blob: c7a73bee1114d63d904c31aaee7adad809ae7f01 [file] [log] [blame]
/*
* Copyright (C) 2016 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "guest/hals/hwcomposer/common/cpu_composer.h"
#include <algorithm>
#include <cstdlib>
#include <utility>
#include <vector>
#include <hardware/hwcomposer.h>
#include <hardware/hwcomposer_defs.h>
#include <libyuv.h>
#include <log/log.h>
#include <system/graphics.h>
#include "common/libs/utils/size_utils.h"
#include "guest/hals/hwcomposer/common/geometry_utils.h"
namespace cuttlefish {
namespace {
bool LayerNeedsScaling(const hwc_layer_1_t& layer) {
int from_w = layer.sourceCrop.right - layer.sourceCrop.left;
int from_h = layer.sourceCrop.bottom - layer.sourceCrop.top;
int to_w = layer.displayFrame.right - layer.displayFrame.left;
int to_h = layer.displayFrame.bottom - layer.displayFrame.top;
bool not_rot_scale = from_w != to_w || from_h != to_h;
bool rot_scale = from_w != to_h || from_h != to_w;
bool needs_rot = layer.transform & HAL_TRANSFORM_ROT_90;
return needs_rot ? rot_scale : not_rot_scale;
}
bool LayerNeedsBlending(const hwc_layer_1_t& layer) {
return layer.blending != HWC_BLENDING_NONE;
}
bool LayerNeedsAttenuation(const hwc_layer_1_t& layer) {
return layer.blending == HWC_BLENDING_COVERAGE;
}
struct BufferSpec;
typedef int (*ConverterFunction)(const BufferSpec& src, const BufferSpec& dst,
bool v_flip);
int DoCopy(const BufferSpec& src, const BufferSpec& dst, bool v_flip);
int ConvertFromYV12(const BufferSpec& src, const BufferSpec& dst, bool v_flip);
ConverterFunction GetConverter(uint32_t format) {
switch (format) {
case HAL_PIXEL_FORMAT_RGBA_8888:
case HAL_PIXEL_FORMAT_RGBX_8888:
case HAL_PIXEL_FORMAT_IMPLEMENTATION_DEFINED:
return &DoCopy;
case HAL_PIXEL_FORMAT_YV12:
return &ConvertFromYV12;
// Unsupported formats
// TODO(jemoreira): Conversion from these formats should be implemented as
// we find evidence of its usage.
// case HAL_PIXEL_FORMAT_BGRA_8888:
// case HAL_PIXEL_FORMAT_RGB_888:
// case HAL_PIXEL_FORMAT_RGB_565:
// case HAL_PIXEL_FORMAT_sRGB_A_8888:
// case HAL_PIXEL_FORMAT_sRGB_X_8888:
// case HAL_PIXEL_FORMAT_Y8:
// case HAL_PIXEL_FORMAT_Y16:
// case HAL_PIXEL_FORMAT_RAW_SENSOR:
// case HAL_PIXEL_FORMAT_BLOB:
// case HAL_PIXEL_FORMAT_YCbCr_420_888:
// case HAL_PIXEL_FORMAT_YCbCr_422_SP:
// case HAL_PIXEL_FORMAT_YCrCb_420_SP:
// case HAL_PIXEL_FORMAT_YCbCr_422_I:
default:
ALOGW("Unsupported format: 0x%04x, returning null converter function",
format);
}
return NULL;
}
// Whether we support a given format
bool IsFormatSupported(uint32_t format) { return GetConverter(format) != NULL; }
bool CanCompositeLayer(const hwc_layer_1_t& layer) {
if (layer.handle == NULL) {
ALOGW("%s received a layer with a null handler", __FUNCTION__);
return false;
}
int format = reinterpret_cast<const private_handle_t*>(layer.handle)->format;
if (!IsFormatSupported(format)) {
ALOGD("Unsupported pixel format: 0x%x, doing software composition instead",
format);
return false;
}
return true;
}
/*******************************************************************************
Libyuv's convert functions only allow the combination of any rotation (multiple
of 90 degrees) and a vertical flip, but not horizontal flips.
Surfaceflinger's transformations are expressed in terms of a vertical flip, a
horizontal flip and/or a single 90 degrees clockwise rotation (see
NATIVE_WINDOW_TRANSFORM_HINT documentation on system/window.h for more insight).
The following code allows to turn a horizontal flip into a 180 degrees rotation
and a vertical flip.
*******************************************************************************/
libyuv::RotationMode GetRotationFromTransform(uint32_t transform) {
uint32_t rotation =
(transform & HAL_TRANSFORM_ROT_90) ? 1 : 0; // 1 * ROT90 bit
rotation += (transform & HAL_TRANSFORM_FLIP_H) ? 2 : 0; // 2 * VFLIP bit
return static_cast<libyuv::RotationMode>(90 * rotation);
}
bool GetVFlipFromTransform(uint32_t transform) {
// vertical flip xor horizontal flip
return ((transform & HAL_TRANSFORM_FLIP_V) >> 1) ^
(transform & HAL_TRANSFORM_FLIP_H);
}
struct BufferSpec {
uint8_t* buffer;
size_t size;
int width;
int height;
int stride;
int crop_x;
int crop_y;
int crop_width;
int crop_height;
uint32_t format;
BufferSpec(uint8_t* buffer, size_t size, int width, int height, int stride)
: buffer(buffer),
size(size),
width(width),
height(height),
stride(stride),
crop_x(0),
crop_y(0),
crop_width(width),
crop_height(height),
format(HAL_PIXEL_FORMAT_RGBA_8888) {}
};
int ConvertFromYV12(const BufferSpec& src, const BufferSpec& dst, bool v_flip) {
// use the stride in pixels as the source width
int stride_in_pixels = src.stride / formatToBytesPerPixel(src.format);
// The following calculation of plane offsets and alignments are based on
// swiftshader's Sampler::setTextureLevel() implementation
// (Renderer/Sampler.cpp:225)
uint8_t* src_y = src.buffer;
int stride_y = stride_in_pixels;
uint8_t* src_v = src_y + stride_y * src.height;
int stride_v = cuttlefish::AlignToPowerOf2(stride_y / 2, 4);
uint8_t* src_u = src_v + stride_v * src.height / 2;
int stride_u = cuttlefish::AlignToPowerOf2(stride_y / 2, 4);
// Adjust for crop
src_y += src.crop_y * stride_y + src.crop_x;
src_v += (src.crop_y / 2) * stride_v + (src.crop_x / 2);
src_u += (src.crop_y / 2) * stride_u + (src.crop_x / 2);
uint8_t* dst_buffer = dst.buffer + dst.crop_y * dst.stride +
dst.crop_x * formatToBytesPerPixel(dst.format);
// YV12 is the same as I420, with the U and V planes swapped
return libyuv::I420ToARGB(src_y, stride_y, src_v, stride_v, src_u, stride_u,
dst_buffer, dst.stride, dst.crop_width,
v_flip ? -dst.crop_height : dst.crop_height);
}
int DoConversion(const BufferSpec& src, const BufferSpec& dst, bool v_flip) {
return (*GetConverter(src.format))(src, dst, v_flip);
}
int DoCopy(const BufferSpec& src, const BufferSpec& dst, bool v_flip) {
// Point to the upper left corner of the crop rectangle
uint8_t* src_buffer = src.buffer + src.crop_y * src.stride +
src.crop_x * formatToBytesPerPixel(src.format);
uint8_t* dst_buffer = dst.buffer + dst.crop_y * dst.stride +
dst.crop_x * formatToBytesPerPixel(dst.format);
int width = src.crop_width;
int height = src.crop_height;
if (v_flip) {
height = -height;
}
// HAL formats are named based on the order of the pixel componets on the
// byte stream, while libyuv formats are named based on the order of those
// pixel components in an integer written from left to right. So
// libyuv::FOURCC_ARGB is equivalent to HAL_PIXEL_FORMAT_BGRA_8888.
return libyuv::ARGBCopy(src_buffer, src.stride, dst_buffer, dst.stride, width,
height);
}
int DoRotation(const BufferSpec& src, const BufferSpec& dst,
libyuv::RotationMode rotation, bool v_flip) {
// Point to the upper left corner of the crop rectangles
uint8_t* src_buffer = src.buffer + src.crop_y * src.stride +
src.crop_x * formatToBytesPerPixel(src.format);
uint8_t* dst_buffer = dst.buffer + dst.crop_y * dst.stride +
dst.crop_x * formatToBytesPerPixel(dst.format);
int width = src.crop_width;
int height = src.crop_height;
if (v_flip) {
height = -height;
}
return libyuv::ARGBRotate(src_buffer, src.stride, dst_buffer, dst.stride,
width, height, rotation);
}
int DoScaling(const BufferSpec& src, const BufferSpec& dst, bool v_flip) {
// Point to the upper left corner of the crop rectangles
uint8_t* src_buffer = src.buffer + src.crop_y * src.stride +
src.crop_x * formatToBytesPerPixel(src.format);
uint8_t* dst_buffer = dst.buffer + dst.crop_y * dst.stride +
dst.crop_x * formatToBytesPerPixel(dst.format);
int src_width = src.crop_width;
int src_height = src.crop_height;
int dst_width = dst.crop_width;
int dst_height = dst.crop_height;
if (v_flip) {
src_height = -src_height;
}
return libyuv::ARGBScale(src_buffer, src.stride, src_width, src_height,
dst_buffer, dst.stride, dst_width, dst_height,
libyuv::kFilterBilinear);
}
int DoAttenuation(const BufferSpec& src, const BufferSpec& dest, bool v_flip) {
// Point to the upper left corner of the crop rectangles
uint8_t* src_buffer = src.buffer + src.crop_y * src.stride +
src.crop_x * formatToBytesPerPixel(src.format);
uint8_t* dst_buffer = dest.buffer + dest.crop_y * dest.stride +
dest.crop_x * formatToBytesPerPixel(dest.format);
int width = dest.crop_width;
int height = dest.crop_height;
if (v_flip) {
height = -height;
}
return libyuv::ARGBAttenuate(src_buffer, src.stride, dst_buffer, dest.stride,
width, height);
}
int DoBlending(const BufferSpec& src, const BufferSpec& dest, bool v_flip) {
// Point to the upper left corner of the crop rectangles
uint8_t* src_buffer = src.buffer + src.crop_y * src.stride +
src.crop_x * formatToBytesPerPixel(src.format);
uint8_t* dst_buffer = dest.buffer + dest.crop_y * dest.stride +
dest.crop_x * formatToBytesPerPixel(dest.format);
int width = dest.crop_width;
int height = dest.crop_height;
if (v_flip) {
height = -height;
}
// libyuv's ARGB format is hwcomposer's BGRA format, since blending only cares
// for the position of alpha in the pixel and not the position of the colors
// this function is perfectly usable.
return libyuv::ARGBBlend(src_buffer, src.stride, dst_buffer, dest.stride,
dst_buffer, dest.stride, width, height);
}
} // namespace
void CpuComposer::CompositeLayer(hwc_layer_1_t* src_layer, int buffer_idx) {
libyuv::RotationMode rotation =
GetRotationFromTransform(src_layer->transform);
const private_handle_t* src_priv_handle =
reinterpret_cast<const private_handle_t*>(src_layer->handle);
// TODO(jemoreira): Remove the hardcoded fomat.
bool needs_conversion = src_priv_handle->format != HAL_PIXEL_FORMAT_RGBX_8888;
bool needs_scaling = LayerNeedsScaling(*src_layer);
bool needs_rotation = rotation != libyuv::kRotate0;
bool needs_transpose = needs_rotation && rotation != libyuv::kRotate180;
bool needs_vflip = GetVFlipFromTransform(src_layer->transform);
bool needs_attenuation = LayerNeedsAttenuation(*src_layer);
bool needs_blending = LayerNeedsBlending(*src_layer);
bool needs_copy = !(needs_conversion || needs_scaling || needs_rotation ||
needs_vflip || needs_attenuation || needs_blending);
uint8_t* src_buffer;
uint8_t* dst_buffer =
reinterpret_cast<uint8_t*>(screen_view_->GetBuffer(buffer_idx));
int retval = gralloc_module_->lock(
gralloc_module_, src_layer->handle, GRALLOC_USAGE_SW_READ_OFTEN, 0, 0,
src_priv_handle->x_res, src_priv_handle->y_res,
reinterpret_cast<void**>(&src_buffer));
if (retval) {
ALOGE("Got error code %d from lock function", retval);
return;
}
if (retval) {
ALOGE("Got error code %d from lock function", retval);
// TODO(jemoreira): Use a lock_guard-like object.
gralloc_module_->unlock(gralloc_module_, src_priv_handle);
return;
}
BufferSpec src_layer_spec(src_buffer, src_priv_handle->total_size,
src_priv_handle->x_res, src_priv_handle->y_res,
src_priv_handle->stride_in_pixels *
formatToBytesPerPixel(src_priv_handle->format));
src_layer_spec.crop_x = src_layer->sourceCrop.left;
src_layer_spec.crop_y = src_layer->sourceCrop.top;
src_layer_spec.crop_width =
src_layer->sourceCrop.right - src_layer->sourceCrop.left;
src_layer_spec.crop_height =
src_layer->sourceCrop.bottom - src_layer->sourceCrop.top;
src_layer_spec.format = src_priv_handle->format;
BufferSpec dst_layer_spec(dst_buffer, screen_view_->buffer_size(),
screen_view_->x_res(), screen_view_->y_res(),
screen_view_->line_length());
dst_layer_spec.crop_x = src_layer->displayFrame.left;
dst_layer_spec.crop_y = src_layer->displayFrame.top;
dst_layer_spec.crop_width =
src_layer->displayFrame.right - src_layer->displayFrame.left;
dst_layer_spec.crop_height =
src_layer->displayFrame.bottom - src_layer->displayFrame.top;
// TODO(jemoreira): Remove the hardcoded fomat.
dst_layer_spec.format = HAL_PIXEL_FORMAT_RGBX_8888;
// Add the destination layer to the bottom of the buffer stack
std::vector<BufferSpec> dest_buffer_stack(1, dst_layer_spec);
// If more than operation is to be performed, a temporary buffer is needed for
// each additional operation
// N operations need N destination buffers, the destination layer (the
// framebuffer) is one of them, so only N-1 temporary buffers are needed.
// Vertical flip is not taken into account because it can be done together
// with any other operation.
int needed_tmp_buffers = (needs_conversion ? 1 : 0) +
(needs_scaling ? 1 : 0) + (needs_rotation ? 1 : 0) +
(needs_attenuation ? 1 : 0) +
(needs_blending ? 1 : 0) + (needs_copy ? 1 : 0) - 1;
int x_res = src_layer->displayFrame.right - src_layer->displayFrame.left;
int y_res = src_layer->displayFrame.bottom - src_layer->displayFrame.top;
size_t output_frame_size =
x_res * cuttlefish::AlignToPowerOf2(y_res * screen_view_->bytes_per_pixel(), 4);
while (needed_tmp_buffers > 0) {
BufferSpec tmp(
RotateTmpBuffer(needed_tmp_buffers), output_frame_size, x_res, y_res,
cuttlefish::AlignToPowerOf2(x_res * screen_view_->bytes_per_pixel(), 4));
dest_buffer_stack.push_back(tmp);
needed_tmp_buffers--;
}
// Conversion and scaling should always be the first operations, so that every
// other operation works on equally sized frames (garanteed to fit in the tmp
// buffers)
// TODO(jemoreira): We are converting to ARGB as the first step under the
// assumption that scaling ARGB is faster than scaling I420 (the most common).
// This should be confirmed with testing.
if (needs_conversion) {
BufferSpec& dst_buffer_spec = dest_buffer_stack.back();
if (needs_scaling || needs_transpose) {
// If a rotation or a scaling operation are needed the dimensions at the
// top of the buffer stack are wrong (wrong sizes for scaling, swapped
// width and height for 90 and 270 rotations).
// Make width and height match the crop sizes on the source
int src_width = src_layer_spec.crop_width;
int src_height = src_layer_spec.crop_height;
int dst_stride =
cuttlefish::AlignToPowerOf2(src_width * screen_view_->bytes_per_pixel(), 4);
size_t needed_size = dst_stride * src_height;
dst_buffer_spec.width = src_width;
dst_buffer_spec.height = src_height;
// Ajust the stride accordingly
dst_buffer_spec.stride = dst_stride;
// Crop sizes also need to be adjusted
dst_buffer_spec.crop_width = src_width;
dst_buffer_spec.crop_height = src_height;
dst_buffer_spec.size = needed_size;
// crop_x and y are fine at 0, format is already set to match destination
// In case of a scale, the source frame may be bigger than the default tmp
// buffer size
if (needed_size > tmp_buffer_.size() / kNumTmpBufferPieces) {
dst_buffer_spec.buffer = GetSpecialTmpBuffer(needed_size);
}
}
retval = DoConversion(src_layer_spec, dst_buffer_spec, needs_vflip);
if (retval) {
ALOGE("Got error code %d from DoConversion function", retval);
}
needs_vflip = false;
src_layer_spec = dst_buffer_spec;
dest_buffer_stack.pop_back();
}
if (needs_scaling) {
BufferSpec& dst_buffer_spec = dest_buffer_stack.back();
if (needs_transpose) {
// If a rotation is needed, the temporary buffer has the correct size but
// needs to be transposed and have its stride updated accordingly. The
// crop sizes also needs to be transposed, but not the x and y since they
// are both zero in a temporary buffer (and it is a temporary buffer
// because a rotation will be performed next).
std::swap(dst_buffer_spec.width, dst_buffer_spec.height);
std::swap(dst_buffer_spec.crop_width, dst_buffer_spec.crop_height);
// TODO (jemoreira): Aligment (To align here may cause the needed size to
// be bigger than the buffer, so care should be taken)
dst_buffer_spec.stride =
dst_buffer_spec.width * screen_view_->bytes_per_pixel();
}
retval = DoScaling(src_layer_spec, dst_buffer_spec, needs_vflip);
needs_vflip = false;
if (retval) {
ALOGE("Got error code %d from DoScaling function", retval);
}
src_layer_spec = dst_buffer_spec;
dest_buffer_stack.pop_back();
}
if (needs_rotation) {
retval = DoRotation(src_layer_spec, dest_buffer_stack.back(), rotation,
needs_vflip);
needs_vflip = false;
if (retval) {
ALOGE("Got error code %d from DoTransform function", retval);
}
src_layer_spec = dest_buffer_stack.back();
dest_buffer_stack.pop_back();
}
if (needs_attenuation) {
retval =
DoAttenuation(src_layer_spec, dest_buffer_stack.back(), needs_vflip);
needs_vflip = false;
if (retval) {
ALOGE("Got error code %d from DoBlending function", retval);
}
src_layer_spec = dest_buffer_stack.back();
dest_buffer_stack.pop_back();
}
if (needs_copy) {
retval = DoCopy(src_layer_spec, dest_buffer_stack.back(), needs_vflip);
needs_vflip = false;
if (retval) {
ALOGE("Got error code %d from DoBlending function", retval);
}
src_layer_spec = dest_buffer_stack.back();
dest_buffer_stack.pop_back();
}
// Blending (if needed) should always be the last operation, so that it reads
// and writes in the destination layer and not some temporary buffer.
if (needs_blending) {
retval = DoBlending(src_layer_spec, dest_buffer_stack.back(), needs_vflip);
needs_vflip = false;
if (retval) {
ALOGE("Got error code %d from DoBlending function", retval);
}
// Don't need to assign destination to source in the last one
dest_buffer_stack.pop_back();
}
gralloc_module_->unlock(gralloc_module_, src_priv_handle);
}
/* static */ const int CpuComposer::kNumTmpBufferPieces = 2;
CpuComposer::CpuComposer(std::unique_ptr<ScreenView> screen_view)
: BaseComposer(std::move(screen_view)),
tmp_buffer_(kNumTmpBufferPieces * screen_view_->buffer_size()) {}
int CpuComposer::PrepareLayers(size_t num_layers, hwc_layer_1_t* layers) {
int composited_layers_count = 0;
// Loop over layers in inverse order of z-index
for (size_t layer_index = num_layers; layer_index > 0;) {
// Decrement here to be able to compare unsigned integer with 0 in the
// loop condition
--layer_index;
if (IS_TARGET_FRAMEBUFFER(layers[layer_index].compositionType)) {
continue;
}
if (layers[layer_index].flags & HWC_SKIP_LAYER) {
continue;
}
if (layers[layer_index].compositionType == HWC_BACKGROUND) {
layers[layer_index].compositionType = HWC_FRAMEBUFFER;
continue;
}
layers[layer_index].compositionType = HWC_OVERLAY;
// Hwcomposer cannot draw below software-composed layers, so we need
// to mark those HWC_FRAMEBUFFER as well.
for (size_t top_idx = layer_index + 1; top_idx < num_layers; ++top_idx) {
// layers marked as skip are in a state that makes them unreliable to
// read, so it's best to assume they cover the whole screen
if (layers[top_idx].flags & HWC_SKIP_LAYER ||
(layers[top_idx].compositionType == HWC_FRAMEBUFFER &&
LayersOverlap(layers[layer_index], layers[top_idx]))) {
layers[layer_index].compositionType = HWC_FRAMEBUFFER;
break;
}
}
if (layers[layer_index].compositionType == HWC_OVERLAY &&
!CanCompositeLayer(layers[layer_index])) {
layers[layer_index].compositionType = HWC_FRAMEBUFFER;
}
if (layers[layer_index].compositionType == HWC_OVERLAY) {
++composited_layers_count;
}
}
return composited_layers_count;
}
int CpuComposer::SetLayers(size_t num_layers, hwc_layer_1_t* layers) {
int targetFbs = 0;
int buffer_idx = screen_view_->NextBuffer();
// The framebuffer target layer should be composed if at least one layers was
// marked HWC_FRAMEBUFFER or if it's the only layer in the composition
// (unlikely)
bool fb_target = true;
for (size_t idx = 0; idx < num_layers; idx++) {
if (layers[idx].compositionType == HWC_FRAMEBUFFER) {
// At least one was found
fb_target = true;
break;
}
if (layers[idx].compositionType == HWC_OVERLAY) {
// Not the only layer in the composition
fb_target = false;
}
}
// When the framebuffer target needs to be composed, it has to go first.
if (fb_target) {
for (size_t idx = 0; idx < num_layers; idx++) {
if (IS_TARGET_FRAMEBUFFER(layers[idx].compositionType)) {
CompositeLayer(&layers[idx], buffer_idx);
break;
}
}
}
for (size_t idx = 0; idx < num_layers; idx++) {
if (IS_TARGET_FRAMEBUFFER(layers[idx].compositionType)) {
++targetFbs;
}
if (layers[idx].compositionType == HWC_OVERLAY &&
!(layers[idx].flags & HWC_SKIP_LAYER)) {
CompositeLayer(&layers[idx], buffer_idx);
}
}
if (targetFbs != 1) {
ALOGW("Saw %zu layers, posted=%d", num_layers, targetFbs);
}
screen_view_->Broadcast(buffer_idx);
return 0;
}
uint8_t* CpuComposer::RotateTmpBuffer(unsigned int order) {
return &tmp_buffer_[(order % kNumTmpBufferPieces) * tmp_buffer_.size() /
kNumTmpBufferPieces];
}
uint8_t* CpuComposer::GetSpecialTmpBuffer(size_t needed_size) {
special_tmp_buffer_.resize(needed_size);
return &special_tmp_buffer_[0];
}
} // namespace cuttlefish