sampler_state(device->cso, s, state->samp[s]);
+                nine_convert_sampler_state(device->cso, s, context->samp[s]);
             }
         } else {
             /* Bind dummy sampler. We do not bind dummy sampler when
@@ -633,10 +745,10 @@ update_textures_and_samplers(struct NineDevice9 *device)
                                s - NINE_SAMPLER_PS(0), &device->dummy_sampler_state);
 
             commit_samplers = TRUE;
-            state->changed.sampler[s] = ~0;
+            context->changed.sampler[s] = ~0;
         }
 
-        state->bound_samplers_mask_ps |= (1 << s);
+        context->bound_samplers_mask_ps |= (1 << s);
     }
 
     cso_set_sampler_views(device->cso, PIPE_SHADER_FRAGMENT, num_textures, view);
@@ -645,27 +757,27 @@ update_textures_and_samplers(struct NineDevice9 *device)
         cso_single_sampler_done(device->cso, PIPE_SHADER_FRAGMENT);
 
     commit_samplers = FALSE;
-    sampler_mask = state->programmable_vs ? state->vs->sampler_mask : 0;
-    state->bound_samplers_mask_vs = 0;
+    sampler_mask = context->programmable_vs ? context->vs->sampler_mask : 0;
+    context->bound_samplers_mask_vs = 0;
     for (num_textures = 0, i = 0; i < NINE_MAX_SAMPLERS_VS; ++i) {
         const unsigned s = NINE_SAMPLER_VS(i);
         int sRGB;
 
-        if (!state->texture[s] && !(sampler_mask & (1 << i))) {
+        if (!context->texture[s] && !(sampler_mask & (1 << i))) {
             view[i] = NULL;
             continue;
         }
 
-        if (state->texture[s]) {
-            sRGB = state->samp[s][D3DSAMP_SRGBTEXTURE] ? 1 : 0;
+        if (context->texture[s]) {
+            sRGB = context->samp[s][D3DSAMP_SRGBTEXTURE] ? 1 : 0;
 
-            view[i] = NineBaseTexture9_GetSamplerView(state->texture[s], sRGB);
+            view[i] = NineBaseTexture9_GetSamplerView(context->texture[s], sRGB);
             num_textures = i + 1;
 
-            if (update_sampler_derived(state, s) || (state->changed.sampler[s] & 0x05fe)) {
-                state->changed.sampler[s] = 0;
+            if (update_sampler_derived(context, s) || (context->changed.sampler[s] & 0x05fe)) {
+                context->changed.sampler[s] = 0;
                 commit_samplers = TRUE;
-                nine_convert_sampler_state(device->cso, s, state->samp[s]);
+                nine_convert_sampler_state(device->cso, s, context->samp[s]);
             }
         } else {
             /* Bind dummy sampler. We do not bind dummy sampler when
@@ -681,18 +793,16 @@ update_textures_and_samplers(struct NineDevice9 *device)
                                s - NINE_SAMPLER_VS(0), &device->dummy_sampler_state);
 
             commit_samplers = TRUE;
-            state->changed.sampler[s] = ~0;
+            context->changed.sampler[s] = ~0;
         }
 
-        state->bound_samplers_mask_vs |= (1 << s);
+        context->bound_samplers_mask_vs |= (1 << s);
     }
 
     cso_set_sampler_views(device->cso, PIPE_SHADER_VERTEX, num_textures, view);
 
     if (commit_samplers)
         cso_single_sampler_done(device->cso, PIPE_SHADER_VERTEX);
-
-    state->changed.texture = 0;
 }
 
 /* State commit only */
@@ -700,13 +810,13 @@ update_textures_and_samplers(struct NineDevice9 *device)
 static inline void
 commit_blend(struct NineDevice9 *device)
 {
-    cso_set_blend(device->cso, &device->state.pipe.blend);
+    cso_set_blend(device->cso, &device->context.pipe.blend);
 }
 
 static inline void
 commit_dsa(struct NineDevice9 *device)
 {
-    cso_set_depth_stencil_alpha(device->cso, &device->state.pipe.dsa);
+    cso_set_depth_stencil_alpha(device->cso, &device->context.pipe.dsa);
 }
 
 static inline void
@@ -720,15 +830,15 @@ commit_scissor(struct NineDevice9 *device)
 static inline void
 commit_rasterizer(struct NineDevice9 *device)
 {
-    cso_set_rasterizer(device->cso, &device->state.pipe.rast);
+    cso_set_rasterizer(device->cso, &device->context.pipe.rast);
 }
 
 static inline void
 commit_index_buffer(struct NineDevice9 *device)
 {
     struct pipe_context *pipe = device->pipe;
-    if (device->state.idxbuf)
-        pipe->set_index_buffer(pipe, &device->state.idxbuf->buffer);
+    if (device->context.idxbuf.buffer)
+        pipe->set_index_buffer(pipe, &device->context.idxbuf);
     else
         pipe->set_index_buffer(pipe, NULL);
 }
@@ -738,10 +848,18 @@ commit_vs_constants(struct NineDevice9 *device)
 {
     struct pipe_context *pipe = device->pipe;
 
-    if (unlikely(!device->state.programmable_vs))
-        pipe->set_constant_buffer(pipe, PIPE_SHADER_VERTEX, 0, &device->state.pipe.cb_vs_ff);
-    else
-        pipe->set_constant_buffer(pipe, PIPE_SHADER_VERTEX, 0, &device->state.pipe.cb_vs);
+    if (unlikely(!device->context.programmable_vs))
+        pipe->set_constant_buffer(pipe, PIPE_SHADER_VERTEX, 0, &device->context.pipe.cb_vs_ff);
+    else {
+        if (device->swvp) {
+            pipe->set_constant_buffer(pipe, PIPE_SHADER_VERTEX, 0, &device->context.pipe.cb0_swvp);
+            pipe->set_constant_buffer(pipe, PIPE_SHADER_VERTEX, 1, &device->context.pipe.cb1_swvp);
+            pipe->set_constant_buffer(pipe, PIPE_SHADER_VERTEX, 2, &device->context.pipe.cb2_swvp);
+            pipe->set_constant_buffer(pipe, PIPE_SHADER_VERTEX, 3, &device->context.pipe.cb3_swvp);
+        } else {
+            pipe->set_constant_buffer(pipe, PIPE_SHADER_VERTEX, 0, &device->context.pipe.cb_vs);
+        }
+    }
 }
 
 static inline void
@@ -750,26 +868,26 @@ commit_ps_constants(struct NineDevice9 *device)
     struct pipe_context *pipe = device->pipe;
 
     if (unlikely(!device->state.ps))
-        pipe->set_constant_buffer(pipe, PIPE_SHADER_FRAGMENT, 0, &device->state.pipe.cb_ps_ff);
+        pipe->set_constant_buffer(pipe, PIPE_SHADER_FRAGMENT, 0, &device->context.pipe.cb_ps_ff);
     else
-        pipe->set_constant_buffer(pipe, PIPE_SHADER_FRAGMENT, 0, &device->state.pipe.cb_ps);
+        pipe->set_constant_buffer(pipe, PIPE_SHADER_FRAGMENT, 0, &device->context.pipe.cb_ps);
 }
 
 static inline void
 commit_vs(struct NineDevice9 *device)
 {
-    struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
 
-    device->pipe->bind_vs_state(device->pipe, state->cso.vs);
+    device->pipe->bind_vs_state(device->pipe, context->cso.vs);
 }
 
 
 static inline void
 commit_ps(struct NineDevice9 *device)
 {
-    struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
 
-    device->pipe->bind_fs_state(device->pipe, state->cso.ps);
+    device->pipe->bind_fs_state(device->pipe, context->cso.ps);
 }
 /* State Update */
 
@@ -777,7 +895,8 @@ commit_ps(struct NineDevice9 *device)
    (NINE_STATE_VS |         \
     NINE_STATE_TEXTURE |    \
     NINE_STATE_FOG_SHADER | \
-    NINE_STATE_POINTSIZE_SHADER)
+    NINE_STATE_POINTSIZE_SHADER | \
+    NINE_STATE_SWVP)
 
 #define NINE_STATE_SHADER_CHANGE_PS \
    (NINE_STATE_PS |         \
@@ -830,22 +949,12 @@ update_managed_buffers(struct NineDevice9 *device)
     }
 }
 
-void
-nine_update_state_framebuffer_clear(struct NineDevice9 *device)
-{
-    struct nine_state *state = &device->state;
-
-    validate_textures(device);
-
-    if (state->changed.group & NINE_STATE_FB)
-        update_framebuffer(device, TRUE);
-}
-
-boolean
+static void
 nine_update_state(struct NineDevice9 *device)
 {
     struct pipe_context *pipe = device->pipe;
     struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
     uint32_t group;
 
     DBG("changed state groups: %x\n", state->changed.group);
@@ -860,7 +969,7 @@ nine_update_state(struct NineDevice9 *device)
     update_managed_buffers(device);
 
     /* ff_update may change VS/PS dirty bits */
-    if (unlikely(!state->programmable_vs || !state->ps))
+    if (unlikely(!context->programmable_vs || !state->ps))
         nine_ff_update(device);
     group = state->changed.group;
 
@@ -886,38 +995,38 @@ nine_update_state(struct NineDevice9 *device)
             commit_index_buffer(device);
     }
 
-    if (likely(group & (NINE_STATE_FREQUENT | NINE_STATE_VS | NINE_STATE_PS))) {
+    if (likely(group & (NINE_STATE_FREQUENT | NINE_STATE_VS | NINE_STATE_PS | NINE_STATE_SWVP))) {
         if (group & NINE_STATE_MULTISAMPLE)
             group |= check_multisample(device);
         if (group & NINE_STATE_RASTERIZER)
             prepare_rasterizer(device);
         if (group & (NINE_STATE_TEXTURE | NINE_STATE_SAMPLER))
             update_textures_and_samplers(device);
-        if ((group & (NINE_STATE_VS_CONST | NINE_STATE_VS)) && state->programmable_vs)
+        if ((group & (NINE_STATE_VS_CONST | NINE_STATE_VS | NINE_STATE_SWVP)) && context->programmable_vs)
             prepare_vs_constants_userbuf(device);
         if ((group & (NINE_STATE_PS_CONST | NINE_STATE_PS)) && state->ps)
             prepare_ps_constants_userbuf(device);
     }
 
-    if (state->changed.vtxbuf)
+    if (context->changed.vtxbuf)
         update_vertex_buffers(device);
 
-    if (state->commit & NINE_STATE_COMMIT_BLEND)
+    if (context->commit & NINE_STATE_COMMIT_BLEND)
         commit_blend(device);
-    if (state->commit & NINE_STATE_COMMIT_DSA)
+    if (context->commit & NINE_STATE_COMMIT_DSA)
         commit_dsa(device);
-    if (state->commit & NINE_STATE_COMMIT_RASTERIZER)
+    if (context->commit & NINE_STATE_COMMIT_RASTERIZER)
         commit_rasterizer(device);
-    if (state->commit & NINE_STATE_COMMIT_CONST_VS)
+    if (context->commit & NINE_STATE_COMMIT_CONST_VS)
         commit_vs_constants(device);
-    if (state->commit & NINE_STATE_COMMIT_CONST_PS)
+    if (context->commit & NINE_STATE_COMMIT_CONST_PS)
         commit_ps_constants(device);
-    if (state->commit & NINE_STATE_COMMIT_VS)
+    if (context->commit & NINE_STATE_COMMIT_VS)
         commit_vs(device);
-    if (state->commit & NINE_STATE_COMMIT_PS)
+    if (context->commit & NINE_STATE_COMMIT_PS)
         commit_ps(device);
 
-    state->commit = 0;
+    context->commit = 0;
 
     if (unlikely(state->changed.ucp)) {
         pipe->set_clip_state(pipe, &state->clip);
@@ -929,15 +1038,19 @@ nine_update_state(struct NineDevice9 *device)
             commit_scissor(device);
         if (group & NINE_STATE_BLEND_COLOR) {
             struct pipe_blend_color color;
-            d3dcolor_to_rgba(&color.color[0], state->rs[D3DRS_BLENDFACTOR]);
+            d3dcolor_to_rgba(&color.color[0], context->rs[D3DRS_BLENDFACTOR]);
             pipe->set_blend_color(pipe, &color);
         }
         if (group & NINE_STATE_SAMPLE_MASK) {
-            pipe->set_sample_mask(pipe, state->rs[D3DRS_MULTISAMPLEMASK]);
+            if (context->rt[0]->desc.MultiSampleType <= D3DMULTISAMPLE_NONMASKABLE) {
+                pipe->set_sample_mask(pipe, ~0);
+            } else {
+                pipe->set_sample_mask(pipe, context->rs[D3DRS_MULTISAMPLEMASK]);
+            }
         }
         if (group & NINE_STATE_STENCIL_REF) {
             struct pipe_stencil_ref ref;
-            ref.ref_value[0] = state->rs[D3DRS_STENCILREF];
+            ref.ref_value[0] = context->rs[D3DRS_STENCILREF];
             ref.ref_value[1] = ref.ref_value[0];
             pipe->set_stencil_ref(pipe, &ref);
         }
@@ -947,8 +1060,816 @@ nine_update_state(struct NineDevice9 *device)
         (NINE_STATE_FF | NINE_STATE_VS_CONST | NINE_STATE_PS_CONST);
 
     DBG("finished\n");
+}
+
+#define RESZ_CODE 0x7fa05000
+
+static void
+NineDevice9_ResolveZ( struct NineDevice9 *device )
+{
+    struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
+    const struct util_format_description *desc;
+    struct NineSurface9 *source = state->ds;
+    struct NineBaseTexture9 *destination = context->texture[0];
+    struct pipe_resource *src, *dst;
+    struct pipe_blit_info blit;
+
+    DBG("RESZ resolve\n");
+
+    if (!source || !destination ||
+        destination->base.type != D3DRTYPE_TEXTURE)
+        return;
+
+    src = source->base.resource;
+    dst = destination->base.resource;
+
+    if (!src || !dst)
+        return;
+
+    /* check dst is depth format. we know already for src */
+    desc = util_format_description(dst->format);
+    if (desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS)
+        return;
+
+    memset(&blit, 0, sizeof(blit));
+    blit.src.resource = src;
+    blit.src.level = 0;
+    blit.src.format = src->format;
+    blit.src.box.z = 0;
+    blit.src.box.depth = 1;
+    blit.src.box.x = 0;
+    blit.src.box.y = 0;
+    blit.src.box.width = src->width0;
+    blit.src.box.height = src->height0;
+
+    blit.dst.resource = dst;
+    blit.dst.level = 0;
+    blit.dst.format = dst->format;
+    blit.dst.box.z = 0;
+    blit.dst.box.depth = 1;
+    blit.dst.box.x = 0;
+    blit.dst.box.y = 0;
+    blit.dst.box.width = dst->width0;
+    blit.dst.box.height = dst->height0;
+
+    blit.mask = PIPE_MASK_ZS;
+    blit.filter = PIPE_TEX_FILTER_NEAREST;
+    blit.scissor_enable = FALSE;
+
+    device->pipe->blit(device->pipe, &blit);
+}
+
+#define ALPHA_TO_COVERAGE_ENABLE   MAKEFOURCC('A', '2', 'M', '1')
+#define ALPHA_TO_COVERAGE_DISABLE  MAKEFOURCC('A', '2', 'M', '0')
+
+
+void
+nine_context_set_render_state(struct NineDevice9 *device,
+                              D3DRENDERSTATETYPE State,
+                              DWORD Value)
+{
+    struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
+
+    /* Amd hacks (equivalent to GL extensions) */
+    if (unlikely(State == D3DRS_POINTSIZE)) {
+        if (Value == RESZ_CODE) {
+            NineDevice9_ResolveZ(device);
+            return;
+        }
+
+        if (Value == ALPHA_TO_COVERAGE_ENABLE ||
+            Value == ALPHA_TO_COVERAGE_DISABLE) {
+            context->rs[NINED3DRS_ALPHACOVERAGE] = (Value == ALPHA_TO_COVERAGE_ENABLE);
+            state->changed.group |= NINE_STATE_BLEND;
+            return;
+        }
+    }
+
+    /* NV hack */
+    if (unlikely(State == D3DRS_ADAPTIVETESS_Y)) {
+        if (Value == D3DFMT_ATOC || (Value == D3DFMT_UNKNOWN && context->rs[NINED3DRS_ALPHACOVERAGE])) {
+            context->rs[NINED3DRS_ALPHACOVERAGE] = (Value == D3DFMT_ATOC) ? 3 : 0;
+            context->rs[NINED3DRS_ALPHACOVERAGE] &= context->rs[D3DRS_ALPHATESTENABLE] ? 3 : 2;
+            state->changed.group |= NINE_STATE_BLEND;
+            return;
+        }
+    }
+    if (unlikely(State == D3DRS_ALPHATESTENABLE && (context->rs[NINED3DRS_ALPHACOVERAGE] & 2))) {
+        DWORD alphacoverage_prev = context->rs[NINED3DRS_ALPHACOVERAGE];
+        context->rs[NINED3DRS_ALPHACOVERAGE] = (Value ? 3 : 2);
+        if (context->rs[NINED3DRS_ALPHACOVERAGE] != alphacoverage_prev)
+            state->changed.group |= NINE_STATE_BLEND;
+    }
+
+    context->rs[State] = nine_fix_render_state_value(State, Value);
+    state->changed.group |= nine_render_state_group[State];
+}
+
+void
+nine_context_set_texture(struct NineDevice9 *device,
+                         DWORD Stage,
+                         struct NineBaseTexture9 *tex)
+{
+    struct nine_context *context = &device->context;
+    struct nine_state *state = &device->state;
+
+    context->samplers_shadow &= ~(1 << Stage);
+    if (tex)
+        context->samplers_shadow |= tex->shadow << Stage;
+
+    nine_bind(&context->texture[Stage], tex);
+
+    state->changed.group |= NINE_STATE_TEXTURE;
+}
+
+void
+nine_context_set_sampler_state(struct NineDevice9 *device,
+                               DWORD Sampler,
+                               D3DSAMPLERSTATETYPE Type,
+                               DWORD Value)
+{
+    struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
+
+    if (unlikely(!nine_check_sampler_state_value(Type, Value)))
+        return;
+
+    context->samp[Sampler][Type] = Value;
+    state->changed.group |= NINE_STATE_SAMPLER;
+    context->changed.sampler[Sampler] |= 1 << Type;
+}
+
+void
+nine_context_set_stream_source(struct NineDevice9 *device,
+                               UINT StreamNumber,
+                               struct NineVertexBuffer9 *pVBuf9,
+                               UINT OffsetInBytes,
+                               UINT Stride)
+{
+    struct nine_context *context = &device->context;
+    const unsigned i = StreamNumber;
+
+    context->changed.vtxbuf |= 1 << StreamNumber;
+
+    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);
+}
+
+void
+nine_context_set_stream_source_freq(struct NineDevice9 *device,
+                                    UINT StreamNumber,
+                                    UINT Setting)
+{
+    struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
+
+    context->stream_freq[StreamNumber] = Setting;
+
+    if (Setting & D3DSTREAMSOURCE_INSTANCEDATA)
+        context->stream_instancedata_mask |= 1 << StreamNumber;
+    else
+        context->stream_instancedata_mask &= ~(1 << StreamNumber);
+
+    if (StreamNumber != 0)
+        state->changed.group |= NINE_STATE_STREAMFREQ;
+}
 
-    return TRUE;
+void
+nine_context_set_indices(struct NineDevice9 *device,
+                         struct NineIndexBuffer9 *idxbuf)
+{
+    struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
+    const struct pipe_index_buffer *pipe_idxbuf;
+
+    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);
+
+    state->changed.group |= NINE_STATE_IDXBUF;
+}
+
+void
+nine_context_set_vertex_declaration(struct NineDevice9 *device,
+                                    struct NineVertexDeclaration9 *vdecl)
+{
+    struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
+    BOOL was_programmable_vs = context->programmable_vs;
+
+    nine_bind(&context->vdecl, vdecl);
+
+    context->programmable_vs = context->vs && !(context->vdecl && context->vdecl->position_t);
+    if (was_programmable_vs != context->programmable_vs) {
+        context->commit |= NINE_STATE_COMMIT_CONST_VS;
+        state->changed.group |= NINE_STATE_VS;
+    }
+
+    state->changed.group |= NINE_STATE_VDECL;
+}
+
+void
+nine_context_set_vertex_shader(struct NineDevice9 *device,
+                               struct NineVertexShader9 *pShader)
+{
+    struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
+    BOOL was_programmable_vs = context->programmable_vs;
+
+    nine_bind(&context->vs, pShader);
+
+    context->programmable_vs = context->vs && !(context->vdecl && context->vdecl->position_t);
+
+    /* ff -> non-ff: commit back non-ff constants */
+    if (!was_programmable_vs && context->programmable_vs)
+        context->commit |= NINE_STATE_COMMIT_CONST_VS;
+
+    state->changed.group |= NINE_STATE_VS;
+}
+
+void
+nine_context_set_vertex_shader_constant_f(struct NineDevice9 *device,
+                                          UINT StartRegister,
+                                          const float *pConstantData,
+                                          UINT Vector4fCount)
+{
+    struct nine_state *state = &device->state;
+    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]));
+
+    if (device->may_swvp) {
+        Vector4fCount = MIN2(StartRegister + Vector4fCount, NINE_MAX_CONST_F) - StartRegister;
+        if (StartRegister < NINE_MAX_CONST_F)
+            memcpy(&context->vs_const_f[StartRegister * 4],
+                   pConstantData,
+                   Vector4fCount * 4 * sizeof(context->vs_const_f[0]));
+    }
+
+    context->changed.vs_const_f = TRUE;
+    state->changed.group |= NINE_STATE_VS_CONST;
+}
+
+
+void
+nine_context_set_vertex_shader_constant_i(struct NineDevice9 *device,
+                                          UINT StartRegister,
+                                          const int *pConstantData,
+                                          UINT Vector4iCount)
+{
+    struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
+    int i;
+
+    if (device->driver_caps.vs_integer) {
+        memcpy(&context->vs_const_i[4 * StartRegister],
+               pConstantData,
+               Vector4iCount * sizeof(int[4]));
+    } else {
+        for (i = 0; i < Vector4iCount; i++) {
+            context->vs_const_i[4 * (StartRegister + i)] = fui((float)(pConstantData[4 * i]));
+            context->vs_const_i[4 * (StartRegister + i) + 1] = fui((float)(pConstantData[4 * i + 1]));
+            context->vs_const_i[4 * (StartRegister + i) + 2] = fui((float)(pConstantData[4 * i + 2]));
+            context->vs_const_i[4 * (StartRegister + i) + 3] = fui((float)(pConstantData[4 * i + 3]));
+        }
+    }
+
+    context->changed.vs_const_i = TRUE;
+    state->changed.group |= NINE_STATE_VS_CONST;
+}
+
+void
+nine_context_set_vertex_shader_constant_b(struct NineDevice9 *device,
+                                          UINT StartRegister,
+                                          const BOOL *pConstantData,
+                                          UINT BoolCount)
+{
+    struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
+    int i;
+    uint32_t bool_true = device->driver_caps.vs_integer ? 0xFFFFFFFF : fui(1.0f);
+
+    for (i = 0; i < BoolCount; i++)
+        context->vs_const_b[StartRegister + i] = pConstantData[i] ? bool_true : 0;
+
+    context->changed.vs_const_b = TRUE;
+    state->changed.group |= NINE_STATE_VS_CONST;
+}
+
+void
+nine_context_set_pixel_shader_constant_f(struct NineDevice9 *device,
+                                        UINT StartRegister,
+                                        const float *pConstantData,
+                                        UINT Vector4fCount)
+{
+    struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
+
+    memcpy(&context->ps_const_f[StartRegister * 4],
+           pConstantData,
+           Vector4fCount * 4 * sizeof(context->ps_const_f[0]));
+
+    context->changed.ps_const_f = TRUE;
+    state->changed.group |= NINE_STATE_PS_CONST;
+}
+
+void
+nine_context_set_pixel_shader_constant_i(struct NineDevice9 *device,
+                                         UINT StartRegister,
+                                         const int *pConstantData,
+                                         UINT Vector4iCount)
+{
+    struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
+    int i;
+
+    if (device->driver_caps.ps_integer) {
+        memcpy(&context->ps_const_i[StartRegister][0],
+               pConstantData,
+               Vector4iCount * sizeof(context->ps_const_i[0]));
+    } else {
+        for (i = 0; i < Vector4iCount; i++) {
+            context->ps_const_i[StartRegister+i][0] = fui((float)(pConstantData[4*i]));
+            context->ps_const_i[StartRegister+i][1] = fui((float)(pConstantData[4*i+1]));
+            context->ps_const_i[StartRegister+i][2] = fui((float)(pConstantData[4*i+2]));
+            context->ps_const_i[StartRegister+i][3] = fui((float)(pConstantData[4*i+3]));
+        }
+    }
+    context->changed.ps_const_i = TRUE;
+    state->changed.group |= NINE_STATE_PS_CONST;
+}
+
+void
+nine_context_set_pixel_shader_constant_b(struct NineDevice9 *device,
+                                         UINT StartRegister,
+                                         const BOOL *pConstantData,
+                                         UINT BoolCount)
+{
+    struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
+    int i;
+    uint32_t bool_true = device->driver_caps.ps_integer ? 0xFFFFFFFF : fui(1.0f);
+
+    for (i = 0; i < BoolCount; i++)
+        context->ps_const_b[StartRegister + i] = pConstantData[i] ? bool_true : 0;
+
+    context->changed.ps_const_b = TRUE;
+    state->changed.group |= NINE_STATE_PS_CONST;
+}
+
+void
+nine_context_set_render_target(struct NineDevice9 *device,
+                               DWORD RenderTargetIndex,
+                               struct NineSurface9 *rt)
+{
+    struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
+    const unsigned i = RenderTargetIndex;
+
+    if (i == 0) {
+        /* viewport and scissor changes */
+        state->changed.group |= NINE_STATE_VIEWPORT | NINE_STATE_SCISSOR | NINE_STATE_MULTISAMPLE;
+
+        if (context->rt[0] &&
+            (context->rt[0]->desc.MultiSampleType <= D3DMULTISAMPLE_NONMASKABLE) !=
+            (rt->desc.MultiSampleType <= D3DMULTISAMPLE_NONMASKABLE))
+            state->changed.group |= NINE_STATE_SAMPLE_MASK;
+    }
+
+    if (context->rt[i] != rt) {
+       nine_bind(&context->rt[i], rt);
+       state->changed.group |= NINE_STATE_FB;
+    }
+}
+
+void
+nine_context_apply_stateblock(struct NineDevice9 *device,
+                              const struct nine_state *src)
+{
+    struct nine_context *context = &device->context;
+    int i;
+
+    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);
+            context->rs[i * 32 + r] = nine_fix_render_state_value(i * 32 + r, src->rs_advertised[i * 32 + r]);
+        }
+    }
+
+    /* Textures */
+    if (src->changed.texture) {
+        uint32_t m = src->changed.texture;
+        unsigned s;
+
+        context->samplers_shadow &= ~m;
+
+        for (s = 0; m; ++s, m >>= 1) {
+            struct NineBaseTexture9 *tex = src->texture[s];
+            if (!(m & 1))
+                continue;
+            if (tex)
+                context->samplers_shadow |= tex->shadow << s;
+            nine_bind(&context->texture[s], src->texture[s]);
+        }
+    }
+
+    /* 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);
+                if (nine_check_sampler_state_value(i, src->samp_advertised[s][i]))
+                    context->samp[s][i] = src->samp_advertised[s][i];
+            }
+            context->changed.sampler[s] |= src->changed.sampler[s];
+        }
+    }
+
+    /* 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)) {
+                if (src->stream[i]) {
+                    context->vtxbuf[i].buffer_offset = src->vtxbuf[i].buffer_offset;
+                    pipe_resource_reference(&context->vtxbuf[i].buffer,
+                        src->stream[i] ? NineVertexBuffer9_GetResource(src->stream[i]) : NULL);
+                    context->vtxbuf[i].stride = src->vtxbuf[i].stride;
+                }
+            }
+            if (src->changed.stream_freq & (1 << i)) {
+                context->stream_freq[i] = src->stream_freq[i];
+                if (src->stream_freq[i] & D3DSTREAMSOURCE_INSTANCEDATA)
+                    context->stream_instancedata_mask |= 1 << i;
+                else
+                    context->stream_instancedata_mask &= ~(1 << i);
+            }
+        }
+        context->changed.vtxbuf |= src->changed.vtxbuf;
+    }
+
+    /* 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_bind(&context->vs, src->vs);
+
+    context->programmable_vs = context->vs && !(context->vdecl && context->vdecl->position_t);
+
+    /* Vertex constants */
+    if (src->changed.group & NINE_STATE_VS_CONST) {
+        struct nine_range *r;
+        if (device->may_swvp) {
+            for (r = src->changed.vs_const_f; r; r = r->next) {
+                int bgn = r->bgn;
+                int end = r->end;
+                memcpy(&context->vs_const_f_swvp[bgn * 4],
+                       &src->vs_const_f[bgn * 4],
+                       (end - bgn) * 4 * sizeof(float));
+                if (bgn < device->max_vs_const_f) {
+                    end = MIN2(end, device->max_vs_const_f);
+                    memcpy(&context->vs_const_f[bgn * 4],
+                           &src->vs_const_f[bgn * 4],
+                           (end - bgn) * 4 * sizeof(float));
+                }
+            }
+        } else {
+            for (r = src->changed.vs_const_f; r; r = r->next) {
+                memcpy(&context->vs_const_f[r->bgn * 4],
+                       &src->vs_const_f[r->bgn * 4],
+                       (r->end - r->bgn) * 4 * sizeof(float));
+            }
+        }
+        for (r = src->changed.vs_const_i; r; r = r->next) {
+            memcpy(&context->vs_const_i[r->bgn * 4],
+                   &src->vs_const_i[r->bgn * 4],
+                   (r->end - r->bgn) * 4 * sizeof(int));
+        }
+        for (r = src->changed.vs_const_b; r; r = r->next) {
+            memcpy(&context->vs_const_b[r->bgn],
+                   &src->vs_const_b[r->bgn],
+                   (r->end - r->bgn) * sizeof(int));
+        }
+        context->changed.vs_const_f = !!src->changed.vs_const_f;
+        context->changed.vs_const_i = !!src->changed.vs_const_i;
+        context->changed.vs_const_b = !!src->changed.vs_const_b;
+    }
+
+    /* 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) {
+            memcpy(&context->ps_const_f[r->bgn * 4],
+                   &src->ps_const_f[r->bgn * 4],
+                   (r->end - r->bgn) * 4 * sizeof(float));
+        }
+        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)
+                    memcpy(context->ps_const_i[i], src->ps_const_i[i], 4 * sizeof(int));
+        }
+        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)
+                    context->ps_const_b[i] = src->ps_const_b[i];
+        }
+        context->changed.ps_const_f = !!src->changed.ps_const_f;
+        context->changed.ps_const_i = !!src->changed.ps_const_i;
+        context->changed.ps_const_b = !!src->changed.ps_const_b;
+    }
+}
+
+static void
+nine_update_state_framebuffer_clear(struct NineDevice9 *device)
+{
+    struct nine_state *state = &device->state;
+
+    validate_textures(device);
+
+    if (state->changed.group & NINE_STATE_FB)
+        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)
+{
+    struct nine_context *context = &device->context;
+    const int sRGB = context->rs[D3DRS_SRGBWRITEENABLE] ? 1 : 0;
+    struct pipe_surface *cbuf, *zsbuf;
+    struct pipe_context *pipe = device->pipe;
+    struct NineSurface9 *zsbuf_surf = device->state.ds;
+    struct NineSurface9 *rt;
+    unsigned bufs = 0;
+    unsigned r, i;
+    union pipe_color_union rgba;
+    unsigned rt_mask = 0;
+    D3DRECT rect;
+
+    nine_update_state_framebuffer_clear(device);
+
+    if (Flags & D3DCLEAR_TARGET) bufs |= PIPE_CLEAR_COLOR;
+    /* Ignore Z buffer if not bound */
+    if (context->pipe.fb.zsbuf != NULL) {
+        if (Flags & D3DCLEAR_ZBUFFER) bufs |= PIPE_CLEAR_DEPTH;
+        if (Flags & D3DCLEAR_STENCIL) bufs |= PIPE_CLEAR_STENCIL;
+    }
+    if (!bufs)
+        return;
+    d3dcolor_to_pipe_color_union(&rgba, Color);
+
+    rect.x1 = device->state.viewport.X;
+    rect.y1 = device->state.viewport.Y;
+    rect.x2 = device->state.viewport.Width + rect.x1;
+    rect.y2 = device->state.viewport.Height + rect.y1;
+
+    /* Both rectangles apply, which is weird, but that's D3D9. */
+    if (context->rs[D3DRS_SCISSORTESTENABLE]) {
+        rect.x1 = MAX2(rect.x1, device->state.scissor.minx);
+        rect.y1 = MAX2(rect.y1, device->state.scissor.miny);
+        rect.x2 = MIN2(rect.x2, device->state.scissor.maxx);
+        rect.y2 = MIN2(rect.y2, device->state.scissor.maxy);
+    }
+
+    if (Count) {
+        /* Maybe apps like to specify a large rect ? */
+        if (pRects[0].x1 <= rect.x1 && pRects[0].x2 >= rect.x2 &&
+            pRects[0].y1 <= rect.y1 && pRects[0].y2 >= rect.y2) {
+            DBG("First rect covers viewport.\n");
+            Count = 0;
+            pRects = NULL;
+        }
+    }
+
+    if (rect.x1 >= context->pipe.fb.width || rect.y1 >= context->pipe.fb.height)
+        return;
+
+    for (i = 0; i < device->caps.NumSimultaneousRTs; ++i) {
+        if (context->rt[i] && context->rt[i]->desc.Format != D3DFMT_NULL)
+            rt_mask |= 1 << i;
+    }
+
+    /* fast path, clears everything at once */
+    if (!Count &&
+        (!(bufs & PIPE_CLEAR_COLOR) || (rt_mask == context->rt_mask)) &&
+        rect.x1 == 0 && rect.y1 == 0 &&
+        /* Case we clear only render target. Check clear region vs rt. */
+        ((!(bufs & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) &&
+         rect.x2 >= context->pipe.fb.width &&
+         rect.y2 >= context->pipe.fb.height) ||
+        /* Case we clear depth buffer (and eventually rt too).
+         * depth buffer size is always >= rt size. Compare to clear region */
+        ((bufs & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) &&
+         rect.x2 >= zsbuf_surf->desc.Width &&
+         rect.y2 >= zsbuf_surf->desc.Height))) {
+        DBG("Clear fast path\n");
+        pipe->clear(pipe, bufs, &rgba, Z, Stencil);
+        return;
+    }
+
+    if (!Count) {
+        Count = 1;
+        pRects = &rect;
+    }
+
+    for (i = 0; i < device->caps.NumSimultaneousRTs; ++i) {
+        rt = context->rt[i];
+        if (!rt || rt->desc.Format == D3DFMT_NULL ||
+            !(bufs & PIPE_CLEAR_COLOR))
+            continue; /* save space, compiler should hoist this */
+        cbuf = NineSurface9_GetSurface(rt, sRGB);
+        for (r = 0; r < Count; ++r) {
+            /* Don't trust users to pass these in the right order. */
+            unsigned x1 = MIN2(pRects[r].x1, pRects[r].x2);
+            unsigned y1 = MIN2(pRects[r].y1, pRects[r].y2);
+            unsigned x2 = MAX2(pRects[r].x1, pRects[r].x2);
+            unsigned y2 = MAX2(pRects[r].y1, pRects[r].y2);
+#ifndef NINE_LAX
+            /* Drop negative rectangles (like wine expects). */
+            if (pRects[r].x1 > pRects[r].x2) continue;
+            if (pRects[r].y1 > pRects[r].y2) continue;
+#endif
+
+            x1 = MAX2(x1, rect.x1);
+            y1 = MAX2(y1, rect.y1);
+            x2 = MIN3(x2, rect.x2, rt->desc.Width);
+            y2 = MIN3(y2, rect.y2, rt->desc.Height);
+
+            DBG("Clearing (%u..%u)x(%u..%u)\n", x1, x2, y1, y2);
+            pipe->clear_render_target(pipe, cbuf, &rgba,
+                                      x1, y1, x2 - x1, y2 - y1, false);
+        }
+    }
+    if (!(bufs & PIPE_CLEAR_DEPTHSTENCIL))
+        return;
+
+    bufs &= PIPE_CLEAR_DEPTHSTENCIL;
+
+    for (r = 0; r < Count; ++r) {
+        unsigned x1 = MIN2(pRects[r].x1, pRects[r].x2);
+        unsigned y1 = MIN2(pRects[r].y1, pRects[r].y2);
+        unsigned x2 = MAX2(pRects[r].x1, pRects[r].x2);
+        unsigned y2 = MAX2(pRects[r].y1, pRects[r].y2);
+#ifndef NINE_LAX
+        /* Drop negative rectangles. */
+        if (pRects[r].x1 > pRects[r].x2) continue;
+        if (pRects[r].y1 > pRects[r].y2) continue;
+#endif
+
+        x1 = MIN2(x1, rect.x1);
+        y1 = MIN2(y1, rect.y1);
+        x2 = MIN3(x2, rect.x2, zsbuf_surf->desc.Width);
+        y2 = MIN3(y2, rect.y2, zsbuf_surf->desc.Height);
+
+        zsbuf = NineSurface9_GetSurface(zsbuf_surf, 0);
+        assert(zsbuf);
+        pipe->clear_depth_stencil(pipe, zsbuf, bufs, Z, Stencil,
+                                  x1, y1, x2 - x1, y2 - y1, false);
+    }
+    return;
+}
+
+
+static inline void
+init_draw_info(struct pipe_draw_info *info,
+               struct NineDevice9 *dev, D3DPRIMITIVETYPE type, UINT count)
+{
+    info->mode = d3dprimitivetype_to_pipe_prim(type);
+    info->count = prim_count_to_vertex_count(type, count);
+    info->start_instance = 0;
+    info->instance_count = 1;
+    if (dev->context.stream_instancedata_mask & dev->context.stream_usage_mask)
+        info->instance_count = MAX2(dev->context.stream_freq[0] & 0x7FFFFF, 1);
+    info->primitive_restart = FALSE;
+    info->restart_index = 0;
+    info->count_from_stream_output = NULL;
+    info->indirect = NULL;
+    info->indirect_params = NULL;
+}
+
+void
+nine_context_draw_primitive(struct NineDevice9 *device,
+                            D3DPRIMITIVETYPE PrimitiveType,
+                            UINT StartVertex,
+                            UINT PrimitiveCount)
+{
+    struct pipe_draw_info info;
+
+    nine_update_state(device);
+
+    init_draw_info(&info, device, PrimitiveType, PrimitiveCount);
+    info.indexed = FALSE;
+    info.start = StartVertex;
+    info.index_bias = 0;
+    info.min_index = info.start;
+    info.max_index = info.count - 1;
+
+    device->pipe->draw_vbo(device->pipe, &info);
+}
+
+void
+nine_context_draw_indexed_primitive(struct NineDevice9 *device,
+                                    D3DPRIMITIVETYPE PrimitiveType,
+                                    INT BaseVertexIndex,
+                                    UINT MinVertexIndex,
+                                    UINT NumVertices,
+                                    UINT StartIndex,
+                                    UINT PrimitiveCount)
+{
+    struct pipe_draw_info info;
+
+    nine_update_state(device);
+
+    init_draw_info(&info, device, PrimitiveType, PrimitiveCount);
+    info.indexed = TRUE;
+    info.start = StartIndex;
+    info.index_bias = BaseVertexIndex;
+    /* These don't include index bias: */
+    info.min_index = MinVertexIndex;
+    info.max_index = MinVertexIndex + NumVertices - 1;
+
+    device->pipe->draw_vbo(device->pipe, &info);
+}
+
+void
+nine_context_draw_primitive_from_vtxbuf(struct NineDevice9 *device,
+                                        D3DPRIMITIVETYPE PrimitiveType,
+                                        UINT PrimitiveCount,
+                                        struct pipe_vertex_buffer *vtxbuf)
+{
+    struct pipe_draw_info info;
+
+    nine_update_state(device);
+
+    init_draw_info(&info, device, PrimitiveType, PrimitiveCount);
+    info.indexed = FALSE;
+    info.start = 0;
+    info.index_bias = 0;
+    info.min_index = 0;
+    info.max_index = info.count - 1;
+
+    device->pipe->set_vertex_buffers(device->pipe, 0, 1, vtxbuf);
+
+    device->pipe->draw_vbo(device->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)
+{
+    struct pipe_draw_info info;
+
+    nine_update_state(device);
+
+    init_draw_info(&info, device, PrimitiveType, PrimitiveCount);
+    info.indexed = TRUE;
+    info.start = 0;
+    info.index_bias = 0;
+    info.min_index = MinVertexIndex;
+    info.max_index = MinVertexIndex + NumVertices - 1;
+    device->pipe->set_vertex_buffers(device->pipe, 0, 1, vbuf);
+    device->pipe->set_index_buffer(device->pipe, ibuf);
+
+    device->pipe->draw_vbo(device->pipe, &info);
 }
 
 /* State defaults */
@@ -1105,18 +2026,19 @@ static const DWORD nine_samp_state_defaults[NINED3DSAMP_LAST + 1] =
     [D3DSAMP_ELEMENTINDEX] = 0,
     [D3DSAMP_DMAPOFFSET] = 0,
     [NINED3DSAMP_MINLOD] = 0,
-    [NINED3DSAMP_SHADOW] = 0
+    [NINED3DSAMP_SHADOW] = 0,
+    [NINED3DSAMP_CUBETEX] = 0
 };
 
 void nine_state_restore_non_cso(struct NineDevice9 *device)
 {
     struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
 
     state->changed.group = NINE_STATE_ALL;
-    state->changed.vtxbuf = (1ULL << device->caps.MaxStreams) - 1;
+    context->changed.vtxbuf = (1ULL << device->caps.MaxStreams) - 1;
     state->changed.ucp = (1 << PIPE_MAX_CLIP_PLANES) - 1;
-    state->changed.texture = NINE_PS_SAMPLERS_MASK | NINE_VS_SAMPLERS_MASK;
-    state->commit |= NINE_STATE_COMMIT_CONST_VS | NINE_STATE_COMMIT_CONST_PS;
+    context->commit |= NINE_STATE_COMMIT_CONST_VS | NINE_STATE_COMMIT_CONST_PS;
 }
 
 void
@@ -1124,11 +2046,12 @@ nine_state_set_defaults(struct NineDevice9 *device, const D3DCAPS9 *caps,
                         boolean is_reset)
 {
     struct nine_state *state = &device->state;
+    struct nine_context *context = &device->context;
     unsigned s;
 
     /* Initialize defaults.
      */
-    memcpy(state->rs, nine_render_state_defaults, sizeof(state->rs));
+    memcpy(context->rs, nine_render_state_defaults, sizeof(context->rs));
 
     for (s = 0; s < ARRAY_SIZE(state->ff.tex_stage); ++s) {
         memcpy(&state->ff.tex_stage[s], nine_tex_stage_state_defaults,
@@ -1137,30 +2060,41 @@ nine_state_set_defaults(struct NineDevice9 *device, const D3DCAPS9 *caps,
     }
     state->ff.tex_stage[0][D3DTSS_COLOROP] = D3DTOP_MODULATE;
     state->ff.tex_stage[0][D3DTSS_ALPHAOP] = D3DTOP_SELECTARG1;
-    memset(&state->bumpmap_vars, 0, sizeof(state->bumpmap_vars));
+    memset(&context->bumpmap_vars, 0, sizeof(context->bumpmap_vars));
 
-    for (s = 0; s < ARRAY_SIZE(state->samp); ++s) {
-        memcpy(&state->samp[s], nine_samp_state_defaults,
-               sizeof(state->samp[s]));
+    for (s = 0; s < NINE_MAX_SAMPLERS; ++s) {
+        memcpy(&context->samp[s], nine_samp_state_defaults,
+               sizeof(context->samp[s]));
+        memcpy(&state->samp_advertised[s], nine_samp_state_defaults,
+               sizeof(state->samp_advertised[s]));
     }
 
-    if (state->vs_const_f)
-        memset(state->vs_const_f, 0, device->vs_const_size);
-    if (state->ps_const_f)
-        memset(state->ps_const_f, 0, device->ps_const_size);
+    memset(state->vs_const_f, 0, VS_CONST_F_SIZE(device));
+    memset(context->vs_const_f, 0, device->vs_const_size);
+    if (context->vs_const_f_swvp)
+        memset(context->vs_const_f_swvp, 0, NINE_MAX_CONST_F_SWVP * sizeof(float[4]));
+    memset(state->vs_const_i, 0, VS_CONST_I_SIZE(device));
+    memset(context->vs_const_i, 0, VS_CONST_I_SIZE(device));
+    memset(state->vs_const_b, 0, VS_CONST_B_SIZE(device));
+    memset(context->vs_const_b, 0, VS_CONST_B_SIZE(device));
+    memset(state->ps_const_f, 0, device->ps_const_size);
+    memset(context->ps_const_f, 0, device->ps_const_size);
+    memset(state->ps_const_i, 0, sizeof(state->ps_const_i));
+    memset(context->ps_const_i, 0, sizeof(context->ps_const_i));
+    memset(state->ps_const_b, 0, sizeof(state->ps_const_b));
+    memset(context->ps_const_b, 0, sizeof(context->ps_const_b));
 
     /* Cap dependent initial state:
      */
-    state->rs[D3DRS_POINTSIZE_MAX] = fui(caps->MaxPointSize);
+    context->rs[D3DRS_POINTSIZE_MAX] = fui(caps->MaxPointSize);
 
-    memcpy(state->rs_advertised, state->rs, sizeof(state->rs));
+    memcpy(state->rs_advertised, context->rs, sizeof(context->rs));
 
     /* Set changed flags to initialize driver.
      */
     state->changed.group = NINE_STATE_ALL;
-    state->changed.vtxbuf = (1ULL << device->caps.MaxStreams) - 1;
+    context->changed.vtxbuf = (1ULL << device->caps.MaxStreams) - 1;
     state->changed.ucp = (1 << PIPE_MAX_CLIP_PLANES) - 1;
-    state->changed.texture = NINE_PS_SAMPLERS_MASK | NINE_VS_SAMPLERS_MASK;
 
     state->ff.changed.transform[0] = ~0;
     state->ff.changed.transform[D3DTS_WORLD / 32] |= 1 << (D3DTS_WORLD % 32);
@@ -1170,12 +2104,12 @@ nine_state_set_defaults(struct NineDevice9 *device, const D3DCAPS9 *caps,
         state->viewport.MaxZ = 1.0f;
     }
 
-    for (s = 0; s < ARRAY_SIZE(state->changed.sampler); ++s)
-        state->changed.sampler[s] = ~0;
+    for (s = 0; s < NINE_MAX_SAMPLERS; ++s)
+        context->changed.sampler[s] = ~0;
 
     if (!is_reset) {
-        state->dummy_vbo_bound_at = -1;
-        state->vbo_bound_done = FALSE;
+        context->dummy_vbo_bound_at = -1;
+        context->vbo_bound_done = FALSE;
     }
 }
 
@@ -1190,10 +2124,9 @@ nine_state_clear(struct nine_state *state, const boolean device)
     nine_bind(&state->vs, NULL);
     nine_bind(&state->ps, NULL);
     nine_bind(&state->vdecl, NULL);
-    for (i = 0; i < PIPE_MAX_ATTRIBS; ++i) {
+    for (i = 0; i < PIPE_MAX_ATTRIBS; ++i)
         nine_bind(&state->stream[i], NULL);
-        pipe_resource_reference(&state->vtxbuf[i].buffer, NULL);
-    }
+
     nine_bind(&state->idxbuf, NULL);
     for (i = 0; i < NINE_MAX_SAMPLERS; ++i) {
         if (device &&
@@ -1204,6 +2137,373 @@ nine_state_clear(struct nine_state *state, const boolean device)
     }
 }
 
+void
+nine_context_clear(struct nine_context *context)
+{
+    unsigned i;
+
+    for (i = 0; i < ARRAY_SIZE(context->rt); ++i)
+       nine_bind(&context->rt[i], NULL);
+    nine_bind(&context->vs, NULL);
+    nine_bind(&context->vdecl, NULL);
+    for (i = 0; i < PIPE_MAX_ATTRIBS; ++i)
+        pipe_resource_reference(&context->vtxbuf[i].buffer, NULL);
+    pipe_resource_reference(&context->idxbuf.buffer, NULL);
+
+    for (i = 0; i < NINE_MAX_SAMPLERS; ++i)
+        nine_bind(&context->texture[i], NULL);
+}
+
+void
+nine_state_init_sw(struct NineDevice9 *device)
+{
+    struct pipe_context *pipe_sw = device->pipe_sw;
+    struct pipe_rasterizer_state rast;
+    struct pipe_blend_state blend;
+    struct pipe_depth_stencil_alpha_state dsa;
+    struct pipe_framebuffer_state fb;
+
+    /* Only used with Streamout */
+    memset(&rast, 0, sizeof(rast));
+    rast.rasterizer_discard = true;
+    rast.point_quad_rasterization = 1; /* to make llvmpipe happy */
+    cso_set_rasterizer(device->cso_sw, &rast);
+
+    /* dummy settings */
+    memset(&blend, 0, sizeof(blend));
+    memset(&dsa, 0, sizeof(dsa));
+    memset(&fb, 0, sizeof(fb));
+    cso_set_blend(device->cso_sw, &blend);
+    cso_set_depth_stencil_alpha(device->cso_sw, &dsa);
+    cso_set_framebuffer(device->cso_sw, &fb);
+    cso_set_viewport_dims(device->cso_sw, 1.0, 1.0, false);
+    cso_set_fragment_shader_handle(device->cso_sw, util_make_empty_fragment_shader(pipe_sw));
+}
+
+/* There is duplication with update_vertex_elements.
+ * TODO: Share the code */
+
+static void
+update_vertex_elements_sw(struct NineDevice9 *device)
+{
+    struct nine_state *state = &device->state;
+    const struct NineVertexDeclaration9 *vdecl = device->state.vdecl;
+    const struct NineVertexShader9 *vs;
+    unsigned n, b, i;
+    int index;
+    char vdecl_index_map[16]; /* vs->num_inputs <= 16 */
+    char used_streams[device->caps.MaxStreams];
+    int dummy_vbo_stream = -1;
+    BOOL need_dummy_vbo = FALSE;
+    struct pipe_vertex_element ve[PIPE_MAX_ATTRIBS];
+    bool programmable_vs = state->vs && !(state->vdecl && state->vdecl->position_t);
+
+    memset(vdecl_index_map, -1, 16);
+    memset(used_streams, 0, device->caps.MaxStreams);
+    vs = programmable_vs ? device->state.vs : device->ff.vs;
+
+    if (vdecl) {
+        for (n = 0; n < vs->num_inputs; ++n) {
+            DBG("looking up input %u (usage %u) from vdecl(%p)\n",
+                n, vs->input_map[n].ndecl, vdecl);
+
+            for (i = 0; i < vdecl->nelems; i++) {
+                if (vdecl->usage_map[i] == vs->input_map[n].ndecl) {
+                    vdecl_index_map[n] = i;
+                    used_streams[vdecl->elems[i].vertex_buffer_index] = 1;
+                    break;
+                }
+            }
+            if (vdecl_index_map[n] < 0)
+                need_dummy_vbo = TRUE;
+        }
+    } else {
+        /* No vertex declaration. Likely will never happen in practice,
+         * but we need not crash on this */
+        need_dummy_vbo = TRUE;
+    }
+
+    if (need_dummy_vbo) {
+        for (i = 0; i < device->caps.MaxStreams; i++ ) {
+            if (!used_streams[i]) {
+                dummy_vbo_stream = i;
+                break;
+            }
+        }
+    }
+    /* TODO handle dummy_vbo */
+    assert (!need_dummy_vbo);
+
+    for (n = 0; n < vs->num_inputs; ++n) {
+        index = vdecl_index_map[n];
+        if (index >= 0) {
+            ve[n] = vdecl->elems[index];
+            b = ve[n].vertex_buffer_index;
+            /* XXX wine just uses 1 here: */
+            if (state->stream_freq[b] & D3DSTREAMSOURCE_INSTANCEDATA)
+                ve[n].instance_divisor = state->stream_freq[b] & 0x7FFFFF;
+        } else {
+            /* if the vertex declaration is incomplete compared to what the
+             * vertex shader needs, we bind a dummy vbo with 0 0 0 0.
+             * This is not precised by the spec, but is the behaviour
+             * tested on win */
+            ve[n].vertex_buffer_index = dummy_vbo_stream;
+            ve[n].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+            ve[n].src_offset = 0;
+            ve[n].instance_divisor = 0;
+        }
+    }
+
+    cso_set_vertex_elements(device->cso_sw, vs->num_inputs, ve);
+}
+
+static void
+update_vertex_buffers_sw(struct NineDevice9 *device, int start_vertice, int num_vertices)
+{
+    struct pipe_context *pipe = device->pipe;
+    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;
+    struct pipe_vertex_buffer vtxbuf;
+    uint32_t mask = 0xf;
+    unsigned i;
+
+    DBG("mask=%x\n", mask);
+
+    /* TODO: handle dummy_vbo_bound_at */
+
+    for (i = 0; mask; mask >>= 1, ++i) {
+        if (mask & 1) {
+            if (state->stream[i]) {
+                struct pipe_resource *buf;
+                struct pipe_box box;
+
+                vtxbuf = state->vtxbuf[i];
+                vtxbuf.buffer = NineVertexBuffer9_GetResource(state->stream[i]);
+
+                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,
+                         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,
+                                  0,
+                                  box.width,
+                                  16,
+                                  vtxbuf.user_buffer,
+                                  &(vtxbuf.buffer_offset),
+                                  &(vtxbuf.buffer));
+                    u_upload_unmap(device->vertex_sw_uploader);
+                    vtxbuf.user_buffer = NULL;
+                }
+                pipe_sw->set_vertex_buffers(pipe_sw, i, 1, &vtxbuf);
+                if (vtxbuf.buffer)
+                    pipe_resource_reference(&vtxbuf.buffer, NULL);
+            } else
+                pipe_sw->set_vertex_buffers(pipe_sw, i, 1, NULL);
+        }
+    }
+}
+
+static void
+update_vs_constants_sw(struct NineDevice9 *device)
+{
+    struct nine_state *state = &device->state;
+    struct pipe_context *pipe_sw = device->pipe_sw;
+
+    DBG("updating\n");
+
+    {
+        struct pipe_constant_buffer cb;
+        const void *buf;
+
+        cb.buffer = NULL;
+        cb.buffer_offset = 0;
+        cb.buffer_size = 4096 * sizeof(float[4]);
+        cb.user_buffer = state->vs_const_f;
+
+        if (state->vs->lconstf.ranges) {
+            const struct nine_lconstf *lconstf =  &device->state.vs->lconstf;
+            const struct nine_range *r = lconstf->ranges;
+            unsigned n = 0;
+            float *dst = device->state.vs_lconstf_temp;
+            float *src = (float *)cb.user_buffer;
+            memcpy(dst, src, 8192 * sizeof(float[4]));
+            while (r) {
+                unsigned p = r->bgn;
+                unsigned c = r->end - r->bgn;
+                memcpy(&dst[p * 4], &lconstf->data[n * 4], c * 4 * sizeof(float));
+                n += c;
+                r = r->next;
+            }
+            cb.user_buffer = dst;
+        }
+
+        buf = cb.user_buffer;
+        if (!device->driver_caps.user_sw_cbufs) {
+            u_upload_data(device->constbuf_sw_uploader,
+                          0,
+                          cb.buffer_size,
+                          16,
+                          cb.user_buffer,
+                          &(cb.buffer_offset),
+                          &(cb.buffer));
+            u_upload_unmap(device->constbuf_sw_uploader);
+            cb.user_buffer = NULL;
+        }
+
+        pipe_sw->set_constant_buffer(pipe_sw, PIPE_SHADER_VERTEX, 0, &cb);
+        if (cb.buffer)
+            pipe_resource_reference(&cb.buffer, NULL);
+
+        cb.user_buffer = (char *)buf + 4096 * sizeof(float[4]);
+        if (!device->driver_caps.user_sw_cbufs) {
+            u_upload_data(device->constbuf_sw_uploader,
+                          0,
+                          cb.buffer_size,
+                          16,
+                          cb.user_buffer,
+                          &(cb.buffer_offset),
+                          &(cb.buffer));
+            u_upload_unmap(device->constbuf_sw_uploader);
+            cb.user_buffer = NULL;
+        }
+
+        pipe_sw->set_constant_buffer(pipe_sw, PIPE_SHADER_VERTEX, 1, &cb);
+        if (cb.buffer)
+            pipe_resource_reference(&cb.buffer, NULL);
+    }
+
+    {
+        struct pipe_constant_buffer cb;
+
+        cb.buffer = NULL;
+        cb.buffer_offset = 0;
+        cb.buffer_size = 2048 * sizeof(float[4]);
+        cb.user_buffer = state->vs_const_i;
+
+        if (!device->driver_caps.user_sw_cbufs) {
+            u_upload_data(device->constbuf_sw_uploader,
+                          0,
+                          cb.buffer_size,
+                          16,
+                          cb.user_buffer,
+                          &(cb.buffer_offset),
+                          &(cb.buffer));
+            u_upload_unmap(device->constbuf_sw_uploader);
+            cb.user_buffer = NULL;
+        }
+
+        pipe_sw->set_constant_buffer(pipe_sw, PIPE_SHADER_VERTEX, 2, &cb);
+        if (cb.buffer)
+            pipe_resource_reference(&cb.buffer, NULL);
+    }
+
+    {
+        struct pipe_constant_buffer cb;
+
+        cb.buffer = NULL;
+        cb.buffer_offset = 0;
+        cb.buffer_size = 512 * sizeof(float[4]);
+        cb.user_buffer = state->vs_const_b;
+
+        if (!device->driver_caps.user_sw_cbufs) {
+            u_upload_data(device->constbuf_sw_uploader,
+                          0,
+                          cb.buffer_size,
+                          16,
+                          cb.user_buffer,
+                          &(cb.buffer_offset),
+                          &(cb.buffer));
+            u_upload_unmap(device->constbuf_sw_uploader);
+            cb.user_buffer = NULL;
+        }
+
+        pipe_sw->set_constant_buffer(pipe_sw, PIPE_SHADER_VERTEX, 3, &cb);
+        if (cb.buffer)
+            pipe_resource_reference(&cb.buffer, NULL);
+    }
+
+    {
+        struct pipe_constant_buffer cb;
+        const D3DVIEWPORT9 *vport = &device->state.viewport;
+        float viewport_data[8] = {(float)vport->Width * 0.5f,
+            (float)vport->Height * -0.5f, vport->MaxZ - vport->MinZ, 0.f,
+            (float)vport->Width * 0.5f + (float)vport->X,
+            (float)vport->Height * 0.5f + (float)vport->Y,
+            vport->MinZ, 0.f};
+
+        cb.buffer = NULL;
+        cb.buffer_offset = 0;
+        cb.buffer_size = 2 * sizeof(float[4]);
+        cb.user_buffer = viewport_data;
+
+        {
+            u_upload_data(device->constbuf_sw_uploader,
+                          0,
+                          cb.buffer_size,
+                          16,
+                          cb.user_buffer,
+                          &(cb.buffer_offset),
+                          &(cb.buffer));
+            u_upload_unmap(device->constbuf_sw_uploader);
+            cb.user_buffer = NULL;
+        }
+
+        pipe_sw->set_constant_buffer(pipe_sw, PIPE_SHADER_VERTEX, 4, &cb);
+        if (cb.buffer)
+            pipe_resource_reference(&cb.buffer, NULL);
+    }
+
+}
+
+void
+nine_state_prepare_draw_sw(struct NineDevice9 *device, struct NineVertexDeclaration9 *vdecl_out,
+                           int start_vertice, int num_vertices, struct pipe_stream_output_info *so)
+{
+    struct nine_state *state = &device->state;
+    bool programmable_vs = state->vs && !(state->vdecl && state->vdecl->position_t);
+    struct NineVertexShader9 *vs = programmable_vs ? device->state.vs : device->ff.vs;
+
+    assert(programmable_vs);
+
+    DBG("Preparing draw\n");
+    cso_set_vertex_shader_handle(device->cso_sw,
+                                 NineVertexShader9_GetVariantProcessVertices(vs, vdecl_out, so));
+    update_vertex_elements_sw(device);
+    update_vertex_buffers_sw(device, start_vertice, num_vertices);
+    update_vs_constants_sw(device);
+    DBG("Preparation succeeded\n");
+}
+
+void
+nine_state_after_draw_sw(struct NineDevice9 *device)
+{
+    struct nine_state_sw_internal *sw_internal = &device->state_sw_internal;
+    struct pipe_context *pipe = device->pipe;
+    struct pipe_context *pipe_sw = device->pipe_sw;
+    int i;
+
+    for (i = 0; i < 4; i++) {
+        pipe_sw->set_vertex_buffers(pipe_sw, i, 1, NULL);
+        if (sw_internal->transfers_so[i])
+            pipe->transfer_unmap(pipe, sw_internal->transfers_so[i]);
+        sw_internal->transfers_so[i] = NULL;
+    }
+}
+
+void
+nine_state_destroy_sw(struct NineDevice9 *device)
+{
+    (void) device;
+    /* Everything destroyed with cso */
+}
+
 /*
 static const DWORD nine_render_states_pixel[] =
 {