Generalize bitmap support and add remaining GL formats.
Fix bug in command fifo looping case.
diff --git a/rsAllocation.cpp b/rsAllocation.cpp
index 7b8bc80..3a01a75 100644
--- a/rsAllocation.cpp
+++ b/rsAllocation.cpp
@@ -47,6 +47,7 @@
Allocation::~Allocation()
{
+ LOGE("Allocation %p destryed", this);
}
void Allocation::setCpuWritable(bool)
@@ -77,6 +78,13 @@
//LOGE("uploadToTexture %i, lod %i", mTextureID, lodOffset);
+ GLenum type = mType->getElement()->getGLType();
+ GLenum format = mType->getElement()->getGLFormat();
+
+ if (!type || !format) {
+ return;
+ }
+
if (!mTextureID) {
glGenTextures(1, &mTextureID);
}
@@ -87,9 +95,9 @@
adapt.setLOD(lod+lodOffset);
uint16_t * ptr = static_cast<uint16_t *>(adapt.getElement(0,0));
- glTexImage2D(GL_TEXTURE_2D, lod, GL_RGB,
- adapt.getDimX(), adapt.getDimY(),
- 0, GL_RGB, GL_UNSIGNED_SHORT_5_6_5, ptr);
+ glTexImage2D(GL_TEXTURE_2D, lod, format,
+ adapt.getDimX(), adapt.getDimY(),
+ 0, format, type, ptr);
}
}
@@ -121,7 +129,7 @@
memcpy(ptr, data, count * eSize);
}
-void Allocation::subData(uint32_t xoff, uint32_t yoff,
+void Allocation::subData(uint32_t xoff, uint32_t yoff,
uint32_t w, uint32_t h, const void *data)
{
uint32_t eSize = mType->getElementSizeBytes();
@@ -147,7 +155,7 @@
/////////////////
-//
+//
namespace android {
@@ -192,7 +200,7 @@
{
}
-static void mip(const Adapter2D &out, const Adapter2D &in)
+static void mip565(const Adapter2D &out, const Adapter2D &in)
{
uint32_t w = out.getDimX();
uint32_t h = out.getDimY();
@@ -203,7 +211,26 @@
const uint16_t *i2 = static_cast<uint16_t *>(in.getElement(0, y*2+1));
for (uint32_t x=0; x < h; x++) {
- *oPtr = rsBoxFilter565(i1[0], i1[2], i2[0], i2[1]);
+ *oPtr = rsBoxFilter565(i1[0], i1[1], i2[0], i2[1]);
+ oPtr ++;
+ i1 += 2;
+ i2 += 2;
+ }
+ }
+}
+
+static void mip8888(const Adapter2D &out, const Adapter2D &in)
+{
+ uint32_t w = out.getDimX();
+ uint32_t h = out.getDimY();
+
+ for (uint32_t y=0; y < w; y++) {
+ uint32_t *oPtr = static_cast<uint32_t *>(out.getElement(0, y));
+ const uint32_t *i1 = static_cast<uint32_t *>(in.getElement(0, y*2));
+ const uint32_t *i2 = static_cast<uint32_t *>(in.getElement(0, y*2+1));
+
+ for (uint32_t x=0; x < h; x++) {
+ *oPtr = rsBoxFilter8888(i1[0], i1[1], i2[0], i2[1]);
oPtr ++;
i1 += 2;
i2 += 2;
@@ -255,21 +282,25 @@
static ElementConverter_t pickConverter(RsElementPredefined dstFmt, RsElementPredefined srcFmt)
{
- if ((dstFmt == RS_ELEMENT_RGB_565) &&
+ if ((dstFmt == RS_ELEMENT_RGB_565) &&
(srcFmt == RS_ELEMENT_RGB_565)) {
return elementConverter_cpy_16;
}
- if ((dstFmt == RS_ELEMENT_RGB_565) &&
+ if ((dstFmt == RS_ELEMENT_RGB_565) &&
(srcFmt == RS_ELEMENT_RGB_888)) {
return elementConverter_888_to_565;
}
- if ((dstFmt == RS_ELEMENT_RGB_565) &&
+ if ((dstFmt == RS_ELEMENT_RGB_565) &&
(srcFmt == RS_ELEMENT_RGBA_8888)) {
return elementConverter_8888_to_565;
}
+ if ((dstFmt == RS_ELEMENT_RGBA_8888) &&
+ (srcFmt == RS_ELEMENT_RGBA_8888)) {
+ return elementConverter_cpy_32;
+ }
LOGE("pickConverter, unsuported combo");
return 0;
@@ -303,7 +334,7 @@
for(uint32_t lod=0; lod < (texAlloc->getType()->getLODCount() -1); lod++) {
adapt.setLOD(lod);
adapt2.setLOD(lod + 1);
- mip(adapt2, adapt);
+ mip565(adapt2, adapt);
}
}
@@ -312,6 +343,8 @@
RsAllocation rsi_AllocationCreateFromFile(Context *rsc, const char *file, bool genMips)
{
+ bool use32bpp = false;
+
typedef struct _Win3xBitmapHeader
{
uint16_t type;
@@ -351,7 +384,11 @@
int32_t texWidth = rsHigherPow2(hdr.width);
int32_t texHeight = rsHigherPow2(hdr.height);
- rsi_TypeBegin(rsc, rsi_ElementGetPredefined(rsc, RS_ELEMENT_RGB_565));
+ if (use32bpp) {
+ rsi_TypeBegin(rsc, rsi_ElementGetPredefined(rsc, RS_ELEMENT_RGBA_8888));
+ } else {
+ rsi_TypeBegin(rsc, rsi_ElementGetPredefined(rsc, RS_ELEMENT_RGB_565));
+ }
rsi_TypeAdd(rsc, RS_DIMENSION_X, texWidth);
rsi_TypeAdd(rsc, RS_DIMENSION_Y, texHeight);
if (genMips) {
@@ -372,14 +409,29 @@
Adapter2D adapt(texAlloc);
uint8_t * fileInBuf = new uint8_t[texWidth * 3];
uint32_t yOffset = (hdr.width - hdr.height) / 2;
- uint16_t *tmp = static_cast<uint16_t *>(adapt.getElement(0, yOffset));
- for (int y=0; y < hdr.height; y++) {
- fseek(f, hdr.offset + (y*hdr.width*3), SEEK_SET);
- fread(fileInBuf, 1, hdr.width * 3, f);
- for(int x=0; x < hdr.width; x++) {
- *tmp = rs888to565(fileInBuf[x*3], fileInBuf[x*3 + 1], fileInBuf[x*3 + 2]);
- tmp++;
+ if (use32bpp) {
+ uint8_t *tmp = static_cast<uint8_t *>(adapt.getElement(0, yOffset));
+ for (int y=0; y < hdr.height; y++) {
+ fseek(f, hdr.offset + (y*hdr.width*3), SEEK_SET);
+ fread(fileInBuf, 1, hdr.width * 3, f);
+ for(int x=0; x < hdr.width; x++) {
+ tmp[0] = fileInBuf[x*3 + 2];
+ tmp[1] = fileInBuf[x*3 + 1];
+ tmp[2] = fileInBuf[x*3];
+ tmp[3] = 0xff;
+ tmp += 4;
+ }
+ }
+ } else {
+ uint16_t *tmp = static_cast<uint16_t *>(adapt.getElement(0, yOffset));
+ for (int y=0; y < hdr.height; y++) {
+ fseek(f, hdr.offset + (y*hdr.width*3), SEEK_SET);
+ fread(fileInBuf, 1, hdr.width * 3, f);
+ for(int x=0; x < hdr.width; x++) {
+ *tmp = rs888to565(fileInBuf[x*3 + 2], fileInBuf[x*3 + 1], fileInBuf[x*3]);
+ tmp++;
+ }
}
}
@@ -391,7 +443,11 @@
for(uint32_t lod=0; lod < (texAlloc->getType()->getLODCount() -1); lod++) {
adapt.setLOD(lod);
adapt2.setLOD(lod + 1);
- mip(adapt2, adapt);
+ if (use32bpp) {
+ mip8888(adapt2, adapt);
+ } else {
+ mip565(adapt2, adapt);
+ }
}
}
diff --git a/rsElement.cpp b/rsElement.cpp
index bd11f72..5a44f47 100644
--- a/rsElement.cpp
+++ b/rsElement.cpp
@@ -16,6 +16,8 @@
#include "rsContext.h"
+#include <GLES/gl.h>
+
using namespace android;
using namespace android::renderscript;
@@ -235,6 +237,108 @@
return offset;
}
+uint32_t Element::getGLType() const
+{
+ int bits[4];
+
+ if (mComponentCount > 4) {
+ return 0;
+ }
+
+ for (uint32_t ct=0; ct < mComponentCount; ct++) {
+ bits[ct] = mComponents[ct]->getBits();
+ if (mComponents[ct]->getType() != Component::UNSIGNED) {
+ return 0;
+ }
+ if (!mComponents[ct]->getIsNormalized()) {
+ return 0;
+ }
+ }
+
+ switch(mComponentCount) {
+ case 1:
+ if (bits[0] == 8) {
+ return GL_UNSIGNED_BYTE;
+ }
+ return 0;
+ case 2:
+ if ((bits[0] == 8) &&
+ (bits[1] == 8)) {
+ return GL_UNSIGNED_BYTE;
+ }
+ return 0;
+ case 3:
+ if ((bits[0] == 8) &&
+ (bits[1] == 8) &&
+ (bits[2] == 8)) {
+ return GL_UNSIGNED_BYTE;
+ }
+ if ((bits[0] == 5) &&
+ (bits[1] == 6) &&
+ (bits[2] == 5)) {
+ return GL_UNSIGNED_SHORT_5_6_5;
+ }
+ return 0;
+ case 4:
+ if ((bits[0] == 8) &&
+ (bits[1] == 8) &&
+ (bits[2] == 8) &&
+ (bits[3] == 8)) {
+ return GL_UNSIGNED_BYTE;
+ }
+ if ((bits[0] == 4) &&
+ (bits[1] == 4) &&
+ (bits[2] == 4) &&
+ (bits[3] == 4)) {
+ return GL_UNSIGNED_SHORT_4_4_4_4;
+ }
+ if ((bits[0] == 5) &&
+ (bits[1] == 5) &&
+ (bits[2] == 5) &&
+ (bits[3] == 1)) {
+ return GL_UNSIGNED_SHORT_5_5_5_1;
+ }
+ }
+ return 0;
+}
+
+uint32_t Element::getGLFormat() const
+{
+ switch(mComponentCount) {
+ case 1:
+ if (mComponents[0]->getKind() == Component::ALPHA) {
+ return GL_ALPHA;
+ }
+ if (mComponents[0]->getKind() == Component::LUMINANCE) {
+ return GL_LUMINANCE;
+ }
+ break;
+ case 2:
+ if ((mComponents[0]->getKind() == Component::LUMINANCE) &&
+ (mComponents[1]->getKind() == Component::ALPHA)) {
+ return GL_LUMINANCE_ALPHA;
+ }
+ break;
+ case 3:
+ if ((mComponents[0]->getKind() == Component::RED) &&
+ (mComponents[1]->getKind() == Component::GREEN) &&
+ (mComponents[2]->getKind() == Component::BLUE)) {
+ return GL_RGB;
+ }
+ break;
+ case 4:
+ if ((mComponents[0]->getKind() == Component::RED) &&
+ (mComponents[1]->getKind() == Component::GREEN) &&
+ (mComponents[2]->getKind() == Component::BLUE) &&
+ (mComponents[3]->getKind() == Component::ALPHA)) {
+ return GL_RGBA;
+ }
+ break;
+ }
+ return 0;
+}
+
+
ElementState::ElementState()
{
}
diff --git a/rsElement.h b/rsElement.h
index 7852ffc..2434977 100644
--- a/rsElement.h
+++ b/rsElement.h
@@ -36,6 +36,8 @@
void setComponent(uint32_t idx, Component *c);
+ uint32_t getGLType() const;
+ uint32_t getGLFormat() const;
size_t getSizeBits() const;
diff --git a/rsLocklessFifo.cpp b/rsLocklessFifo.cpp
index 3f51e04..67ab434 100644
--- a/rsLocklessFifo.cpp
+++ b/rsLocklessFifo.cpp
@@ -74,6 +74,7 @@
freeSpace = 0;
}
+ //LOGE("free %i", freeSpace);
return freeSpace;
}
@@ -85,8 +86,8 @@
void * LocklessCommandFifo::reserve(uint32_t sizeInBytes)
{
- // Add space for command header;
- sizeInBytes += 4;
+ // Add space for command header and loop token;
+ sizeInBytes += 8;
//dumpState("reserve");
if (getFreeSpace() < sizeInBytes) {
@@ -153,16 +154,17 @@
void LocklessCommandFifo::makeSpace(uint32_t bytes)
{
+ //dumpState("make space");
if ((mPut+bytes) > mEnd) {
// Need to loop regardless of where get is.
- while((mGet > mPut) && (mPut+4 >= mGet)) {
+ while((mGet > mPut) && (mBuffer+4 >= mGet)) {
sleep(1);
}
// Toss in a reset then the normal wait for space will do the rest.
reinterpret_cast<uint16_t *>(mPut)[0] = 0;
reinterpret_cast<uint16_t *>(mPut)[1] = 0;
- mPut += 4;
+ mPut = mBuffer;
}
// it will fit here so we just need to wait for space.
diff --git a/rsUtils.h b/rsUtils.h
index f40e2ce..5a43fb3 100644
--- a/rsUtils.h
+++ b/rsUtils.h
@@ -96,13 +96,19 @@
static inline uint16_t rsBoxFilter565(uint16_t i1, uint16_t i2, uint16_t i3, uint16_t i4)
{
uint32_t r = ((i1 & 0x1f) + (i2 & 0x1f) + (i3 & 0x1f) + (i4 & 0x1f));
- uint32_t g = ((i1 >> 5) & 0x3f) + ((i2 >> 5) & 0x3f) + ((i3 >> 5) & 0x3f) + ((i1 >> 5) & 0x3f);
+ uint32_t g = ((i1 >> 5) & 0x3f) + ((i2 >> 5) & 0x3f) + ((i3 >> 5) & 0x3f) + ((i4 >> 5) & 0x3f);
uint32_t b = ((i1 >> 11) + (i2 >> 11) + (i3 >> 11) + (i4 >> 11));
return (r >> 2) | ((g >> 2) << 5) | ((b >> 2) << 11);
}
-
-
+static inline uint32_t rsBoxFilter8888(uint32_t i1, uint32_t i2, uint32_t i3, uint32_t i4)
+{
+ uint32_t r = (i1 & 0xff) + (i2 & 0xff) + (i3 & 0xff) + (i4 & 0xff);
+ uint32_t g = ((i1 >> 8) & 0xff) + ((i2 >> 8) & 0xff) + ((i3 >> 8) & 0xff) + ((i4 >> 8) & 0xff);
+ uint32_t b = ((i1 >> 16) & 0xff) + ((i2 >> 16) & 0xff) + ((i3 >> 16) & 0xff) + ((i4 >> 16) & 0xff);
+ uint32_t a = ((i1 >> 24) & 0xff) + ((i2 >> 24) & 0xff) + ((i3 >> 24) & 0xff) + ((i4 >> 24) & 0xff);
+ return (r >> 2) | ((g >> 2) << 8) | ((b >> 2) << 16) | ((a >> 2) << 24);
+}