vertexshader9.c revision 7ec681f3 
1/*
2 * Copyright 2011 Joakim Sindholt <opensource@zhasha.com>
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE. */
22
23#include "nine_helpers.h"
24#include "nine_shader.h"
25
26#include "vertexdeclaration9.h"
27#include "vertexshader9.h"
28
29#include "device9.h"
30#include "pipe/p_context.h"
31#include "cso_cache/cso_context.h"
32
33#define DBG_CHANNEL DBG_VERTEXSHADER
34
35HRESULT
36NineVertexShader9_ctor( struct NineVertexShader9 *This,
37                        struct NineUnknownParams *pParams,
38                        const DWORD *pFunction, void *cso )
39{
40    struct NineDevice9 *device;
41    struct nine_shader_info info;
42    struct pipe_context *pipe;
43    HRESULT hr;
44    unsigned i;
45
46    DBG("This=%p pParams=%p pFunction=%p cso=%p\n",
47        This, pParams, pFunction, cso);
48
49    hr = NineUnknown_ctor(&This->base, pParams);
50    if (FAILED(hr))
51        return hr;
52
53    if (cso) {
54        This->ff_cso = cso;
55        return D3D_OK;
56    }
57
58    device = This->base.device;
59
60    info.type = PIPE_SHADER_VERTEX;
61    info.byte_code = pFunction;
62    info.const_i_base = NINE_CONST_I_BASE(device->max_vs_const_f) / 16;
63    info.const_b_base = NINE_CONST_B_BASE(device->max_vs_const_f) / 16;
64    info.sampler_mask_shadow = 0x0;
65    info.fetch4 = 0x0;
66    info.sampler_ps1xtypes = 0x0;
67    info.fog_enable = 0;
68    info.point_size_min = 0;
69    info.point_size_max = 0;
70    info.add_constants_defs.c_combination = NULL;
71    info.add_constants_defs.int_const_added = NULL;
72    info.add_constants_defs.bool_const_added = NULL;
73    info.swvp_on = !!(device->params.BehaviorFlags & D3DCREATE_SOFTWARE_VERTEXPROCESSING);
74    info.process_vertices = false;
75
76    pipe = nine_context_get_pipe_acquire(device);
77    hr = nine_translate_shader(device, &info, pipe);
78    if (hr == D3DERR_INVALIDCALL &&
79        (device->params.BehaviorFlags & D3DCREATE_MIXED_VERTEXPROCESSING)) {
80        /* Retry with a swvp shader. It will require swvp to be on. */
81        info.swvp_on = true;
82        hr = nine_translate_shader(device, &info, pipe);
83    }
84    nine_context_get_pipe_release(device);
85    if (hr == D3DERR_INVALIDCALL)
86        ERR("Encountered buggy shader\n");
87    if (FAILED(hr))
88        return hr;
89    This->byte_code.version = info.version;
90    This->swvp_only = info.swvp_on;
91
92    This->byte_code.tokens = mem_dup(pFunction, info.byte_size);
93    if (!This->byte_code.tokens)
94        return E_OUTOFMEMORY;
95    This->byte_code.size = info.byte_size;
96
97    This->variant.cso = info.cso;
98    This->variant.const_ranges = info.const_ranges;
99    This->variant.const_used_size = info.const_used_size;
100    This->last_cso = info.cso;
101    This->last_const_ranges = info.const_ranges;
102    This->last_const_used_size = info.const_used_size;
103    This->last_key = (uint32_t) (info.swvp_on << 9);
104
105    This->lconstf = info.lconstf;
106    This->sampler_mask = info.sampler_mask;
107    This->position_t = info.position_t;
108    This->point_size = info.point_size;
109
110    memcpy(This->int_slots_used, info.int_slots_used, sizeof(This->int_slots_used));
111    memcpy(This->bool_slots_used, info.bool_slots_used, sizeof(This->bool_slots_used));
112
113    This->const_int_slots = info.const_int_slots;
114    This->const_bool_slots = info.const_bool_slots;
115
116    This->c_combinations = NULL;
117
118    for (i = 0; i < info.num_inputs && i < ARRAY_SIZE(This->input_map); ++i)
119        This->input_map[i].ndecl = info.input_map[i];
120    This->num_inputs = i;
121
122    return D3D_OK;
123}
124
125void
126NineVertexShader9_dtor( struct NineVertexShader9 *This )
127{
128    DBG("This=%p\n", This);
129
130    if (This->base.device) {
131        struct pipe_context *pipe = nine_context_get_pipe_multithread(This->base.device);
132        struct nine_shader_variant *var = &This->variant;
133        struct nine_shader_variant_so *var_so = &This->variant_so;
134
135        do {
136            if (var->cso) {
137                if (This->base.device->context.cso_shader.vs == var->cso)
138                    pipe->bind_vs_state(pipe, NULL);
139                pipe->delete_vs_state(pipe, var->cso);
140                FREE(var->const_ranges);
141            }
142            var = var->next;
143        } while (var);
144
145        while (var_so && var_so->vdecl) {
146            if (var_so->cso) {
147                This->base.device->pipe_sw->delete_vs_state(This->base.device->pipe_sw, var_so->cso);
148            }
149            var_so = var_so->next;
150        }
151
152        if (This->ff_cso) {
153            if (This->ff_cso == This->base.device->context.cso_shader.vs)
154                pipe->bind_vs_state(pipe, NULL);
155            pipe->delete_vs_state(pipe, This->ff_cso);
156        }
157    }
158    nine_shader_variants_free(&This->variant);
159    nine_shader_variants_so_free(&This->variant_so);
160
161    nine_shader_constant_combination_free(This->c_combinations);
162
163    FREE((void *)This->byte_code.tokens); /* const_cast */
164
165    FREE(This->lconstf.data);
166    FREE(This->lconstf.ranges);
167
168    NineUnknown_dtor(&This->base);
169}
170
171HRESULT NINE_WINAPI
172NineVertexShader9_GetFunction( struct NineVertexShader9 *This,
173                               void *pData,
174                               UINT *pSizeOfData )
175{
176    user_assert(pSizeOfData, D3DERR_INVALIDCALL);
177
178    if (!pData) {
179        *pSizeOfData = This->byte_code.size;
180        return D3D_OK;
181    }
182    user_assert(*pSizeOfData >= This->byte_code.size, D3DERR_INVALIDCALL);
183
184    memcpy(pData, This->byte_code.tokens, This->byte_code.size);
185
186    return D3D_OK;
187}
188
189void *
190NineVertexShader9_GetVariant( struct NineVertexShader9 *This,
191                              unsigned **const_ranges,
192                              unsigned *const_used_size )
193{
194    /* GetVariant is called from nine_context, thus we can
195     * get pipe directly */
196    struct pipe_context *pipe = This->base.device->context.pipe;
197    void *cso;
198    uint64_t key;
199
200    key = This->next_key;
201    if (key == This->last_key) {
202        *const_ranges = This->last_const_ranges;
203        *const_used_size = This->last_const_used_size;
204        return This->last_cso;
205    }
206
207    cso = nine_shader_variant_get(&This->variant, const_ranges, const_used_size, key);
208    if (!cso) {
209        struct NineDevice9 *device = This->base.device;
210        struct nine_shader_info info;
211        HRESULT hr;
212
213        info.type = PIPE_SHADER_VERTEX;
214        info.const_i_base = NINE_CONST_I_BASE(device->max_vs_const_f) / 16;
215        info.const_b_base = NINE_CONST_B_BASE(device->max_vs_const_f) / 16;
216        info.byte_code = This->byte_code.tokens;
217        info.sampler_mask_shadow = key & 0xf;
218        info.fetch4 = 0x0;
219        info.fog_enable = device->context.rs[D3DRS_FOGENABLE];
220        info.point_size_min = asfloat(device->context.rs[D3DRS_POINTSIZE_MIN]);
221        info.point_size_max = asfloat(device->context.rs[D3DRS_POINTSIZE_MAX]);
222        info.add_constants_defs.c_combination =
223            nine_shader_constant_combination_get(This->c_combinations, (key >> 16) & 0xff);
224        info.add_constants_defs.int_const_added = &This->int_slots_used;
225        info.add_constants_defs.bool_const_added = &This->bool_slots_used;
226        info.swvp_on = device->context.swvp;
227        info.process_vertices = false;
228
229        hr = nine_translate_shader(This->base.device, &info, pipe);
230        if (FAILED(hr))
231            return NULL;
232        nine_shader_variant_add(&This->variant, key, info.cso,
233                                info.const_ranges, info.const_used_size);
234        cso = info.cso;
235        *const_ranges = info.const_ranges;
236        *const_used_size = info.const_used_size;
237    }
238
239    This->last_key = key;
240    This->last_cso = cso;
241    This->last_const_ranges = *const_ranges;
242    This->last_const_used_size = *const_used_size;
243
244    return cso;
245}
246
247void *
248NineVertexShader9_GetVariantProcessVertices( struct NineVertexShader9 *This,
249                                             struct NineVertexDeclaration9 *vdecl_out,
250                                             struct pipe_stream_output_info *so )
251{
252    struct nine_shader_info info;
253    HRESULT hr;
254    void *cso;
255
256    cso = nine_shader_variant_so_get(&This->variant_so, vdecl_out, so);
257    if (cso)
258        return cso;
259
260    info.type = PIPE_SHADER_VERTEX;
261    info.const_i_base = 0;
262    info.const_b_base = 0;
263    info.byte_code = This->byte_code.tokens;
264    info.sampler_mask_shadow = 0;
265    info.fetch4 = 0x0;
266    info.fog_enable = false;
267    info.point_size_min = 0;
268    info.point_size_max = 0;
269    info.add_constants_defs.c_combination = NULL;
270    info.add_constants_defs.int_const_added = NULL;
271    info.add_constants_defs.bool_const_added = NULL;
272    info.swvp_on = true;
273    info.vdecl_out = vdecl_out;
274    info.process_vertices = true;
275    hr = nine_translate_shader(This->base.device, &info, This->base.device->pipe_sw);
276    if (FAILED(hr))
277        return NULL;
278    *so = info.so;
279    nine_shader_variant_so_add(&This->variant_so, vdecl_out, so, info.cso);
280    return info.cso;
281}
282
283IDirect3DVertexShader9Vtbl NineVertexShader9_vtable = {
284    (void *)NineUnknown_QueryInterface,
285    (void *)NineUnknown_AddRef,
286    (void *)NineUnknown_Release,
287    (void *)NineUnknown_GetDevice,
288    (void *)NineVertexShader9_GetFunction
289};
290
291static const GUID *NineVertexShader9_IIDs[] = {
292    &IID_IDirect3DVertexShader9,
293    &IID_IUnknown,
294    NULL
295};
296
297HRESULT
298NineVertexShader9_new( struct NineDevice9 *pDevice,
299                       struct NineVertexShader9 **ppOut,
300                       const DWORD *pFunction, void *cso )
301{
302    if (cso) {
303        NINE_DEVICE_CHILD_BIND_NEW(VertexShader9, ppOut, pDevice, pFunction, cso);
304    } else {
305        NINE_DEVICE_CHILD_NEW(VertexShader9, ppOut, pDevice, pFunction, cso);
306    }
307}
308