nvc0: implement resource_resolve

This commit is contained in:
Christoph Bumiller
2011-09-14 18:30:37 +02:00
parent 9cae933f4a
commit 7744e867b7
4 changed files with 532 additions and 5 deletions

View File

@@ -223,6 +223,9 @@ nvc0_screen_destroy(struct pipe_screen *pscreen)
}
screen->base.channel->user_private = NULL;
if (screen->blitctx)
FREE(screen->blitctx);
nouveau_bo_ref(NULL, &screen->text);
nouveau_bo_ref(NULL, &screen->tls);
nouveau_bo_ref(NULL, &screen->txc);
@@ -633,6 +636,9 @@ nvc0_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
screen->mm_VRAM_fe0 = nouveau_mm_create(dev, NOUVEAU_BO_VRAM, 0xfe0);
if (!nvc0_blitctx_create(screen))
goto fail;
nouveau_fence_new(&screen->base, &screen->base.fence.current, FALSE);
return pscreen;

View File

@@ -19,6 +19,8 @@ struct nvc0_context;
#define NVC0_SCREEN_RESIDENT_BO_COUNT 5
struct nvc0_blitctx;
struct nvc0_screen {
struct nouveau_screen base;
struct nouveau_winsys *nvws;
@@ -36,6 +38,8 @@ struct nvc0_screen {
struct nouveau_resource *text_heap;
struct nouveau_resource *lib_code; /* allocated from text_heap */
struct nvc0_blitctx *blitctx;
struct {
struct nouveau_bo *bo[NVC0_SCRATCH_NR_BUFFERS];
uint8_t *buf;
@@ -73,6 +77,8 @@ nvc0_screen(struct pipe_screen *screen)
return (struct nvc0_screen *)screen;
}
boolean nvc0_blitctx_create(struct nvc0_screen *);
void nvc0_screen_make_buffers_resident(struct nvc0_screen *);
int nvc0_screen_tic_alloc(struct nvc0_screen *, void *);

View File

@@ -62,12 +62,14 @@ nvc0_program_update_context_state(struct nvc0_context *nvc0,
static INLINE boolean
nvc0_program_validate(struct nvc0_context *nvc0, struct nvc0_program *prog)
{
if (prog->translated)
if (prog->res)
return TRUE;
prog->translated = nvc0_program_translate(prog);
if (!prog->translated)
return FALSE;
if (!prog->translated) {
prog->translated = nvc0_program_translate(prog);
if (!prog->translated)
return FALSE;
}
return nvc0_program_upload_code(nvc0, prog);
}

View File

@@ -421,14 +421,527 @@ nvc0_clear(struct pipe_context *pipe, unsigned buffers,
}
}
struct nvc0_blitctx
{
struct nvc0_screen *screen;
struct {
struct pipe_framebuffer_state fb;
struct nvc0_program *vp;
struct nvc0_program *tcp;
struct nvc0_program *tep;
struct nvc0_program *gp;
struct nvc0_program *fp;
unsigned num_textures[5];
unsigned num_samplers[5];
struct pipe_sampler_view *texture;
struct nv50_tsc_entry *sampler;
unsigned dirty;
unsigned clip_nr;
} saved;
struct nvc0_program vp;
struct nvc0_program fp;
struct nv50_tsc_entry sampler[2]; /* nearest, bilinear */
uint32_t fp_offset;
uint16_t color_mask;
uint8_t filter;
};
static void
nvc0_blitctx_make_vp(struct nvc0_blitctx *blit)
{
static const uint32_t code[] =
{
0xfff01c66, 0x06000080, /* vfetch b128 { $r0 $r1 $r2 $r3 } a[0x80] */
0xfff11c26, 0x06000090, /* vfetch b64 { $r4 $r5 } a[0x90]*/
0x03f01c66, 0x0a7e0070, /* export b128 o[0x70] { $r0 $r1 $r2 $r3 } */
0x13f01c26, 0x0a7e0080, /* export b64 o[0x80] { $r4 $r5 } */
0x00001de7, 0x80000000, /* exit */
};
blit->vp.type = PIPE_SHADER_VERTEX;
blit->vp.translated = TRUE;
blit->vp.code = (uint32_t *)code; /* no relocations -> no modification */
blit->vp.code_size = sizeof(code);
blit->vp.max_gpr = 6;
blit->vp.vp.edgeflag = PIPE_MAX_ATTRIBS;
blit->vp.hdr[0] = 0x00020461; /* vertprog magic */
blit->vp.hdr[4] = 0x000ff000; /* no outputs read */
blit->vp.hdr[6] = 0x0000003f; /* a[0x80], a[0x90] */
blit->vp.hdr[13] = 0x0003f000; /* o[0x70], o[0x80] */
}
static void
nvc0_blitctx_make_fp(struct nvc0_blitctx *blit)
{
static const uint32_t code[] = /* use nvc0dis */
{
/* 2 coords RGBA in, RGBA out, also for Z32_FLOAT(_S8X24_USCALED)
* NOTE:
* NVC0 doesn't like tex 3d on non-3d textures, but there should
* only be 2d and 2d-array MS resources anyway.
*/
0xfff01c00, 0xc07e0080,
0xfff05c00, 0xc07e0084,
0x00001c86, 0x8013c000,
0x00001de7, 0x80000000,
/* size: 0x70 + padding */
0, 0, 0, 0,
/* 2 coords ZS in, S encoded in R, Z encoded in GBA (8_UNORM)
* Setup float outputs in a way that conversion to UNORM yields the
* desired byte value.
*/
/* NOTE: need to repeat header */
0x00021462, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
0x80000000, 0x0000000f, 0x00000000, 0x00000000, 0x00000000,
0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
0x00000000, 0x00000000, 0x00000000, 0x0000000f, 0x00000000,
0xfff01c00, 0xc07e0080,
0xfff05c00, 0xc07e0084,
0x00001c86, 0x8010c000,
0xfc009c02, 0x312dffff,
0x05001c88,
0x09009e88,
0x04001c02, 0x30ee0202,
0xfc205c02, 0x38000003,
0x0020dc02, 0x3803fc00,
0x00209c02, 0x380003fc,
0x05005c88,
0x0d00dc88,
0x09209e04, 0x18000000,
0x04105c02, 0x30ee0202,
0x0430dc02, 0x30ce0202,
0x04209c02, 0x30de0202,
0x00001de7, 0x80000000,
/* size: 0xc8 + padding */
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/* 2 coords ZS in, Z encoded in RGB, S encoded in A (U8_UNORM) */
0x00021462, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
0x80000000, 0x0000000f, 0x00000000, 0x00000000, 0x00000000,
0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
0x00000000, 0x00000000, 0x00000000, 0x0000000f, 0x00000000,
0xfff01c00, 0xc07e0080,
0xfff05c00, 0xc07e0084,
0x00001c86, 0x8010c000,
0xfc009c02, 0x312dffff,
0x05081c88,
0x09009e88,
0x0430dc02, 0x30ee0202,
0xfc201c02, 0x38000003,
0x00205c02, 0x380003fc,
0x00209c02, 0x3803fc00,
0x01001c88,
0x05005c88,
0x09209e04, 0x18000000,
0x04001c02, 0x30ee0202,
0x04105c02, 0x30de0202,
0x04209c02, 0x30ce0202,
0x00001de7, 0x80000000,
};
blit->fp.type = PIPE_SHADER_FRAGMENT;
blit->fp.translated = TRUE;
blit->fp.code = (uint32_t *)code; /* const_cast */
blit->fp.code_size = sizeof(code);
blit->fp.max_gpr = 4;
blit->fp.hdr[0] = 0x00021462; /* fragprog magic */
blit->fp.hdr[5] = 0x80000000;
blit->fp.hdr[6] = 0x0000000f; /* 2 linear */
blit->fp.hdr[18] = 0x0000000f; /* 1 colour output */
}
static void
nvc0_blitctx_make_sampler(struct nvc0_blitctx *blit)
{
/* clamp to edge, min/max lod = 0, nearest filtering */
blit->sampler[0].id = -1;
blit->sampler[0].tsc[0] = 0x00000092;
blit->sampler[0].tsc[1] = 0x00000051;
/* clamp to edge, min/max lod = 0, bilinear filtering */
blit->sampler[1].id = -1;
blit->sampler[1].tsc[0] = 0x00000092;
blit->sampler[1].tsc[1] = 0x00000062;
}
/* Since shaders cannot export stencil, we cannot copy stencil values when
* rendering to ZETA, so we attach the ZS surface to a colour render target.
*/
static INLINE enum pipe_format
nvc0_blit_zeta_to_colour_format(enum pipe_format format)
{
switch (format) {
case PIPE_FORMAT_Z16_UNORM: return PIPE_FORMAT_R16_UNORM;
case PIPE_FORMAT_Z24_UNORM_S8_USCALED:
case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
case PIPE_FORMAT_Z24X8_UNORM: return PIPE_FORMAT_R8G8B8A8_UNORM;
case PIPE_FORMAT_Z32_FLOAT: return PIPE_FORMAT_R32_FLOAT;
case PIPE_FORMAT_Z32_FLOAT_S8X24_USCALED: return PIPE_FORMAT_R32G32_FLOAT;
default:
assert(0);
return PIPE_FORMAT_NONE;
}
}
static void
nvc0_blitctx_get_color_mask_and_fp(struct nvc0_blitctx *blit,
enum pipe_format format, uint8_t mask)
{
blit->color_mask = 0;
switch (format) {
case PIPE_FORMAT_Z24X8_UNORM:
case PIPE_FORMAT_Z24_UNORM_S8_USCALED:
blit->fp_offset = 0x180;
if (mask & PIPE_MASK_Z)
blit->color_mask |= 0x0111;
if (mask & PIPE_MASK_S)
blit->color_mask |= 0x1000;
break;
case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
blit->fp_offset = 0x80;
if (mask & PIPE_MASK_Z)
blit->color_mask |= 0x1110;
if (mask & PIPE_MASK_S)
blit->color_mask |= 0x0001;
break;
default:
blit->fp_offset = 0;
if (mask & (PIPE_MASK_R | PIPE_MASK_Z)) blit->color_mask |= 0x0001;
if (mask & (PIPE_MASK_G | PIPE_MASK_S)) blit->color_mask |= 0x0010;
if (mask & PIPE_MASK_B) blit->color_mask |= 0x0100;
if (mask & PIPE_MASK_A) blit->color_mask |= 0x1000;
break;
}
}
static void
nvc0_blit_set_dst(struct nvc0_context *nvc0,
struct pipe_resource *res, unsigned level, unsigned layer)
{
struct pipe_context *pipe = &nvc0->base.pipe;
struct pipe_surface templ;
if (util_format_is_depth_or_stencil(res->format))
templ.format = nvc0_blit_zeta_to_colour_format(res->format);
else
templ.format = res->format;
templ.usage = PIPE_USAGE_STREAM;
templ.u.tex.level = level;
templ.u.tex.first_layer = templ.u.tex.last_layer = layer;
nvc0->framebuffer.cbufs[0] = nvc0_miptree_surface_new(pipe, res, &templ);
nvc0->framebuffer.nr_cbufs = 1;
nvc0->framebuffer.zsbuf = NULL;
nvc0->framebuffer.width = nvc0->framebuffer.cbufs[0]->width;
nvc0->framebuffer.height = nvc0->framebuffer.cbufs[0]->height;
}
static INLINE void
nvc0_blit_fixup_tic_entry(struct pipe_sampler_view *view)
{
struct nv50_tic_entry *ent = nv50_tic_entry(view);
ent->tic[2] &= ~(1 << 31); /* scaled coordinates, ok with 3d textures ? */
/* magic: */
ent->tic[3] = 0x20000000; /* affects quality of near vertical edges in MS8 */
}
static void
nvc0_blit_set_src(struct nvc0_context *nvc0,
struct pipe_resource *res, unsigned level, unsigned layer)
{
struct pipe_context *pipe = &nvc0->base.pipe;
struct pipe_sampler_view templ;
int s;
templ.format = res->format;
templ.u.tex.first_layer = templ.u.tex.last_layer = layer;
templ.u.tex.first_level = templ.u.tex.last_level = level;
templ.swizzle_r = PIPE_SWIZZLE_RED;
templ.swizzle_g = PIPE_SWIZZLE_GREEN;
templ.swizzle_b = PIPE_SWIZZLE_BLUE;
templ.swizzle_a = PIPE_SWIZZLE_ALPHA;
nvc0->textures[4][0] = nvc0_create_sampler_view(pipe, res, &templ);
nvc0_blit_fixup_tic_entry(nvc0->textures[4][0]);
for (s = 0; s <= 3; ++s)
nvc0->num_textures[s] = 0;
nvc0->num_textures[4] = 1;
}
static void
nvc0_blitctx_prepare_state(struct nvc0_blitctx *blit)
{
struct nouveau_channel *chan = blit->screen->base.channel;
/* TODO: maybe make this a MACRO (if we need more logic) ? */
/* blend state */
BEGIN_RING(chan, RING_3D(COLOR_MASK(0)), 1);
OUT_RING (chan, blit->color_mask);
BEGIN_RING(chan, RING_3D(BLEND_ENABLE(0)), 1);
OUT_RING (chan, 0);
IMMED_RING(chan, RING_3D(LOGIC_OP_ENABLE), 0);
/* rasterizer state */
BEGIN_RING(chan, RING_3D(FRAG_COLOR_CLAMP_EN), 1);
OUT_RING (chan, 0);
IMMED_RING(chan, RING_3D(MULTISAMPLE_ENABLE), 0);
BEGIN_RING(chan, RING_3D(MSAA_MASK(0)), 4);
OUT_RING (chan, 0xffff);
OUT_RING (chan, 0xffff);
OUT_RING (chan, 0xffff);
OUT_RING (chan, 0xffff);
BEGIN_RING(chan, RING_3D(POLYGON_MODE_FRONT), 1);
OUT_RING (chan, NVC0_3D_POLYGON_MODE_FRONT_FILL);
BEGIN_RING(chan, RING_3D(POLYGON_MODE_BACK), 1);
OUT_RING (chan, NVC0_3D_POLYGON_MODE_BACK_FILL);
IMMED_RING(chan, RING_3D(POLYGON_SMOOTH_ENABLE), 0);
IMMED_RING(chan, RING_3D(POLYGON_OFFSET_FILL_ENABLE), 0);
IMMED_RING(chan, RING_3D(POLYGON_STIPPLE_ENABLE), 0);
IMMED_RING(chan, RING_3D(CULL_FACE_ENABLE), 0);
/* zsa state */
IMMED_RING(chan, RING_3D(DEPTH_TEST_ENABLE), 0);
IMMED_RING(chan, RING_3D(STENCIL_ENABLE), 0);
IMMED_RING(chan, RING_3D(ALPHA_TEST_ENABLE), 0);
/* transform feedback ? */
}
static void
nvc0_blitctx_pre_blit(struct nvc0_blitctx *blit, struct nvc0_context *nvc0)
{
int s;
blit->saved.fb.width = nvc0->framebuffer.width;
blit->saved.fb.height = nvc0->framebuffer.height;
blit->saved.fb.nr_cbufs = nvc0->framebuffer.nr_cbufs;
blit->saved.fb.cbufs[0] = nvc0->framebuffer.cbufs[0];
blit->saved.fb.zsbuf = nvc0->framebuffer.zsbuf;
blit->saved.vp = nvc0->vertprog;
blit->saved.tcp = nvc0->tctlprog;
blit->saved.tep = nvc0->tevlprog;
blit->saved.gp = nvc0->gmtyprog;
blit->saved.fp = nvc0->fragprog;
nvc0->vertprog = &blit->vp;
nvc0->fragprog = &blit->fp;
nvc0->tctlprog = NULL;
nvc0->tevlprog = NULL;
nvc0->gmtyprog = NULL;
blit->saved.clip_nr = nvc0->clip.nr;
nvc0->clip.nr = 0;
for (s = 0; s <= 4; ++s) {
blit->saved.num_textures[s] = nvc0->num_textures[s];
blit->saved.num_samplers[s] = nvc0->num_samplers[s];
}
blit->saved.texture = nvc0->textures[4][0];
blit->saved.sampler = nvc0->samplers[4][0];
nvc0->samplers[4][0] = &blit->sampler[blit->filter];
for (s = 0; s <= 3; ++s)
nvc0->num_samplers[s] = 0;
nvc0->num_samplers[4] = 1;
blit->saved.dirty = nvc0->dirty;
nvc0->dirty = NVC0_NEW_FRAMEBUFFER |
NVC0_NEW_VERTPROG | NVC0_NEW_FRAGPROG |
NVC0_NEW_TCTLPROG | NVC0_NEW_TEVLPROG | NVC0_NEW_GMTYPROG |
NVC0_NEW_TEXTURES | NVC0_NEW_SAMPLERS;
}
static void
nvc0_blitctx_post_blit(struct nvc0_context *nvc0, struct nvc0_blitctx *blit)
{
int s;
pipe_surface_reference(&nvc0->framebuffer.cbufs[0], NULL);
nvc0->framebuffer.width = blit->saved.fb.width;
nvc0->framebuffer.height = blit->saved.fb.height;
nvc0->framebuffer.nr_cbufs = blit->saved.fb.nr_cbufs;
nvc0->framebuffer.cbufs[0] = blit->saved.fb.cbufs[0];
nvc0->framebuffer.zsbuf = blit->saved.fb.zsbuf;
nvc0->vertprog = blit->saved.vp;
nvc0->tctlprog = blit->saved.tcp;
nvc0->tevlprog = blit->saved.tep;
nvc0->gmtyprog = blit->saved.gp;
nvc0->fragprog = blit->saved.fp;
nvc0->clip.nr = blit->saved.clip_nr;
pipe_sampler_view_reference(&nvc0->textures[4][0], NULL);
for (s = 0; s <= 4; ++s) {
nvc0->num_textures[s] = blit->saved.num_textures[s];
nvc0->num_samplers[s] = blit->saved.num_samplers[s];
}
nvc0->textures[4][0] = blit->saved.texture;
nvc0->samplers[4][0] = blit->saved.sampler;
nvc0->dirty = blit->saved.dirty |
(NVC0_NEW_FRAMEBUFFER | NVC0_NEW_SCISSOR | NVC0_NEW_SAMPLE_MASK |
NVC0_NEW_RASTERIZER | NVC0_NEW_ZSA | NVC0_NEW_BLEND |
NVC0_NEW_TEXTURES | NVC0_NEW_SAMPLERS |
NVC0_NEW_VERTPROG | NVC0_NEW_FRAGPROG |
NVC0_NEW_TCTLPROG | NVC0_NEW_TEVLPROG | NVC0_NEW_GMTYPROG);
}
static void
nvc0_resource_resolve(struct pipe_context *pipe,
const struct pipe_resolve_info *info)
{
struct nvc0_context *nvc0 = nvc0_context(pipe);
struct nvc0_screen *screen = nvc0->screen;
struct nvc0_blitctx *blit = screen->blitctx;
struct nouveau_channel *chan = screen->base.channel;
struct pipe_resource *src = info->src.res;
struct pipe_resource *dst = info->dst.res;
float x0, x1, y0, y1;
float x_range, y_range;
/* Would need more shader variants or, better, just change the TIC target.
* But no API creates 3D MS textures ...
*/
if (src->target == PIPE_TEXTURE_3D)
return;
nvc0_blitctx_get_color_mask_and_fp(blit, dst->format, info->mask);
blit->filter = util_format_is_depth_or_stencil(dst->format) ? 0 : 1;
nvc0_blitctx_pre_blit(blit, nvc0);
nvc0_blit_set_dst(nvc0, dst, info->dst.level, info->dst.layer);
nvc0_blit_set_src(nvc0, src, 0, info->src.layer);
nvc0_blitctx_prepare_state(blit);
nvc0_state_validate(nvc0, ~0, 36);
x_range =
(float)(info->src.x1 - info->src.x0) /
(float)(info->dst.x1 - info->dst.x0);
y_range =
(float)(info->src.y1 - info->src.y0) /
(float)(info->dst.y1 - info->dst.y0);
x0 = (float)info->src.x0 - x_range * (float)info->dst.x0;
y0 = (float)info->src.y0 - y_range * (float)info->dst.y0;
x1 = x0 + 16384.0f * x_range;
y1 = y0 + 16384.0f * y_range;
x0 *= (float)(1 << nv50_miptree(src)->ms_x);
x1 *= (float)(1 << nv50_miptree(src)->ms_x);
y0 *= (float)(1 << nv50_miptree(src)->ms_y);
y1 *= (float)(1 << nv50_miptree(src)->ms_y);
BEGIN_RING(chan, RING_3D(SP_START_ID(5)), 1);
OUT_RING (chan,
blit->fp.code_base + blit->fp_offset);
IMMED_RING(chan, RING_3D(VIEWPORT_TRANSFORM_EN), 0);
/* Draw a large triangle in screen coordinates covering the whole
* render target, with scissors defining the destination region.
* The vertex is supplied with non-normalized texture coordinates
* arranged in a way to yield the desired offset and scale.
*/
BEGIN_RING(chan, RING_3D(SCISSOR_HORIZ(0)), 2);
OUT_RING (chan, (info->dst.x1 << 16) | info->dst.x0);
OUT_RING (chan, (info->dst.y1 << 16) | info->dst.y0);
IMMED_RING(chan, RING_3D(VERTEX_BEGIN_GL),
NVC0_3D_VERTEX_BEGIN_GL_PRIMITIVE_TRIANGLES);
BEGIN_RING(chan, RING_3D(VTX_ATTR_DEFINE), 3);
OUT_RING (chan, 0x74201);
OUT_RINGf (chan, x0);
OUT_RINGf (chan, y0);
BEGIN_RING(chan, RING_3D(VTX_ATTR_DEFINE), 3);
OUT_RING (chan, 0x74200);
OUT_RINGf (chan, 0.0f);
OUT_RINGf (chan, 0.0f);
BEGIN_RING(chan, RING_3D(VTX_ATTR_DEFINE), 3);
OUT_RING (chan, 0x74201);
OUT_RINGf (chan, x1);
OUT_RINGf (chan, y0);
BEGIN_RING(chan, RING_3D(VTX_ATTR_DEFINE), 3);
OUT_RING (chan, 0x74200);
OUT_RINGf (chan, 16384 << nv50_miptree(dst)->ms_x);
OUT_RINGf (chan, 0.0f);
BEGIN_RING(chan, RING_3D(VTX_ATTR_DEFINE), 3);
OUT_RING (chan, 0x74201);
OUT_RINGf (chan, x0);
OUT_RINGf (chan, y1);
BEGIN_RING(chan, RING_3D(VTX_ATTR_DEFINE), 3);
OUT_RING (chan, 0x74200);
OUT_RINGf (chan, 0.0f);
OUT_RINGf (chan, 16384 << nv50_miptree(dst)->ms_y);
IMMED_RING(chan, RING_3D(VERTEX_END_GL), 0);
/* re-enable normally constant state */
IMMED_RING(chan, RING_3D(VIEWPORT_TRANSFORM_EN), 1);
nvc0_blitctx_post_blit(nvc0, blit);
}
boolean
nvc0_blitctx_create(struct nvc0_screen *screen)
{
screen->blitctx = CALLOC_STRUCT(nvc0_blitctx);
if (!screen->blitctx) {
NOUVEAU_ERR("failed to allocate blit context\n");
return FALSE;
}
screen->blitctx->screen = screen;
nvc0_blitctx_make_vp(screen->blitctx);
nvc0_blitctx_make_fp(screen->blitctx);
nvc0_blitctx_make_sampler(screen->blitctx);
screen->blitctx->color_mask = 0x1111;
return TRUE;
}
void
nvc0_init_surface_functions(struct nvc0_context *nvc0)
{
struct pipe_context *pipe = &nvc0->base.pipe;
pipe->resource_copy_region = nvc0_resource_copy_region;
pipe->resource_resolve = nvc0_resource_resolve;
pipe->clear_render_target = nvc0_clear_render_target;
pipe->clear_depth_stencil = nvc0_clear_depth_stencil;
}