ShuriZma
/
xenia-canary
mirror of https://github.com/xenia-canary/xenia-canary.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Partial success on predicated tiling (depth buffer still bad).

This commit is contained in:
Ben Vanik 2015-03-21 22:32:23 -07:00
parent 7c3225ee41
commit fa58eaa317
4 changed files with 77 additions and 50 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/xenia/gpu/gl4/blitter.cc
View File

@ -229,7 +229,7 @@ void Blitter::Draw(GLuint src_texture, Rect2D src_rect, Rect2D dest_rect,
break;
}
glViewport(0, 0, dest_rect.width, dest_rect.height);
glViewport(dest_rect.x, dest_rect.y, dest_rect.width, dest_rect.height);
// TODO(benvanik): avoid this?
GLint src_texture_width;

63
src/xenia/gpu/gl4/command_processor.cc
View File

@ -578,16 +578,12 @@ void CommandProcessor::IssueSwap() {
// Guess frontbuffer dimensions.
// Command buffer seems to set these right before the XE_SWAP.
// uint32_t window_scissor_tl = regs[XE_GPU_REG_PA_SC_WINDOW_SCISSOR_TL].u32;
// uint32_t window_scissor_br = regs[XE_GPU_REG_PA_SC_WINDOW_SCISSOR_BR].u32;
// swap_params.x = window_scissor_tl & 0x7FFF;
// swap_params.y = (window_scissor_tl >> 16) & 0x7FFF;
// swap_params.width = window_scissor_br & 0x7FFF - swap_params.x;
// swap_params.height = (window_scissor_br >> 16) & 0x7FFF - swap_params.y;
swap_params.x = 0;
swap_params.y = 0;
swap_params.width = 1280;
swap_params.height = 720;
uint32_t window_scissor_tl = regs[XE_GPU_REG_PA_SC_WINDOW_SCISSOR_TL].u32;
uint32_t window_scissor_br = regs[XE_GPU_REG_PA_SC_WINDOW_SCISSOR_BR].u32;
swap_params.x = window_scissor_tl & 0x7FFF;
swap_params.y = (window_scissor_tl >> 16) & 0x7FFF;
swap_params.width = window_scissor_br & 0x7FFF - swap_params.x;
swap_params.height = (window_scissor_br >> 16) & 0x7FFF - swap_params.y;
PrepareForWait();
swap_handler_(swap_params);
@ -2208,6 +2204,7 @@ CommandProcessor::UpdateStatus CommandProcessor::UpdateDepthStencilState() {
} else {
glDisable(GL_DEPTH_TEST);
}
// glDisable(GL_DEPTH_TEST);
// A2XX_RB_DEPTHCONTROL_Z_WRITE_ENABLE
glDepthMask((regs.rb_depthcontrol & 0x00000004) ? GL_TRUE : GL_FALSE);
// A2XX_RB_DEPTHCONTROL_EARLY_Z_ENABLE
@ -2631,17 +2628,14 @@ bool CommandProcessor::IssueCopy() {
// glPixelStorei(GL_PACK_ROW_LENGTH, 0);
// glPixelStorei(GL_PACK_IMAGE_HEIGHT, 0);
// Destination pointer in guest memory.
// We have GL throw bytes directly into it.
// TODO(benvanik): copy to staging texture then PBO back?
void* ptr = membase_ + GpuToCpu(copy_dest_base);
// TODO(benvanik): any way to scissor this? a200 has:
// REG_A2XX_RB_COPY_DEST_OFFSET = A2XX_RB_COPY_DEST_OFFSET_X(tile->xoff) |
// A2XX_RB_COPY_DEST_OFFSET_Y(tile->yoff);
// but I can't seem to find something similar.
uint32_t width = copy_dest_pitch;
uint32_t height = copy_dest_height;
uint32_t dest_logical_width = copy_dest_pitch;
uint32_t dest_logical_height = copy_dest_height;
uint32_t dest_block_width = poly::round_up(dest_logical_width, 32);
uint32_t dest_block_height = poly::round_up(dest_logical_height, 32);
uint32_t window_offset = regs[XE_GPU_REG_PA_SC_WINDOW_OFFSET].u32;
int16_t window_offset_x = window_offset & 0x7FFF;
@ -2699,6 +2693,19 @@ bool CommandProcessor::IssueCopy() {
dest_max_y - dest_min_y);
Rect2D src_rect(0, 0, dest_rect.width, dest_rect.height);
// The dest base address passed in has already been offset by the window
// offset, so to ensure texture lookup works we need to offset it.
// TODO(benvanik): allow texture cache to lookup partial textures.
// TODO(benvanik): change based on format.
int32_t dest_offset = window_offset_y * copy_dest_pitch * 4;
dest_offset += window_offset_x * 32 * 4;
copy_dest_base += dest_offset;
// Destination pointer in guest memory.
// We have GL throw bytes directly into it.
// TODO(benvanik): copy to staging texture then PBO back?
void* ptr = membase_ + GpuToCpu(copy_dest_base);
// Make active so glReadPixels reads from us.
switch (copy_command) {
case CopyCommand::kRaw: {
@ -2708,7 +2715,8 @@ bool CommandProcessor::IssueCopy() {
// Source from a bound render target.
// TODO(benvanik): RAW copy.
last_framebuffer_texture_ = texture_cache_.CopyTexture(
context_->blitter(), copy_dest_base, width, height,
context_->blitter(), copy_dest_base, dest_logical_width,
dest_logical_height, dest_block_width, dest_block_height,
ColorFormatToTextureFormat(copy_dest_format),
copy_dest_swap ? true : false, color_targets[copy_src_select],
src_rect, dest_rect);
@ -2718,9 +2726,11 @@ bool CommandProcessor::IssueCopy() {
} else {
// Source from the bound depth/stencil target.
// TODO(benvanik): RAW copy.
texture_cache_.CopyTexture(
context_->blitter(), copy_dest_base, width, height, src_format,
copy_dest_swap ? true : false, depth_target, src_rect, dest_rect);
texture_cache_.CopyTexture(context_->blitter(), copy_dest_base,
dest_logical_width, dest_logical_height,
dest_block_width, dest_block_height,
src_format, copy_dest_swap ? true : false,
depth_target, src_rect, dest_rect);
if (!FLAGS_disable_framebuffer_readback) {
// glReadPixels(x, y, w, h, GL_DEPTH_STENCIL, read_type, ptr);
}
@ -2733,7 +2743,8 @@ bool CommandProcessor::IssueCopy() {
// Either copy the readbuffer into an existing texture or create a new
// one in the cache so we can service future upload requests.
last_framebuffer_texture_ = texture_cache_.ConvertTexture(
context_->blitter(), copy_dest_base, width, height,
context_->blitter(), copy_dest_base, dest_logical_width,
dest_logical_height, dest_block_width, dest_block_height,
ColorFormatToTextureFormat(copy_dest_format),
copy_dest_swap ? true : false, color_targets[copy_src_select],
src_rect, dest_rect);
@ -2742,9 +2753,11 @@ bool CommandProcessor::IssueCopy() {
}
} else {
// Source from the bound depth/stencil target.
texture_cache_.ConvertTexture(
context_->blitter(), copy_dest_base, width, height, src_format,
copy_dest_swap ? true : false, depth_target, src_rect, dest_rect);
texture_cache_.ConvertTexture(context_->blitter(), copy_dest_base,
dest_logical_width, dest_logical_height,
dest_block_width, dest_block_height,
src_format, copy_dest_swap ? true : false,
depth_target, src_rect, dest_rect);
if (!FLAGS_disable_framebuffer_readback) {
// glReadPixels(x, y, w, h, GL_DEPTH_STENCIL, read_type, ptr);
}

43
src/xenia/gpu/gl4/texture_cache.cc
View File

@ -409,8 +409,8 @@ TextureCache::TextureEntry* TextureCache::LookupOrInsertTexture(
it != read_buffer_textures_.end(); ++it) {
auto read_buffer_entry = *it;
if (read_buffer_entry->guest_address == texture_info.guest_address &&
read_buffer_entry->width == texture_info.size_2d.logical_width &&
read_buffer_entry->height == texture_info.size_2d.logical_height) {
read_buffer_entry->block_width == texture_info.size_2d.block_width &&
read_buffer_entry->block_height == texture_info.size_2d.block_height) {
// Found! Acquire the handle and remove the readbuffer entry.
read_buffer_textures_.erase(it);
entry->handle = read_buffer_entry->handle;
@ -515,8 +515,8 @@ TextureCache::TextureEntry* TextureCache::LookupAddress(uint32_t guest_address,
const auto& texture_info = it->second->texture_info;
if (texture_info.guest_address == guest_address &&
texture_info.dimension == Dimension::k2D &&
texture_info.size_2d.logical_width == width &&
texture_info.size_2d.logical_height == height) {
texture_info.size_2d.input_width == width &&
texture_info.size_2d.input_height == height) {
return it->second;
}
}
@ -524,16 +524,20 @@ TextureCache::TextureEntry* TextureCache::LookupAddress(uint32_t guest_address,
}
GLuint TextureCache::CopyTexture(Blitter* blitter, uint32_t guest_address,
uint32_t width, uint32_t height,
TextureFormat format, bool swap_channels,
GLuint src_texture, Rect2D src_rect,
Rect2D dest_rect) {
return ConvertTexture(blitter, guest_address, width, height, format,
swap_channels, src_texture, src_rect, dest_rect);
uint32_t logical_width,
uint32_t logical_height, uint32_t block_width,
uint32_t block_height, TextureFormat format,
bool swap_channels, GLuint src_texture,
Rect2D src_rect, Rect2D dest_rect) {
return ConvertTexture(blitter, guest_address, logical_width, logical_height,
block_width, block_height, format, swap_channels,
src_texture, src_rect, dest_rect);
}
GLuint TextureCache::ConvertTexture(Blitter* blitter, uint32_t guest_address,
uint32_t width, uint32_t height,
uint32_t logical_width,
uint32_t logical_height,
uint32_t block_width, uint32_t block_height,
TextureFormat format, bool swap_channels,
GLuint src_texture, Rect2D src_rect,
Rect2D dest_rect) {
@ -546,7 +550,8 @@ GLuint TextureCache::ConvertTexture(Blitter* blitter, uint32_t guest_address,
// See if we have used a texture at this address before. If we have, we can
// reuse it.
// TODO(benvanik): better lookup matching format/etc?
auto texture_entry = LookupAddress(guest_address, width, height, format);
auto texture_entry =
LookupAddress(guest_address, block_width, block_height, format);
if (texture_entry) {
// Have existing texture.
assert_false(texture_entry->pending_invalidation);
@ -571,8 +576,9 @@ GLuint TextureCache::ConvertTexture(Blitter* blitter, uint32_t guest_address,
for (auto it = read_buffer_textures_.begin();
it != read_buffer_textures_.end(); ++it) {
const auto& entry = *it;
if (entry->guest_address == guest_address && entry->width == width &&
entry->height == height && entry->format == format) {
if (entry->guest_address == guest_address &&
entry->logical_width == logical_width &&
entry->logical_height == logical_height && entry->format == format) {
// Found an existing entry - just reupload.
if (config.format == GL_DEPTH_STENCIL) {
blitter->CopyDepthTexture(src_texture, src_rect, entry->handle,
@ -591,14 +597,17 @@ GLuint TextureCache::ConvertTexture(Blitter* blitter, uint32_t guest_address,
// of time we'll dump it.
auto entry = std::make_unique<ReadBufferTexture>();
entry->guest_address = guest_address;
entry->width = width;
entry->height = height;
entry->logical_width = logical_width;
entry->logical_height = logical_height;
entry->block_width = block_width;
entry->block_height = block_height;
entry->format = format;
glCreateTextures(GL_TEXTURE_2D, 1, &entry->handle);
glTextureParameteri(entry->handle, GL_TEXTURE_BASE_LEVEL, 0);
glTextureParameteri(entry->handle, GL_TEXTURE_MAX_LEVEL, 1);
glTextureStorage2D(entry->handle, 1, config.internal_format, width, height);
glTextureStorage2D(entry->handle, 1, config.internal_format, logical_width,
logical_height);
if (config.format == GL_DEPTH_STENCIL) {
blitter->CopyDepthTexture(src_texture, src_rect, entry->handle, dest_rect);
} else {

19
src/xenia/gpu/gl4/texture_cache.h
View File

@ -59,19 +59,24 @@ class TextureCache {
TextureEntryView* Demand(const TextureInfo& texture_info,
const SamplerInfo& sampler_info);
GLuint CopyTexture(Blitter* blitter, uint32_t guest_address, uint32_t width,
uint32_t height, TextureFormat format, bool swap_channels,
GLuint CopyTexture(Blitter* blitter, uint32_t guest_address,
uint32_t logical_width, uint32_t logical_height,
uint32_t block_width, uint32_t block_height,
TextureFormat format, bool swap_channels,
GLuint src_texture, Rect2D src_rect, Rect2D dest_rect);
GLuint ConvertTexture(Blitter* blitter, uint32_t guest_address,
uint32_t width, uint32_t height, TextureFormat format,
bool swap_channels, GLuint src_texture, Rect2D src_rect,
Rect2D dest_rect);
uint32_t logical_width, uint32_t logical_height,
uint32_t block_width, uint32_t block_height,
TextureFormat format, bool swap_channels,
GLuint src_texture, Rect2D src_rect, Rect2D dest_rect);
private:
struct ReadBufferTexture {
uint32_t guest_address;
uint32_t width;
uint32_t height;
uint32_t logical_width;
uint32_t logical_height;
uint32_t block_width;
uint32_t block_height;
TextureFormat format;
GLuint handle;
};