Merge remote branch 'origin/master' into nv50-compiler
[mesa.git] / src / gallium / drivers / nv50 / nv50_vbo.c
index 50472868063f5bfa32c983300e8a5f61be2d98d8..d41a59d05dbd10e1aa9e689c3ec35664ada40385 100644 (file)
 #include "pipe/p_state.h"
 #include "util/u_inlines.h"
 #include "util/u_format.h"
+#include "util/u_split_prim.h"
 
-#include "nouveau/nouveau_util.h"
 #include "nv50_context.h"
-
-static INLINE uint32_t
-nv50_vbo_type_to_hw(enum pipe_format format)
-{
-       const struct util_format_description *desc;
-
-       desc = util_format_description(format);
-       assert(desc);
-
-       switch (desc->channel[0].type) {
-       case UTIL_FORMAT_TYPE_FLOAT:
-               return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_FLOAT;
-       case UTIL_FORMAT_TYPE_UNSIGNED:
-               if (desc->channel[0].normalized) {
-                       return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_UNORM;
-               }
-               return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_USCALED;
-       case UTIL_FORMAT_TYPE_SIGNED:
-               if (desc->channel[0].normalized) {
-                       return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_SNORM;
-               }
-               return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_SSCALED;
-       /*
-       case PIPE_FORMAT_TYPE_UINT:
-               return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_UINT;
-       case PIPE_FORMAT_TYPE_SINT:
-               return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_SINT; */
-       default:
-               return 0;
-       }
-}
-
-static INLINE uint32_t
-nv50_vbo_size_to_hw(unsigned size, unsigned nr_c)
-{
-       static const uint32_t hw_values[] = {
-               0, 0, 0, 0,
-               NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8,
-               NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8_8,
-               NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8_8_8,
-               NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8_8_8_8,
-               NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16,
-               NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16_16,
-               NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16_16_16,
-               NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16_16_16_16,
-               0, 0, 0, 0,
-               NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32,
-               NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32_32,
-               NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32_32_32,
-               NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32_32_32_32 };
-
-       /* we'd also have R11G11B10 and R10G10B10A2 */
-
-       assert(nr_c > 0 && nr_c <= 4);
-
-       if (size > 32)
-               return 0;
-       size >>= (3 - 2);
-
-       return hw_values[size + (nr_c - 1)];
-}
-
-static INLINE uint32_t
-nv50_vbo_vtxelt_to_hw(struct pipe_vertex_element *ve)
-{
-       uint32_t hw_type, hw_size;
-       enum pipe_format pf = ve->src_format;
-       const struct util_format_description *desc;
-       unsigned size, nr_components;
-
-       desc = util_format_description(pf);
-       assert(desc);
-
-       size = util_format_get_component_bits(pf, UTIL_FORMAT_COLORSPACE_RGB, 0);
-       nr_components = util_format_get_nr_components(pf);
-
-       hw_type = nv50_vbo_type_to_hw(pf);
-       hw_size = nv50_vbo_size_to_hw(size, nr_components);
-
-       if (!hw_type || !hw_size) {
-               NOUVEAU_ERR("unsupported vbo format: %s\n", util_format_name(pf));
-               abort();
-               return 0x24e80000;
-       }
-
-       if (desc->swizzle[0] == UTIL_FORMAT_SWIZZLE_Z) /* BGRA */
-               hw_size |= (1 << 31); /* no real swizzle bits :-( */
-
-       return (hw_type | hw_size);
-}
+#include "nv50_resource.h"
 
 struct instance {
        struct nouveau_bo *bo;
@@ -139,7 +50,7 @@ instance_init(struct nv50_context *nv50, struct instance *a, unsigned first)
                if (a[i].divisor) {
                        vb = &nv50->vtxbuf[ve->vertex_buffer_index];
 
-                       a[i].bo = nouveau_bo(vb->buffer);
+                       a[i].bo = nv50_resource(vb->buffer)->bo;
                        a[i].stride = vb->stride;
                        a[i].step = first % a[i].divisor;
                        a[i].delta = vb->buffer_offset + ve->src_offset +
@@ -172,7 +83,7 @@ instance_step(struct nv50_context *nv50, struct instance *a)
        }
 }
 
-void
+static void
 nv50_draw_arrays_instanced(struct pipe_context *pipe,
                           unsigned mode, unsigned start, unsigned count,
                           unsigned startInstance, unsigned instanceCount)
@@ -188,7 +99,7 @@ nv50_draw_arrays_instanced(struct pipe_context *pipe,
                return;
 
        if (nv50->vbo_fifo) {
-               nv50_push_elements_instanced(pipe, NULL, 0, mode, start,
+               nv50_push_elements_instanced(pipe, NULL, 0, 0, mode, start,
                                             count, startInstance,
                                             instanceCount);
                return;
@@ -219,13 +130,6 @@ nv50_draw_arrays_instanced(struct pipe_context *pipe,
        }
 }
 
-void
-nv50_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start,
-                unsigned count)
-{
-       nv50_draw_arrays_instanced(pipe, mode, start, count, 0, 1);
-}
-
 struct inline_ctx {
        struct nv50_context *nv50;
        void *map;
@@ -307,17 +211,17 @@ inline_edgeflag(void *priv, boolean enabled)
 
 static void
 nv50_draw_elements_inline(struct pipe_context *pipe,
-                         struct pipe_buffer *indexBuffer, unsigned indexSize,
+                         struct pipe_resource *indexBuffer, unsigned indexSize,
                          unsigned mode, unsigned start, unsigned count,
                          unsigned startInstance, unsigned instanceCount)
 {
-       struct pipe_screen *pscreen = pipe->screen;
        struct nv50_context *nv50 = nv50_context(pipe);
        struct nouveau_channel *chan = nv50->screen->tesla->channel;
        struct nouveau_grobj *tesla = nv50->screen->tesla;
+       struct pipe_transfer *transfer;
        struct instance a[16];
        struct inline_ctx ctx;
-       struct u_split_prim s;
+       struct util_split_prim s;
        boolean nzi = FALSE;
        unsigned overhead;
 
@@ -337,7 +241,7 @@ nv50_draw_elements_inline(struct pipe_context *pipe,
        s.edge = inline_edgeflag;
 
        ctx.nv50 = nv50;
-       ctx.map = pipe_buffer_map(pscreen, indexBuffer, PIPE_BUFFER_USAGE_CPU_READ);
+       ctx.map = pipe_buffer_map(pipe, indexBuffer, PIPE_TRANSFER_READ, &transfer);
        assert(ctx.map);
        if (!ctx.map)
                return;
@@ -353,7 +257,7 @@ nv50_draw_elements_inline(struct pipe_context *pipe,
                unsigned max_verts;
                boolean done;
 
-               u_split_prim_init(&s, mode, start, count);
+               util_split_prim_init(&s, mode, start, count);
                do {
                        if (AVAIL_RING(chan) < (overhead + 6)) {
                                FIRE_RING(chan);
@@ -372,7 +276,7 @@ nv50_draw_elements_inline(struct pipe_context *pipe,
 
                        BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
                        OUT_RING  (chan, nv50_prim(s.mode) | (nzi ? (1<<28) : 0));
-                       done = u_split_prim_next(&s, max_verts);
+                       done = util_split_prim_next(&s, max_verts);
                        BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
                        OUT_RING  (chan, 0);
                } while (!done);
@@ -380,13 +284,13 @@ nv50_draw_elements_inline(struct pipe_context *pipe,
                nzi = TRUE;
        }
 
-       pipe_buffer_unmap(pscreen, indexBuffer);
+       pipe_buffer_unmap(pipe, indexBuffer, transfer);
 }
 
-void
+static void
 nv50_draw_elements_instanced(struct pipe_context *pipe,
-                            struct pipe_buffer *indexBuffer,
-                            unsigned indexSize,
+                            struct pipe_resource *indexBuffer,
+                            unsigned indexSize, int indexBias,
                             unsigned mode, unsigned start, unsigned count,
                             unsigned startInstance, unsigned instanceCount)
 {
@@ -402,11 +306,16 @@ nv50_draw_elements_instanced(struct pipe_context *pipe,
 
        if (nv50->vbo_fifo) {
                nv50_push_elements_instanced(pipe, indexBuffer, indexSize,
-                                            mode, start, count, startInstance,
-                                            instanceCount);
+                                            indexBias, mode, start, count,
+                                            startInstance, instanceCount);
                return;
-       } else
-       if (!(indexBuffer->usage & PIPE_BUFFER_USAGE_INDEX) || indexSize == 1) {
+       }
+
+       /* indices are uint32 internally, so large indexBias means negative */
+       BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_BASE, 1);
+       OUT_RING  (chan, indexBias);
+
+       if (!nv50_resource_mapped_by_gpu(indexBuffer) || indexSize == 1) {
                nv50_draw_elements_inline(pipe, indexBuffer, indexSize,
                                          mode, start, count, startInstance,
                                          instanceCount);
@@ -431,7 +340,8 @@ nv50_draw_elements_instanced(struct pipe_context *pipe,
                if (indexSize == 4) {
                        BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32 | 0x30000, 0);
                        OUT_RING  (chan, count);
-                       nouveau_pushbuf_submit(chan, nouveau_bo(indexBuffer),
+                       nouveau_pushbuf_submit(chan, 
+                                              nv50_resource(indexBuffer)->bo,
                                               start << 2, count << 2);
                } else
                if (indexSize == 2) {
@@ -443,7 +353,8 @@ nv50_draw_elements_instanced(struct pipe_context *pipe,
                        OUT_RING  (chan, ((start & 1) << 31) | count);
                        BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16 | 0x30000, 0);
                        OUT_RING  (chan, dwords);
-                       nouveau_pushbuf_submit(chan, nouveau_bo(indexBuffer),
+                       nouveau_pushbuf_submit(chan,
+                                              nv50_resource(indexBuffer)->bo,
                                               vb_start << 1, dwords << 2);
                        BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16_SETUP, 1);
                        OUT_RING  (chan, 0);
@@ -456,12 +367,34 @@ nv50_draw_elements_instanced(struct pipe_context *pipe,
 }
 
 void
-nv50_draw_elements(struct pipe_context *pipe,
-                  struct pipe_buffer *indexBuffer, unsigned indexSize,
-                  unsigned mode, unsigned start, unsigned count)
+nv50_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
 {
-       nv50_draw_elements_instanced(pipe, indexBuffer, indexSize,
-                                    mode, start, count, 0, 1);
+       struct nv50_context *nv50 = nv50_context(pipe);
+
+       if (info->indexed && nv50->idxbuf.buffer) {
+               unsigned offset;
+
+               assert(nv50->idxbuf.offset % nv50->idxbuf.index_size == 0);
+               offset = nv50->idxbuf.offset / nv50->idxbuf.index_size;
+
+               nv50_draw_elements_instanced(pipe,
+                                            nv50->idxbuf.buffer,
+                                            nv50->idxbuf.index_size,
+                                            info->index_bias,
+                                            info->mode,
+                                            info->start + offset,
+                                            info->count,
+                                            info->start_instance,
+                                            info->instance_count);
+       }
+       else {
+               nv50_draw_arrays_instanced(pipe,
+                                          info->mode,
+                                          info->start,
+                                          info->count,
+                                          info->start_instance,
+                                          info->instance_count);
+       }
 }
 
 static INLINE boolean
@@ -473,7 +406,7 @@ nv50_vbo_static_attrib(struct nv50_context *nv50, unsigned attrib,
 {
        struct nouveau_stateobj *so;
        struct nouveau_grobj *tesla = nv50->screen->tesla;
-       struct nouveau_bo *bo = nouveau_bo(vb->buffer);
+       struct nouveau_bo *bo = nv50_resource(vb->buffer)->bo;
        float v[4];
        int ret;
        unsigned nr_components = util_format_get_nr_components(ve->src_format);
@@ -510,7 +443,7 @@ nv50_vbo_static_attrib(struct nv50_context *nv50, unsigned attrib,
                so_data  (so, fui(v[1]));
                break;
        case 1:
-               if (attrib == nv50->vertprog->cfg.edgeflag_in) {
+               if (attrib == nv50->vertprog->vp.edgeflag) {
                        so_method(so, tesla, NV50TCL_EDGEFLAG_ENABLE, 1);
                        so_data  (so, v[0] ? 1 : 0);
                }
@@ -531,11 +464,8 @@ nv50_vtxelt_construct(struct nv50_vtxelt_stateobj *cso)
 {
        unsigned i;
 
-       for (i = 0; i < cso->num_elements; ++i) {
-               struct pipe_vertex_element *ve = &cso->pipe[i];
-
-               cso->hw[i] = nv50_vbo_vtxelt_to_hw(ve);
-       }
+       for (i = 0; i < cso->num_elements; ++i)
+               cso->hw[i] = nv50_format_table[cso->pipe[i].src_format].vtx;
 }
 
 struct nouveau_stateobj *
@@ -551,12 +481,12 @@ nv50_vbo_validate(struct nv50_context *nv50)
 
        nv50->vbo_fifo = 0;
        if (nv50->screen->force_push ||
-           nv50->vertprog->cfg.edgeflag_in < 16)
+           nv50->vertprog->vp.edgeflag < 16)
                nv50->vbo_fifo = 0xffff;
 
        for (i = 0; i < nv50->vtxbuf_nr; i++) {
                if (nv50->vtxbuf[i].stride &&
-                   !(nv50->vtxbuf[i].buffer->usage & PIPE_BUFFER_USAGE_VERTEX))
+                   !nv50_resource_mapped_by_gpu(nv50->vtxbuf[i].buffer))
                        nv50->vbo_fifo = 0xffff;
        }
 
@@ -571,7 +501,7 @@ nv50_vbo_validate(struct nv50_context *nv50)
                struct pipe_vertex_element *ve = &nv50->vtxelt->pipe[i];
                struct pipe_vertex_buffer *vb =
                        &nv50->vtxbuf[ve->vertex_buffer_index];
-               struct nouveau_bo *bo = nouveau_bo(vb->buffer);
+               struct nouveau_bo *bo = nv50_resource(vb->buffer)->bo;
                uint32_t hw = nv50->vtxelt->hw[i];
 
                if (!vb->stride &&
@@ -608,10 +538,10 @@ nv50_vbo_validate(struct nv50_context *nv50)
 
                /* vertex array limits */
                so_method(vtxbuf, tesla, NV50TCL_VERTEX_ARRAY_LIMIT_HIGH(i), 2);
-               so_reloc (vtxbuf, bo, vb->buffer->size - 1,
+               so_reloc (vtxbuf, bo, vb->buffer->width0 - 1,
                          NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD |
                          NOUVEAU_BO_HIGH, 0, 0);
-               so_reloc (vtxbuf, bo, vb->buffer->size - 1,
+               so_reloc (vtxbuf, bo, vb->buffer->width0 - 1,
                          NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD |
                          NOUVEAU_BO_LOW, 0, 0);
        }