OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

nvfx: rework state_fb code to get rid of render temps

Browse Source 
This commit rewrites a lot of the state_fb code to support
rendering to targets not aligned to 64 byte.

This allows us to drop the render temporaries as unaligned
targets are the only use-case where they are really needed. The
temporaries code was used for a lot of things more, but apparently
those also work without temps.

There is one regression in piglit fbo-clear-formats, but this will
be fixed with the use of real hardware clears and doesn't matter in
practice as no real application tries to scissor clear a 2x2 pixel
render target.

Signed-off-by: Lucas Stach <dev@lynxeye.de>
This commit is contained in:
Lucas Stach
2012-01-10 12:41:02 +01:00
committed by Marek Olšák
parent 7f918683c4
commit f428ae6f72
4 changed files with 72 additions and 146 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/gallium/drivers/nvfx/nvfx_context.h
View File

@@ -247,7 +247,7 @@ extern void nvfx_vtxfmt_validate(struct nvfx_context *nvfx);
/* nvfx_fb.c */
extern int nvfx_framebuffer_prepare(struct nvfx_context *nvfx);
extern void nvfx_framebuffer_validate(struct nvfx_context *nvfx, unsigned prepare_result);
extern void nvfx_framebuffer_validate(struct nvfx_context *nvfx);
void
nvfx_framebuffer_relocate(struct nvfx_context *nvfx);

59
src/gallium/drivers/nvfx/nvfx_state_emit.c
View File

@@ -199,7 +199,6 @@ nvfx_state_validate_common(struct nvfx_context *nvfx)
struct nouveau_grobj *eng3d = nvfx->screen->eng3d;
unsigned dirty;
unsigned still_dirty = 0;
int new_fb_mode = -1; /* 1 = all swizzled, 0 = make all linear */
boolean flush_tex_cache = FALSE;
unsigned render_temps;
@@ -213,29 +212,12 @@ nvfx_state_validate_common(struct nvfx_context *nvfx)
nvfx->relocs_needed = NVFX_RELOCATE_ALL;
}
/* These can trigger use the of 3D engine to copy temporaries.
* That will recurse here and thus dirty all 3D state, so we need to this before anything else, and in a loop..
* This converges to having clean temps, then binding both fragtexes and framebuffers.
*/
while(nvfx->dirty & (NVFX_NEW_FB | NVFX_NEW_SAMPLER))
{
if(nvfx->dirty & NVFX_NEW_SAMPLER)
{
nvfx->dirty &=~ NVFX_NEW_SAMPLER;
nvfx_fragtex_validate(nvfx);
if(nvfx->dirty & NVFX_NEW_SAMPLER) {
nvfx->dirty &=~ NVFX_NEW_SAMPLER;
nvfx_fragtex_validate(nvfx);
// TODO: only set this if really necessary
flush_tex_cache = TRUE;
}
if(nvfx->dirty & NVFX_NEW_FB)
{
nvfx->dirty &=~ NVFX_NEW_FB;
new_fb_mode = nvfx_framebuffer_prepare(nvfx);
// TODO: make sure this doesn't happen, i.e. fbs have matching formats
assert(new_fb_mode >= 0);
}
// TODO: only set this if really necessary
flush_tex_cache = TRUE;
}
dirty = nvfx->dirty;
@@ -305,8 +287,8 @@ nvfx_state_validate_common(struct nvfx_context *nvfx)
}
}
if(new_fb_mode >= 0)
nvfx_framebuffer_validate(nvfx, new_fb_mode);
if(dirty & NVFX_NEW_FB)
nvfx_framebuffer_validate(nvfx);
if(dirty & NVFX_NEW_BLEND)
sb_emit(chan, nvfx->blend->sb, nvfx->blend->sb_len);
@@ -324,19 +306,19 @@ nvfx_state_validate_common(struct nvfx_context *nvfx)
etracer, neverball, foobillard, glest totally misrender
TODO: find the right fix
*/
if(dirty & (NVFX_NEW_VIEWPORT | NVFX_NEW_RAST | NVFX_NEW_ZSA) || (new_fb_mode >= 0))
if(dirty & (NVFX_NEW_VIEWPORT | NVFX_NEW_RAST | NVFX_NEW_ZSA))
{
nvfx_state_viewport_validate(nvfx);
}
if(dirty & NVFX_NEW_ZSA || (new_fb_mode >= 0))
if(dirty & (NVFX_NEW_ZSA | NVFX_NEW_FB))
{
BEGIN_RING(chan, eng3d, NV30_3D_DEPTH_WRITE_ENABLE, 2);
OUT_RING(chan, nvfx->framebuffer.zsbuf && nvfx->zsa->pipe.depth.writemask);
OUT_RING(chan, nvfx->framebuffer.zsbuf && nvfx->zsa->pipe.depth.enabled);
OUT_RING(chan, nvfx->framebuffer.zsbuf && nvfx->zsa->pipe.depth.enabled);
}
if((new_fb_mode >= 0) || (dirty & NVFX_NEW_FRAGPROG))
if(dirty & (NVFX_NEW_FRAGPROG | NVFX_NEW_FB))
nvfx_coord_conventions_validate(nvfx);
if(flush_tex_cache)
@@ -353,25 +335,6 @@ nvfx_state_validate_common(struct nvfx_context *nvfx)
nvfx->dirty = dirty & still_dirty;
render_temps = nvfx->state.render_temps;
if(render_temps)
{
for(int i = 0; i < nvfx->framebuffer.nr_cbufs; ++i)
{
if(render_temps & (1 << i)) {
assert(((struct nvfx_surface*)nvfx->framebuffer.cbufs[i])->temp);
util_dirty_surface_set_dirty(nvfx_surface_get_dirty_surfaces(nvfx->framebuffer.cbufs[i]),
(struct util_dirty_surface*)nvfx->framebuffer.cbufs[i]);
}
}
if(render_temps & 0x80) {
assert(((struct nvfx_surface*)nvfx->framebuffer.zsbuf)->temp);
util_dirty_surface_set_dirty(nvfx_surface_get_dirty_surfaces(nvfx->framebuffer.zsbuf),
(struct util_dirty_surface*)nvfx->framebuffer.zsbuf);
}
}
return TRUE;
}

155
src/gallium/drivers/nvfx/nvfx_state_fb.c
View File

@@ -3,96 +3,25 @@
#include "util/u_format.h"
static inline boolean
nvfx_surface_linear_renderable(struct pipe_surface* surf)
nvfx_surface_linear_target(struct pipe_surface* surf)
{
/* TODO: precompute this in nvfx_surface creation */
return (surf->texture->flags & NOUVEAU_RESOURCE_FLAG_LINEAR)
&& !(((struct nvfx_surface*)surf)->offset & 63)
&& !(((struct nvfx_surface*)surf)->pitch & 63);
return !!((struct nvfx_miptree*)surf->texture)->linear_pitch;
}
static inline boolean
nvfx_surface_swizzled_renderable(struct pipe_framebuffer_state* fb, struct pipe_surface* surf)
{
/* TODO: precompute this in nvfx_surface creation */
return !((struct nvfx_miptree*)surf->texture)->linear_pitch
&& (surf->texture->target != PIPE_TEXTURE_3D || u_minify(surf->texture->depth0, surf->u.tex.level) <= 1)
&& !(((struct nvfx_surface*)surf)->offset & 127)
&& (surf->width == fb->width)
&& (surf->height == fb->height)
&& !((struct nvfx_surface*)surf)->temp
&& (surf->format == PIPE_FORMAT_B8G8R8A8_UNORM || surf->format == PIPE_FORMAT_B8G8R8X8_UNORM || surf->format == PIPE_FORMAT_B5G6R5_UNORM);
}
static boolean
nvfx_surface_get_render_target(struct pipe_surface* surf, int all_swizzled, struct nvfx_render_target* target)
static void
nvfx_surface_get_render_target(struct pipe_surface* surf,
struct nvfx_render_target* target)
{
struct nvfx_surface* ns = (struct nvfx_surface*)surf;
if(!ns->temp)
{
target->bo = ((struct nvfx_miptree*)surf->texture)->base.bo;
target->offset = ns->offset;
target->pitch = align(ns->pitch, 64);
assert(target->pitch);
return FALSE;
}
else
{
target->offset = 0;
target->pitch = ns->temp->linear_pitch;
target->bo = ns->temp->base.bo;
assert(target->pitch);
return TRUE;
}
}
int
nvfx_framebuffer_prepare(struct nvfx_context *nvfx)
{
struct pipe_framebuffer_state *fb = &nvfx->framebuffer;
int i, color_format = 0, zeta_format = 0;
int all_swizzled = 1;
if(!nvfx->is_nv4x)
assert(fb->nr_cbufs <= 1);
else
assert(fb->nr_cbufs <= 4);
for (i = 0; i < fb->nr_cbufs; i++) {
if (color_format) {
if(color_format != fb->cbufs[i]->format)
return -1;
} else
color_format = fb->cbufs[i]->format;
if(!nvfx_surface_swizzled_renderable(fb, fb->cbufs[i]))
all_swizzled = 0;
}
if (fb->zsbuf) {
/* TODO: return FALSE if we have a format not supporting a depth buffer (e.g. r8); currently those are not supported at all */
if(!nvfx_surface_swizzled_renderable(fb, fb->zsbuf))
all_swizzled = 0;
if(all_swizzled && util_format_get_blocksize(color_format) != util_format_get_blocksize(zeta_format))
all_swizzled = 0;
}
for (i = 0; i < fb->nr_cbufs; i++) {
if(!((struct nvfx_surface*)fb->cbufs[i])->temp && !all_swizzled && !nvfx_surface_linear_renderable(fb->cbufs[i]))
nvfx_surface_create_temp(&nvfx->pipe, fb->cbufs[i]);
}
if(fb->zsbuf) {
if(!((struct nvfx_surface*)fb->zsbuf)->temp && !all_swizzled && !nvfx_surface_linear_renderable(fb->zsbuf))
nvfx_surface_create_temp(&nvfx->pipe, fb->zsbuf);
}
return all_swizzled;
target->bo = ((struct nvfx_miptree*)surf->texture)->base.bo;
target->offset = ns->offset;
target->pitch = align(ns->pitch, 64);
assert(target->pitch);
}
void
nvfx_framebuffer_validate(struct nvfx_context *nvfx, unsigned prepare_result)
nvfx_framebuffer_validate(struct nvfx_context *nvfx)
{
struct pipe_framebuffer_state *fb = &nvfx->framebuffer;
struct nouveau_channel *chan = nvfx->screen->base.channel;
@@ -102,37 +31,69 @@ nvfx_framebuffer_validate(struct nvfx_context *nvfx, unsigned prepare_result)
unsigned rt_flags = NOUVEAU_BO_RDWR | NOUVEAU_BO_VRAM;
unsigned w = fb->width;
unsigned h = fb->height;
int all_swizzled =1 , cb_format = 0;
/* do some sanity checks on the render target state and check if the targets
* are swizzled
*/
nvfx->is_nv4x ? assert(fb->nr_cbufs <= 4) : assert(fb->nr_cbufs <= 1);
if(fb->nr_cbufs && fb->zsbuf)
assert(util_format_get_blocksize(fb->cbufs[0]->format) ==
util_format_get_blocksize(fb->zsbuf->format));
for(i = 0; i < fb->nr_cbufs; i++) {
if(cb_format)
assert(cb_format == fb->cbufs[i]->format);
else
cb_format = fb->cbufs[i]->format;
if(nvfx_surface_linear_target(fb->cbufs[i]))
all_swizzled = 0;
}
if(fb->zsbuf && nvfx_surface_linear_target(fb->zsbuf))
all_swizzled = 0;
rt_enable = (NV30_3D_RT_ENABLE_COLOR0 << fb->nr_cbufs) - 1;
if (rt_enable & (NV30_3D_RT_ENABLE_COLOR1 |
NV40_3D_RT_ENABLE_COLOR2 | NV40_3D_RT_ENABLE_COLOR3))
if(rt_enable & (NV30_3D_RT_ENABLE_COLOR1 |
NV40_3D_RT_ENABLE_COLOR2 | NV40_3D_RT_ENABLE_COLOR3))
rt_enable |= NV30_3D_RT_ENABLE_MRT;
nvfx->state.render_temps = 0;
for (i = 0; i < fb->nr_cbufs; i++)
nvfx->state.render_temps |= nvfx_surface_get_render_target(fb->cbufs[i], prepare_result, &nvfx->hw_rt[i]) << i;
for(i = 0; i < fb->nr_cbufs; i++)
nvfx_surface_get_render_target(fb->cbufs[i], &nvfx->hw_rt[i]);
for(; i < 4; ++i)
nvfx->hw_rt[i].bo = NULL;
nvfx->hw_zeta.bo = NULL;
if (fb->zsbuf) {
nvfx->state.render_temps |= nvfx_surface_get_render_target(fb->zsbuf, prepare_result, &nvfx->hw_zeta) << 7;
assert(util_format_get_stride(fb->zsbuf->format, fb->width) <= nvfx->hw_zeta.pitch);
assert(nvfx->hw_zeta.offset + nvfx->hw_zeta.pitch * fb->height <= nvfx->hw_zeta.bo->size);
if(fb->zsbuf) {
nvfx_surface_get_render_target(fb->zsbuf, &nvfx->hw_zeta);
assert(util_format_get_stride(fb->zsbuf->format, fb->width) <=
nvfx->hw_zeta.pitch);
}
if (prepare_result) {
assert(!(fb->width & (fb->width - 1)) && !(fb->height & (fb->height - 1)));
if(all_swizzled) {
/* hardware rounds down render target offset to 64 bytes,
* but surfaces with a size of 2x2 pixel (16bpp) or 1x1 pixel (32bpp)
* have an unaligned start address, for those two important square
* formats we can hack around this limitation by adjusting the viewport
*/
if(nvfx->hw_rt[0].offset & 63) {
int delta = nvfx->hw_rt[0].offset & 63;
h = 2;
w = 16;
nvfx->viewport.translate[0] += delta /
(util_format_get_blocksize(fb->cbufs[0]->format) * 2);
nvfx->dirty |= NVFX_NEW_VIEWPORT;
}
rt_format = NV30_3D_RT_FORMAT_TYPE_SWIZZLED |
(util_logbase2(fb->width) << NV30_3D_RT_FORMAT_LOG2_WIDTH__SHIFT) |
(util_logbase2(fb->height) << NV30_3D_RT_FORMAT_LOG2_HEIGHT__SHIFT);
} else
(util_logbase2(w) << NV30_3D_RT_FORMAT_LOG2_WIDTH__SHIFT) |
(util_logbase2(h) << NV30_3D_RT_FORMAT_LOG2_HEIGHT__SHIFT);
} else {
rt_format = NV30_3D_RT_FORMAT_TYPE_LINEAR;
}
if(fb->nr_cbufs > 0) {
switch (fb->cbufs[0]->format) {

2
src/gallium/drivers/nvfx/nvfx_surface.c
View File

@@ -428,6 +428,8 @@ nvfx_surface_copy_temp(struct pipe_context* pipe, struct pipe_surface* surf, int
void
nvfx_surface_create_temp(struct pipe_context* pipe, struct pipe_surface* surf)
{
assert (0);
struct nvfx_surface* ns = (struct nvfx_surface*)surf;
struct pipe_resource template;
memset(&template, 0, sizeof(struct pipe_resource));