From f9b53c3e943c06e982e0abbd2f87245d63c17776 Mon Sep 17 00:00:00 2001
From: Rohit Nirmal <rohitnirmal9@gmail.com>
Date: Sun, 30 Nov 2014 01:44:30 -0600
Subject: Silence a few -Wsign-compare warnings.

---
 src/video_core/debug_utils/debug_utils.cpp | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

(limited to 'src/video_core/debug_utils/debug_utils.cpp')

diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
index 275b06b7c..8a5f11424 100644
--- a/src/video_core/debug_utils/debug_utils.cpp
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -155,7 +155,7 @@ void DumpShader(const u32* binary_data, u32 binary_size, const u32* swizzle_data
 
     // This is put into a try-catch block to make sure we notice unknown configurations.
     std::vector<OutputRegisterInfo> output_info_table;
-        for (int i = 0; i < 7; ++i) {
+        for (unsigned i = 0; i < 7; ++i) {
             using OutputAttributes = Pica::Regs::VSOutputAttributes;
 
             // TODO: It's still unclear how the attribute components map to the register!
@@ -375,8 +375,8 @@ void DumpTexture(const Pica::Regs::TextureConfig& texture_config, u8* data) {
     png_write_info(png_ptr, info_ptr);
 
     buf = new u8[row_stride * texture_config.height];
-    for (int y = 0; y < texture_config.height; ++y) {
-        for (int x = 0; x < texture_config.width; ++x) {
+    for (unsigned y = 0; y < texture_config.height; ++y) {
+        for (unsigned x = 0; x < texture_config.width; ++x) {
             // Cf. rasterizer code for an explanation of this algorithm.
             int texel_index_within_tile = 0;
             for (int block_size_index = 0; block_size_index < 3; ++block_size_index) {
@@ -402,7 +402,7 @@ void DumpTexture(const Pica::Regs::TextureConfig& texture_config, u8* data) {
     }
 
     // Write image data
-    for (auto y = 0; y < texture_config.height; ++y)
+    for (unsigned y = 0; y < texture_config.height; ++y)
     {
         u8* row_ptr = (u8*)buf + y * row_stride;
         u8* ptr = row_ptr;
-- 
cgit v1.2.3


From 2c71ec70527abd091d69f1fdd30aaf95d815214a Mon Sep 17 00:00:00 2001
From: Tony Wasserka <NeoBrainX@gmail.com>
Date: Sat, 25 Oct 2014 18:02:26 +0200
Subject: Pica/DebugUtils: Add breakpoint functionality.

---
 src/video_core/debug_utils/debug_utils.cpp | 43 ++++++++++++++++++++++++++++++
 1 file changed, 43 insertions(+)

(limited to 'src/video_core/debug_utils/debug_utils.cpp')

diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
index 8a5f11424..11f87d988 100644
--- a/src/video_core/debug_utils/debug_utils.cpp
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -3,6 +3,8 @@
 // Refer to the license.txt file included.
 
 #include <algorithm>
+#include <condition_variable>
+#include <list>
 #include <map>
 #include <fstream>
 #include <mutex>
@@ -12,6 +14,7 @@
 #include <png.h>
 #endif
 
+#include "common/log.h"
 #include "common/file_util.h"
 
 #include "video_core/pica.h"
@@ -20,6 +23,46 @@
 
 namespace Pica {
 
+void DebugContext::OnEvent(Event event, void* data) {
+    if (!breakpoints[event].enabled)
+        return;
+
+    {
+        std::unique_lock<std::mutex> lock(breakpoint_mutex);
+
+        // TODO: Should stop the CPU thread here once we multithread emulation.
+
+        active_breakpoint = event;
+        at_breakpoint = true;
+
+        // Tell all observers that we hit a breakpoint
+        for (auto& breakpoint_observer : breakpoint_observers) {
+            breakpoint_observer->OnPicaBreakPointHit(event, data);
+        }
+
+        // Wait until another thread tells us to Resume()
+        resume_from_breakpoint.wait(lock, [&]{ return !at_breakpoint; });
+    }
+}
+
+void DebugContext::Resume() {
+    {
+        std::unique_lock<std::mutex> lock(breakpoint_mutex);
+
+        // Tell all observers that we are about to resume
+        for (auto& breakpoint_observer : breakpoint_observers) {
+            breakpoint_observer->OnPicaResume();
+        }
+
+        // Resume the waiting thread (i.e. OnEvent())
+        at_breakpoint = false;
+    }
+
+    resume_from_breakpoint.notify_one();
+}
+
+std::shared_ptr<DebugContext> g_debug_context; // TODO: Get rid of this global
+
 namespace DebugUtils {
 
 void GeometryDumper::AddTriangle(Vertex& v0, Vertex& v1, Vertex& v2) {
-- 
cgit v1.2.3


From fd194d95b0f1522b970a2b77f9ea490fb8c3ef08 Mon Sep 17 00:00:00 2001
From: Tony Wasserka <NeoBrainX@gmail.com>
Date: Sun, 24 Aug 2014 14:39:52 +0200
Subject: citra-qt: Add texture viewer to Pica command list.

The texture viewer is enabled when selecting a write command to one of the texture config registers.
---
 src/video_core/debug_utils/debug_utils.cpp | 57 +++++++++++++++++++-----------
 1 file changed, 36 insertions(+), 21 deletions(-)

(limited to 'src/video_core/debug_utils/debug_utils.cpp')

diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
index 11f87d988..59909c827 100644
--- a/src/video_core/debug_utils/debug_utils.cpp
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -2,6 +2,8 @@
 // Licensed under GPLv2
 // Refer to the license.txt file included.
 
+#include <cassert>
+
 #include <algorithm>
 #include <condition_variable>
 #include <list>
@@ -17,6 +19,7 @@
 #include "common/log.h"
 #include "common/file_util.h"
 
+#include "video_core/math.h"
 #include "video_core/pica.h"
 
 #include "debug_utils.h"
@@ -355,6 +358,30 @@ std::unique_ptr<PicaTrace> FinishPicaTracing()
     return std::move(ret);
 }
 
+const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const TextureInfo& info) {
+    assert(info.format == Pica::Regs::TextureFormat::RGB8);
+
+    // Cf. rasterizer code for an explanation of this algorithm.
+    int texel_index_within_tile = 0;
+    for (int block_size_index = 0; block_size_index < 3; ++block_size_index) {
+        int sub_tile_width = 1 << block_size_index;
+        int sub_tile_height = 1 << block_size_index;
+
+        int sub_tile_index = (x & sub_tile_width) << block_size_index;
+        sub_tile_index += 2 * ((y & sub_tile_height) << block_size_index);
+        texel_index_within_tile += sub_tile_index;
+    }
+
+    const int block_width = 8;
+    const int block_height = 8;
+
+    int coarse_x = (x / block_width) * block_width;
+    int coarse_y = (y / block_height) * block_height;
+
+    const u8* source_ptr = source + coarse_x * block_height * 3 + coarse_y * info.stride + texel_index_within_tile * 3;
+    return { source_ptr[2], source_ptr[1], source_ptr[0], 255 };
+}
+
 void DumpTexture(const Pica::Regs::TextureConfig& texture_config, u8* data) {
     // NOTE: Permanently enabling this just trashes hard disks for no reason.
     //       Hence, this is currently disabled.
@@ -420,27 +447,15 @@ void DumpTexture(const Pica::Regs::TextureConfig& texture_config, u8* data) {
     buf = new u8[row_stride * texture_config.height];
     for (unsigned y = 0; y < texture_config.height; ++y) {
         for (unsigned x = 0; x < texture_config.width; ++x) {
-            // Cf. rasterizer code for an explanation of this algorithm.
-            int texel_index_within_tile = 0;
-            for (int block_size_index = 0; block_size_index < 3; ++block_size_index) {
-                int sub_tile_width = 1 << block_size_index;
-                int sub_tile_height = 1 << block_size_index;
-
-                int sub_tile_index = (x & sub_tile_width) << block_size_index;
-                sub_tile_index += 2 * ((y & sub_tile_height) << block_size_index);
-                texel_index_within_tile += sub_tile_index;
-            }
-
-            const int block_width = 8;
-            const int block_height = 8;
-
-            int coarse_x = (x / block_width) * block_width;
-            int coarse_y = (y / block_height) * block_height;
-
-            u8* source_ptr = (u8*)data + coarse_x * block_height * 3 + coarse_y * row_stride + texel_index_within_tile * 3;
-            buf[3 * x + y * row_stride    ] = source_ptr[2];
-            buf[3 * x + y * row_stride + 1] = source_ptr[1];
-            buf[3 * x + y * row_stride + 2] = source_ptr[0];
+            TextureInfo info;
+            info.width = texture_config.width;
+            info.height = texture_config.height;
+            info.stride = row_stride;
+            info.format = registers.texture0_format;
+            Math::Vec4<u8> texture_color = LookupTexture(data, x, y, info);
+            buf[3 * x + y * row_stride    ] = texture_color.r();
+            buf[3 * x + y * row_stride + 1] = texture_color.g();
+            buf[3 * x + y * row_stride + 2] = texture_color.b();
         }
     }
 
-- 
cgit v1.2.3


From 2793619dcef9fb2f97db5f0258ca950e18fe7f13 Mon Sep 17 00:00:00 2001
From: Tony Wasserka <NeoBrainX@gmail.com>
Date: Sun, 24 Aug 2014 17:23:02 +0200
Subject: citra_qt: Add enhanced texture debugging widgets.

Double-clicking a texture parameter command in the pica command lists will spawn these as a new tab in the pica command list dock area.
---
 src/video_core/debug_utils/debug_utils.cpp | 12 ++++++++++++
 1 file changed, 12 insertions(+)

(limited to 'src/video_core/debug_utils/debug_utils.cpp')

diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
index 59909c827..31ce09faf 100644
--- a/src/video_core/debug_utils/debug_utils.cpp
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -382,6 +382,18 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
     return { source_ptr[2], source_ptr[1], source_ptr[0], 255 };
 }
 
+TextureInfo TextureInfo::FromPicaRegister(const Regs::TextureConfig& config,
+                                          const Regs::TextureFormat& format)
+{
+    TextureInfo info;
+    info.address = config.GetPhysicalAddress();
+    info.width = config.width;
+    info.height = config.height;
+    info.format = format;
+    info.stride = Pica::Regs::BytesPerPixel(info.format) * info.width;
+    return info;
+}
+
 void DumpTexture(const Pica::Regs::TextureConfig& texture_config, u8* data) {
     // NOTE: Permanently enabling this just trashes hard disks for no reason.
     //       Hence, this is currently disabled.
-- 
cgit v1.2.3


From 0cd27a511ecd170484b672263c09192b579e31ac Mon Sep 17 00:00:00 2001
From: Tony Wasserka <NeoBrainX@gmail.com>
Date: Thu, 4 Dec 2014 19:41:03 +0100
Subject: Some code cleanup.

---
 src/video_core/debug_utils/debug_utils.cpp | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

(limited to 'src/video_core/debug_utils/debug_utils.cpp')

diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
index 31ce09faf..71b03f31c 100644
--- a/src/video_core/debug_utils/debug_utils.cpp
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -2,8 +2,6 @@
 // Licensed under GPLv2
 // Refer to the license.txt file included.
 
-#include <cassert>
-
 #include <algorithm>
 #include <condition_variable>
 #include <list>
@@ -359,7 +357,7 @@ std::unique_ptr<PicaTrace> FinishPicaTracing()
 }
 
 const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const TextureInfo& info) {
-    assert(info.format == Pica::Regs::TextureFormat::RGB8);
+    _dbg_assert_(GPU, info.format == Pica::Regs::TextureFormat::RGB8);
 
     // Cf. rasterizer code for an explanation of this algorithm.
     int texel_index_within_tile = 0;
-- 
cgit v1.2.3


From 0600e2d8b5b30bd68c8b19cb1f2051e096e7caa9 Mon Sep 17 00:00:00 2001
From: Yuri Kunde Schlesner <yuriks@yuriks.net>
Date: Fri, 5 Dec 2014 23:53:49 -0200
Subject: Convert old logging calls to new logging macros

---
 src/video_core/debug_utils/debug_utils.cpp | 18 +++++++++---------
 1 file changed, 9 insertions(+), 9 deletions(-)

(limited to 'src/video_core/debug_utils/debug_utils.cpp')

diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
index 71b03f31c..1a20f19ec 100644
--- a/src/video_core/debug_utils/debug_utils.cpp
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -248,8 +248,8 @@ void DumpShader(const u32* binary_data, u32 binary_size, const u32* swizzle_data
                         it->component_mask = it->component_mask | component_mask;
                     }
                 } catch (const std::out_of_range& ) {
-                    _dbg_assert_msg_(GPU, 0, "Unknown output attribute mapping");
-                    ERROR_LOG(GPU, "Unknown output attribute mapping: %03x, %03x, %03x, %03x",
+                    _dbg_assert_msg_(HW_GPU, 0, "Unknown output attribute mapping");
+                    LOG_ERROR(HW_GPU, "Unknown output attribute mapping: %03x, %03x, %03x, %03x",
                               (int)output_attributes[i].map_x.Value(),
                               (int)output_attributes[i].map_y.Value(),
                               (int)output_attributes[i].map_z.Value(),
@@ -309,7 +309,7 @@ static int is_pica_tracing = false;
 void StartPicaTracing()
 {
     if (is_pica_tracing) {
-        ERROR_LOG(GPU, "StartPicaTracing called even though tracing already running!");
+        LOG_WARNING(HW_GPU, "StartPicaTracing called even though tracing already running!");
         return;
     }
 
@@ -342,7 +342,7 @@ void OnPicaRegWrite(u32 id, u32 value)
 std::unique_ptr<PicaTrace> FinishPicaTracing()
 {
     if (!is_pica_tracing) {
-        ERROR_LOG(GPU, "FinishPicaTracing called even though tracing already running!");
+        LOG_WARNING(HW_GPU, "FinishPicaTracing called even though tracing isn't running!");
         return {};
     }
 
@@ -357,7 +357,7 @@ std::unique_ptr<PicaTrace> FinishPicaTracing()
 }
 
 const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const TextureInfo& info) {
-    _dbg_assert_(GPU, info.format == Pica::Regs::TextureFormat::RGB8);
+    _dbg_assert_(Debug_GPU, info.format == Pica::Regs::TextureFormat::RGB8);
 
     // Cf. rasterizer code for an explanation of this algorithm.
     int texel_index_within_tile = 0;
@@ -421,7 +421,7 @@ void DumpTexture(const Pica::Regs::TextureConfig& texture_config, u8* data) {
     // Initialize write structure
     png_ptr = png_create_write_struct(PNG_LIBPNG_VER_STRING, nullptr, nullptr, nullptr);
     if (png_ptr == nullptr) {
-        ERROR_LOG(GPU, "Could not allocate write struct\n");
+        LOG_ERROR(Debug_GPU, "Could not allocate write struct\n");
         goto finalise;
 
     }
@@ -429,13 +429,13 @@ void DumpTexture(const Pica::Regs::TextureConfig& texture_config, u8* data) {
     // Initialize info structure
     info_ptr = png_create_info_struct(png_ptr);
     if (info_ptr == nullptr) {
-        ERROR_LOG(GPU, "Could not allocate info struct\n");
+        LOG_ERROR(Debug_GPU, "Could not allocate info struct\n");
         goto finalise;
     }
 
     // Setup Exception handling
     if (setjmp(png_jmpbuf(png_ptr))) {
-        ERROR_LOG(GPU, "Error during png creation\n");
+        LOG_ERROR(Debug_GPU, "Error during png creation\n");
         goto finalise;
     }
 
@@ -582,7 +582,7 @@ void DumpTevStageConfig(const std::array<Pica::Regs::TevStageConfig,6>& stages)
         stage_info += "Stage " + std::to_string(index) + ": " + GetColorCombinerStr(tev_stage) + "   " + GetAlphaCombinerStr(tev_stage) + "\n";
     }
 
-    DEBUG_LOG(GPU, "%s", stage_info.c_str());
+    LOG_TRACE(HW_GPU, "%s", stage_info.c_str());
 }
 
 } // namespace
-- 
cgit v1.2.3


From c81f1a9ebc9a5f9df9add64e282d9a0c0da96e79 Mon Sep 17 00:00:00 2001
From: Tony Wasserka <NeoBrainX@gmail.com>
Date: Sat, 6 Dec 2014 21:20:56 +0100
Subject: Pica/DebugUtils: Add support for RGBA8, RGBA5551, RGBA4 and A8
 texture formats.

---
 src/video_core/debug_utils/debug_utils.cpp | 49 ++++++++++++++++++++++++++++--
 1 file changed, 46 insertions(+), 3 deletions(-)

(limited to 'src/video_core/debug_utils/debug_utils.cpp')

diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
index 1a20f19ec..89bf08b99 100644
--- a/src/video_core/debug_utils/debug_utils.cpp
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -357,7 +357,6 @@ std::unique_ptr<PicaTrace> FinishPicaTracing()
 }
 
 const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const TextureInfo& info) {
-    _dbg_assert_(Debug_GPU, info.format == Pica::Regs::TextureFormat::RGB8);
 
     // Cf. rasterizer code for an explanation of this algorithm.
     int texel_index_within_tile = 0;
@@ -376,8 +375,52 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
     int coarse_x = (x / block_width) * block_width;
     int coarse_y = (y / block_height) * block_height;
 
-    const u8* source_ptr = source + coarse_x * block_height * 3 + coarse_y * info.stride + texel_index_within_tile * 3;
-    return { source_ptr[2], source_ptr[1], source_ptr[0], 255 };
+    switch (info.format) {
+    case Regs::TextureFormat::RGBA8:
+    {
+        const u8* source_ptr = source + coarse_x * block_height * 4 + coarse_y * info.stride + texel_index_within_tile * 4;
+        return { source_ptr[3], source_ptr[2], source_ptr[1], 255 };
+    }
+
+    case Regs::TextureFormat::RGB8:
+    {
+        const u8* source_ptr = source + coarse_x * block_height * 3 + coarse_y * info.stride + texel_index_within_tile * 3;
+        return { source_ptr[2], source_ptr[1], source_ptr[0], 255 };
+    }
+
+    case Regs::TextureFormat::RGBA5551:
+    {
+        const u16 source_ptr = *(const u16*)(source + coarse_x * block_height * 2 + coarse_y * info.stride + texel_index_within_tile * 2);
+        u8 r = (source_ptr >> 11) & 0x1F;
+        u8 g = ((source_ptr) >> 6) & 0x1F;
+        u8 b = (source_ptr >> 1) & 0x1F;
+        u8 a = 1;
+        return Math::MakeVec<u8>((r << 3) | (r >> 2), (g << 3) | (g >> 2), (b << 3) | (b >> 2), a * 255);
+    }
+
+    case Regs::TextureFormat::RGBA4:
+    {
+        const u8* source_ptr = source + coarse_x * block_height * 2 + coarse_y * info.stride + texel_index_within_tile * 2;
+        u8 r = source_ptr[1] >> 4;
+        u8 g = source_ptr[1] & 0xFF;
+        u8 b = source_ptr[0] >> 4;
+        r = (r << 4) | r;
+        g = (g << 4) | g;
+        b = (b << 4) | b;
+        return { r, g, b, 255 };
+    }
+
+    case Regs::TextureFormat::A8:
+    {
+        const u8* source_ptr = source + coarse_x * block_height + coarse_y * info.stride + texel_index_within_tile;
+        return { *source_ptr, *source_ptr, *source_ptr, 255 };
+    }
+
+    default:
+        LOG_ERROR(HW_GPU, "Unknown texture format: %x", (u32)info.format);
+        _dbg_assert_(HW_GPU, 0);
+        return {};
+    }
 }
 
 TextureInfo TextureInfo::FromPicaRegister(const Regs::TextureConfig& config,
-- 
cgit v1.2.3


From 3df88d59b0ba43f1c3360cfdaaccd461cacff72c Mon Sep 17 00:00:00 2001
From: Tony Wasserka <NeoBrainX@gmail.com>
Date: Sat, 6 Dec 2014 21:52:21 +0100
Subject: Pica: Merge texture lookup logic for DebugUtils and Rasterizer.

This effectively adds support for a lot texture formats in the rasterizer.
---
 src/video_core/debug_utils/debug_utils.cpp | 44 ++++++++++++++++++++++++------
 1 file changed, 36 insertions(+), 8 deletions(-)

(limited to 'src/video_core/debug_utils/debug_utils.cpp')

diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
index 89bf08b99..6c26138da 100644
--- a/src/video_core/debug_utils/debug_utils.cpp
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -356,9 +356,29 @@ std::unique_ptr<PicaTrace> FinishPicaTracing()
     return std::move(ret);
 }
 
-const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const TextureInfo& info) {
-
-    // Cf. rasterizer code for an explanation of this algorithm.
+const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const TextureInfo& info, bool disable_alpha) {
+
+    // Images are split into 8x8 tiles. Each tile is composed of four 4x4 subtiles each
+    // of which is composed of four 2x2 subtiles each of which is composed of four texels.
+    // Each structure is embedded into the next-bigger one in a diagonal pattern, e.g.
+    // texels are laid out in a 2x2 subtile like this:
+    // 2 3
+    // 0 1
+    //
+    // The full 8x8 tile has the texels arranged like this:
+    //
+    // 42 43 46 47 58 59 62 63
+    // 40 41 44 45 56 57 60 61
+    // 34 35 38 39 50 51 54 55
+    // 32 33 36 37 48 49 52 53
+    // 10 11 14 15 26 27 30 31
+    // 08 09 12 13 24 25 28 29
+    // 02 03 06 07 18 19 22 23
+    // 00 01 04 05 16 17 20 21
+
+    // TODO(neobrain): Not sure if this swizzling pattern is used for all textures.
+    // To be flexible in case different but similar patterns are used, we keep this
+    // somewhat inefficient code around for now.
     int texel_index_within_tile = 0;
     for (int block_size_index = 0; block_size_index < 3; ++block_size_index) {
         int sub_tile_width = 1 << block_size_index;
@@ -379,7 +399,7 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
     case Regs::TextureFormat::RGBA8:
     {
         const u8* source_ptr = source + coarse_x * block_height * 4 + coarse_y * info.stride + texel_index_within_tile * 4;
-        return { source_ptr[3], source_ptr[2], source_ptr[1], 255 };
+        return { source_ptr[3], source_ptr[2], source_ptr[1], disable_alpha ? 255 : source_ptr[0] };
     }
 
     case Regs::TextureFormat::RGB8:
@@ -394,8 +414,8 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
         u8 r = (source_ptr >> 11) & 0x1F;
         u8 g = ((source_ptr) >> 6) & 0x1F;
         u8 b = (source_ptr >> 1) & 0x1F;
-        u8 a = 1;
-        return Math::MakeVec<u8>((r << 3) | (r >> 2), (g << 3) | (g >> 2), (b << 3) | (b >> 2), a * 255);
+        u8 a = source_ptr & 1;
+        return Math::MakeVec<u8>((r << 3) | (r >> 2), (g << 3) | (g >> 2), (b << 3) | (b >> 2), disable_alpha ? 255 : (a * 255));
     }
 
     case Regs::TextureFormat::RGBA4:
@@ -404,16 +424,24 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
         u8 r = source_ptr[1] >> 4;
         u8 g = source_ptr[1] & 0xFF;
         u8 b = source_ptr[0] >> 4;
+        u8 a = source_ptr[0] & 0xFF;
         r = (r << 4) | r;
         g = (g << 4) | g;
         b = (b << 4) | b;
-        return { r, g, b, 255 };
+        a = (a << 4) | a;
+        return { r, g, b, disable_alpha ? 255 : a };
     }
 
     case Regs::TextureFormat::A8:
     {
         const u8* source_ptr = source + coarse_x * block_height + coarse_y * info.stride + texel_index_within_tile;
-        return { *source_ptr, *source_ptr, *source_ptr, 255 };
+
+        // TODO: Better control this...
+        if (disable_alpha) {
+            return { *source_ptr, *source_ptr, *source_ptr, 255 };
+        } else {
+            return { 0, 0, 0, *source_ptr };
+        }
     }
 
     default:
-- 
cgit v1.2.3


From 7e210e0229b9caef77c80fea7c056c3913e68129 Mon Sep 17 00:00:00 2001
From: Tony Wasserka <NeoBrainX@gmail.com>
Date: Fri, 24 Oct 2014 00:58:04 +0200
Subject: Pica: Further improve Tev emulation.

---
 src/video_core/debug_utils/debug_utils.cpp | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

(limited to 'src/video_core/debug_utils/debug_utils.cpp')

diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
index 6c26138da..3cc22f436 100644
--- a/src/video_core/debug_utils/debug_utils.cpp
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -573,20 +573,26 @@ void DumpTevStageConfig(const std::array<Pica::Regs::TevStageConfig,6>& stages)
         const std::map<Source, std::string> source_map = {
             { Source::PrimaryColor, "PrimaryColor" },
             { Source::Texture0, "Texture0" },
+            { Source::Texture1, "Texture1" },
+            { Source::Texture2, "Texture2" },
             { Source::Constant, "Constant" },
             { Source::Previous, "Previous" },
         };
 
         const std::map<ColorModifier, std::string> color_modifier_map = {
-            { ColorModifier::SourceColor, { "%source.rgb" } }
+            { ColorModifier::SourceColor, { "%source.rgb" } },
+            { ColorModifier::SourceAlpha, { "%source.aaa" } },
         };
         const std::map<AlphaModifier, std::string> alpha_modifier_map = {
-            { AlphaModifier::SourceAlpha, "%source.a" }
+            { AlphaModifier::SourceAlpha, "%source.a" },
+            { AlphaModifier::OneMinusSourceAlpha, "(255 - %source.a)" },
         };
 
         std::map<Operation, std::string> combiner_map = {
             { Operation::Replace, "%source1" },
             { Operation::Modulate, "(%source1 * %source2) / 255" },
+            { Operation::Add, "(%source1 + %source2)" },
+            { Operation::Lerp, "lerp(%source1, %source2, %source3)" },
         };
 
         auto ReplacePattern =
-- 
cgit v1.2.3


From 40f123b7c0eaf1507d51f6b87192ec2f956e5d5e Mon Sep 17 00:00:00 2001
From: Tony Wasserka <NeoBrainX@gmail.com>
Date: Mon, 15 Dec 2014 21:28:45 +0100
Subject: Pica: Unify ugly address translation hacks.

---
 src/video_core/debug_utils/debug_utils.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'src/video_core/debug_utils/debug_utils.cpp')

diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
index 3cc22f436..08ecd4ccb 100644
--- a/src/video_core/debug_utils/debug_utils.cpp
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -455,7 +455,7 @@ TextureInfo TextureInfo::FromPicaRegister(const Regs::TextureConfig& config,
                                           const Regs::TextureFormat& format)
 {
     TextureInfo info;
-    info.address = config.GetPhysicalAddress();
+    info.physical_address = config.GetPhysicalAddress();
     info.width = config.width;
     info.height = config.height;
     info.format = format;
-- 
cgit v1.2.3


From 1c972ef3b93252a157ec15d0878a2be3e4b46a0e Mon Sep 17 00:00:00 2001
From: Tony Wasserka <NeoBrainX@gmail.com>
Date: Wed, 10 Dec 2014 21:51:00 +0100
Subject: Add support for a ridiculous number of texture formats.

---
 src/video_core/debug_utils/debug_utils.cpp | 65 +++++++++++++++++++++++++++++-
 1 file changed, 64 insertions(+), 1 deletion(-)

(limited to 'src/video_core/debug_utils/debug_utils.cpp')

diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
index 08ecd4ccb..1a7b851d5 100644
--- a/src/video_core/debug_utils/debug_utils.cpp
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -418,6 +418,15 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
         return Math::MakeVec<u8>((r << 3) | (r >> 2), (g << 3) | (g >> 2), (b << 3) | (b >> 2), disable_alpha ? 255 : (a * 255));
     }
 
+    case Regs::TextureFormat::RGB565:
+    {
+        const u16 source_ptr = *(const u16*)(source + coarse_x * block_height * 2 + coarse_y * info.stride + texel_index_within_tile * 2);
+        u8 r = (source_ptr >> 11) & 0x1F;
+        u8 g = ((source_ptr) >> 5) & 0x3F;
+        u8 b = (source_ptr) & 0x1F;
+        return Math::MakeVec<u8>((r << 3) | (r >> 2), (g << 2) | (g >> 4), (b << 3) | (b >> 2), 255);
+    }
+
     case Regs::TextureFormat::RGBA4:
     {
         const u8* source_ptr = source + coarse_x * block_height * 2 + coarse_y * info.stride + texel_index_within_tile * 2;
@@ -432,6 +441,26 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
         return { r, g, b, disable_alpha ? 255 : a };
     }
 
+    case Regs::TextureFormat::IA8:
+    {
+        const u8* source_ptr = source + coarse_x * block_height * 2 + coarse_y * info.stride + texel_index_within_tile * 2;
+
+        // TODO: Better control this...
+        if (disable_alpha) {
+            return { *source_ptr, *(source_ptr+1), 0, 255 };
+        } else {
+            return { *source_ptr, *source_ptr, *source_ptr, *(source_ptr+1)};
+        }
+    }
+
+    case Regs::TextureFormat::I8:
+    {
+        const u8* source_ptr = source + coarse_x * block_height + coarse_y * info.stride + texel_index_within_tile;
+
+        // TODO: Better control this...
+        return { *source_ptr, *source_ptr, *source_ptr, 255 };
+    }
+
     case Regs::TextureFormat::A8:
     {
         const u8* source_ptr = source + coarse_x * block_height + coarse_y * info.stride + texel_index_within_tile;
@@ -444,6 +473,40 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
         }
     }
 
+    case Regs::TextureFormat::IA4:
+    {
+        const u8* source_ptr = source + coarse_x * block_height / 2 + coarse_y * info.stride + texel_index_within_tile / 2;
+
+        // TODO: Order?
+        u8 i = (*source_ptr)&0xF;
+        u8 a = ((*source_ptr) & 0xF0) >> 4;
+        a |= a << 4;
+        i |= i << 4;
+
+        // TODO: Better control this...
+        if (disable_alpha) {
+            return { i, a, 0, 255 };
+        } else {
+            return { i, i, i, a };
+        }
+    }
+
+    case Regs::TextureFormat::A4:
+    {
+        const u8* source_ptr = source + coarse_x * block_height / 2 + coarse_y * info.stride + texel_index_within_tile / 2;
+
+        // TODO: Order?
+        u8 a = (coarse_x % 2) ? ((*source_ptr)&0xF) : (((*source_ptr) & 0xF0) >> 4);
+        a |= a << 4;
+
+        // TODO: Better control this...
+        if (disable_alpha) {
+            return { *source_ptr, *source_ptr, *source_ptr, 255 };
+        } else {
+            return { 0, 0, 0, *source_ptr };
+        }
+    }
+
     default:
         LOG_ERROR(HW_GPU, "Unknown texture format: %x", (u32)info.format);
         _dbg_assert_(HW_GPU, 0);
@@ -459,7 +522,7 @@ TextureInfo TextureInfo::FromPicaRegister(const Regs::TextureConfig& config,
     info.width = config.width;
     info.height = config.height;
     info.format = format;
-    info.stride = Pica::Regs::BytesPerPixel(info.format) * info.width;
+    info.stride = Pica::Regs::NibblesPerPixel(info.format) * info.width / 2;
     return info;
 }
 
-- 
cgit v1.2.3


From cc5746abfe838fa130dd8be58219e00ae292a8fe Mon Sep 17 00:00:00 2001
From: Tony Wasserka <NeoBrainX@gmail.com>
Date: Tue, 16 Dec 2014 01:12:29 +0100
Subject: Pica/DebugUtils: Replace duplicated SHBIN structures in favor of
 nihstro's ones.

---
 src/video_core/debug_utils/debug_utils.cpp | 69 ++++--------------------------
 1 file changed, 8 insertions(+), 61 deletions(-)

(limited to 'src/video_core/debug_utils/debug_utils.cpp')

diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
index 1a7b851d5..7e1cfb92c 100644
--- a/src/video_core/debug_utils/debug_utils.cpp
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -14,6 +14,8 @@
 #include <png.h>
 #endif
 
+#include <nihstro/shader_binary.h>
+
 #include "common/log.h"
 #include "common/file_util.h"
 
@@ -22,6 +24,10 @@
 
 #include "debug_utils.h"
 
+using nihstro::DVLBHeader;
+using nihstro::DVLEHeader;
+using nihstro::DVLPHeader;
+
 namespace Pica {
 
 void DebugContext::OnEvent(Event event, void* data) {
@@ -98,65 +104,6 @@ void GeometryDumper::Dump() {
     }
 }
 
-#pragma pack(1)
-struct DVLBHeader {
-    enum : u32 {
-        MAGIC_WORD = 0x424C5644, // "DVLB"
-    };
-
-    u32 magic_word;
-    u32 num_programs;
-//    u32 dvle_offset_table[];
-};
-static_assert(sizeof(DVLBHeader) == 0x8, "Incorrect structure size");
-
-struct DVLPHeader {
-    enum : u32 {
-        MAGIC_WORD = 0x504C5644, // "DVLP"
-    };
-
-    u32 magic_word;
-    u32 version;
-    u32 binary_offset;  // relative to DVLP start
-    u32 binary_size_words;
-    u32 swizzle_patterns_offset;
-    u32 swizzle_patterns_num_entries;
-    u32 unk2;
-};
-static_assert(sizeof(DVLPHeader) == 0x1C, "Incorrect structure size");
-
-struct DVLEHeader {
-    enum : u32 {
-        MAGIC_WORD = 0x454c5644, // "DVLE"
-    };
-
-    enum class ShaderType : u8 {
-        VERTEX = 0,
-        GEOMETRY = 1,
-    };
-
-    u32 magic_word;
-    u16 pad1;
-    ShaderType type;
-    u8 pad2;
-    u32 main_offset_words; // offset within binary blob
-    u32 endmain_offset_words;
-    u32 pad3;
-    u32 pad4;
-    u32 constant_table_offset;
-    u32 constant_table_size; // number of entries
-    u32 label_table_offset;
-    u32 label_table_size;
-    u32 output_register_table_offset;
-    u32 output_register_table_size;
-    u32 uniform_table_offset;
-    u32 uniform_table_size;
-    u32 symbol_table_offset;
-    u32 symbol_table_size;
-
-};
-static_assert(sizeof(DVLEHeader) == 0x40, "Incorrect structure size");
-#pragma pack()
 
 void DumpShader(const u32* binary_data, u32 binary_size, const u32* swizzle_data, u32 swizzle_size,
                 u32 main_offset, const Regs::VSOutputAttributes* output_attributes)
@@ -276,8 +223,8 @@ void DumpShader(const u32* binary_data, u32 binary_size, const u32* swizzle_data
     dvlp.binary_size_words = binary_size;
     QueueForWriting((u8*)binary_data, binary_size * sizeof(u32));
 
-    dvlp.swizzle_patterns_offset = write_offset - dvlp_offset;
-    dvlp.swizzle_patterns_num_entries = swizzle_size;
+    dvlp.swizzle_info_offset = write_offset - dvlp_offset;
+    dvlp.swizzle_info_num_entries = swizzle_size;
     u32 dummy = 0;
     for (unsigned int i = 0; i < swizzle_size; ++i) {
         QueueForWriting((u8*)&swizzle_data[i], sizeof(swizzle_data[i]));
-- 
cgit v1.2.3


From d81370682fccda1370ba22026aa21a260b506efd Mon Sep 17 00:00:00 2001
From: Tony Wasserka <NeoBrainX@gmail.com>
Date: Fri, 19 Dec 2014 18:49:09 +0100
Subject: Pica/DebugUtils: Make a number of variables static.

Makes for cleaner and faster code.
---
 src/video_core/debug_utils/debug_utils.cpp | 26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)

(limited to 'src/video_core/debug_utils/debug_utils.cpp')

diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
index 7e1cfb92c..0085c117d 100644
--- a/src/video_core/debug_utils/debug_utils.cpp
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -580,7 +580,7 @@ void DumpTevStageConfig(const std::array<Pica::Regs::TevStageConfig,6>& stages)
     for (size_t index = 0; index < stages.size(); ++index) {
         const auto& tev_stage = stages[index];
 
-        const std::map<Source, std::string> source_map = {
+        static const std::map<Source, std::string> source_map = {
             { Source::PrimaryColor, "PrimaryColor" },
             { Source::Texture0, "Texture0" },
             { Source::Texture1, "Texture1" },
@@ -589,23 +589,23 @@ void DumpTevStageConfig(const std::array<Pica::Regs::TevStageConfig,6>& stages)
             { Source::Previous, "Previous" },
         };
 
-        const std::map<ColorModifier, std::string> color_modifier_map = {
+        static const std::map<ColorModifier, std::string> color_modifier_map = {
             { ColorModifier::SourceColor, { "%source.rgb" } },
             { ColorModifier::SourceAlpha, { "%source.aaa" } },
         };
-        const std::map<AlphaModifier, std::string> alpha_modifier_map = {
+        static const std::map<AlphaModifier, std::string> alpha_modifier_map = {
             { AlphaModifier::SourceAlpha, "%source.a" },
             { AlphaModifier::OneMinusSourceAlpha, "(255 - %source.a)" },
         };
 
-        std::map<Operation, std::string> combiner_map = {
+        static const std::map<Operation, std::string> combiner_map = {
             { Operation::Replace, "%source1" },
             { Operation::Modulate, "(%source1 * %source2) / 255" },
             { Operation::Add, "(%source1 + %source2)" },
             { Operation::Lerp, "lerp(%source1, %source2, %source3)" },
         };
 
-        auto ReplacePattern =
+        static auto ReplacePattern =
                 [](const std::string& input, const std::string& pattern, const std::string& replacement) -> std::string {
                     size_t start = input.find(pattern);
                     if (start == std::string::npos)
@@ -615,8 +615,8 @@ void DumpTevStageConfig(const std::array<Pica::Regs::TevStageConfig,6>& stages)
                     ret.replace(start, pattern.length(), replacement);
                     return ret;
                 };
-        auto GetColorSourceStr =
-                [&source_map,&color_modifier_map,&ReplacePattern](const Source& src, const ColorModifier& modifier) {
+        static auto GetColorSourceStr =
+                [](const Source& src, const ColorModifier& modifier) {
                     auto src_it = source_map.find(src);
                     std::string src_str = "Unknown";
                     if (src_it != source_map.end())
@@ -629,8 +629,8 @@ void DumpTevStageConfig(const std::array<Pica::Regs::TevStageConfig,6>& stages)
 
                     return ReplacePattern(modifier_str, "%source", src_str);
                 };
-        auto GetColorCombinerStr =
-                [&](const Regs::TevStageConfig& tev_stage) {
+        static auto GetColorCombinerStr =
+                [](const Regs::TevStageConfig& tev_stage) {
                     auto op_it = combiner_map.find(tev_stage.color_op);
                     std::string op_str = "Unknown op (%source1, %source2, %source3)";
                     if (op_it != combiner_map.end())
@@ -640,8 +640,8 @@ void DumpTevStageConfig(const std::array<Pica::Regs::TevStageConfig,6>& stages)
                     op_str = ReplacePattern(op_str, "%source2", GetColorSourceStr(tev_stage.color_source2, tev_stage.color_modifier2));
                     return   ReplacePattern(op_str, "%source3", GetColorSourceStr(tev_stage.color_source3, tev_stage.color_modifier3));
                 };
-        auto GetAlphaSourceStr =
-                [&source_map,&alpha_modifier_map,&ReplacePattern](const Source& src, const AlphaModifier& modifier) {
+        static auto GetAlphaSourceStr =
+                [](const Source& src, const AlphaModifier& modifier) {
                     auto src_it = source_map.find(src);
                     std::string src_str = "Unknown";
                     if (src_it != source_map.end())
@@ -654,8 +654,8 @@ void DumpTevStageConfig(const std::array<Pica::Regs::TevStageConfig,6>& stages)
 
                     return ReplacePattern(modifier_str, "%source", src_str);
                 };
-        auto GetAlphaCombinerStr =
-                [&](const Regs::TevStageConfig& tev_stage) {
+        static auto GetAlphaCombinerStr =
+                [](const Regs::TevStageConfig& tev_stage) {
                     auto op_it = combiner_map.find(tev_stage.alpha_op);
                     std::string op_str = "Unknown op (%source1, %source2, %source3)";
                     if (op_it != combiner_map.end())
-- 
cgit v1.2.3


From 6e275778c9e7e55cabadb14fdabaa51a55348663 Mon Sep 17 00:00:00 2001
From: Tony Wasserka <NeoBrainX@gmail.com>
Date: Fri, 19 Dec 2014 19:15:47 +0100
Subject: Pica/DebugUtils: Better document LookupTexture.

---
 src/video_core/debug_utils/debug_utils.cpp | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

(limited to 'src/video_core/debug_utils/debug_utils.cpp')

diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
index 0085c117d..1c08ba350 100644
--- a/src/video_core/debug_utils/debug_utils.cpp
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -392,8 +392,10 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
     {
         const u8* source_ptr = source + coarse_x * block_height * 2 + coarse_y * info.stride + texel_index_within_tile * 2;
 
-        // TODO: Better control this...
+        // TODO: compoent order not verified
+
         if (disable_alpha) {
+            // Show intensity as red, alpha as green
             return { *source_ptr, *(source_ptr+1), 0, 255 };
         } else {
             return { *source_ptr, *source_ptr, *source_ptr, *(source_ptr+1)};
@@ -403,8 +405,6 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
     case Regs::TextureFormat::I8:
     {
         const u8* source_ptr = source + coarse_x * block_height + coarse_y * info.stride + texel_index_within_tile;
-
-        // TODO: Better control this...
         return { *source_ptr, *source_ptr, *source_ptr, 255 };
     }
 
@@ -412,7 +412,6 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
     {
         const u8* source_ptr = source + coarse_x * block_height + coarse_y * info.stride + texel_index_within_tile;
 
-        // TODO: Better control this...
         if (disable_alpha) {
             return { *source_ptr, *source_ptr, *source_ptr, 255 };
         } else {
@@ -424,14 +423,15 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
     {
         const u8* source_ptr = source + coarse_x * block_height / 2 + coarse_y * info.stride + texel_index_within_tile / 2;
 
-        // TODO: Order?
+        // TODO: compoent order not verified
+
         u8 i = (*source_ptr)&0xF;
         u8 a = ((*source_ptr) & 0xF0) >> 4;
         a |= a << 4;
         i |= i << 4;
 
-        // TODO: Better control this...
         if (disable_alpha) {
+            // Show intensity as red, alpha as green
             return { i, a, 0, 255 };
         } else {
             return { i, i, i, a };
-- 
cgit v1.2.3


From 88e9efe4b8b370a93bae688dcbe3c03eda905379 Mon Sep 17 00:00:00 2001
From: Tony Wasserka <NeoBrainX@gmail.com>
Date: Fri, 19 Dec 2014 19:20:02 +0100
Subject: Pica/DebugUtils: Fix two warnings.

---
 src/video_core/debug_utils/debug_utils.cpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

(limited to 'src/video_core/debug_utils/debug_utils.cpp')

diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
index 1c08ba350..d9fed58bf 100644
--- a/src/video_core/debug_utils/debug_utils.cpp
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -346,7 +346,7 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
     case Regs::TextureFormat::RGBA8:
     {
         const u8* source_ptr = source + coarse_x * block_height * 4 + coarse_y * info.stride + texel_index_within_tile * 4;
-        return { source_ptr[3], source_ptr[2], source_ptr[1], disable_alpha ? 255 : source_ptr[0] };
+        return { source_ptr[3], source_ptr[2], source_ptr[1], disable_alpha ? (u8)255 : source_ptr[0] };
     }
 
     case Regs::TextureFormat::RGB8:
@@ -385,7 +385,7 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
         g = (g << 4) | g;
         b = (b << 4) | b;
         a = (a << 4) | a;
-        return { r, g, b, disable_alpha ? 255 : a };
+        return { r, g, b, disable_alpha ? (u8)255 : a };
     }
 
     case Regs::TextureFormat::IA8:
-- 
cgit v1.2.3


From 871418e62b079a83d9121dca0ef75b91acbe77cd Mon Sep 17 00:00:00 2001
From: Tony Wasserka <NeoBrainX@gmail.com>
Date: Fri, 19 Dec 2014 19:37:37 +0100
Subject: Pica/DebugUtils: Further cleanups to LookupTexture.

---
 src/video_core/debug_utils/debug_utils.cpp | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

(limited to 'src/video_core/debug_utils/debug_utils.cpp')

diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
index d9fed58bf..328386b7e 100644
--- a/src/video_core/debug_utils/debug_utils.cpp
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -392,13 +392,13 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
     {
         const u8* source_ptr = source + coarse_x * block_height * 2 + coarse_y * info.stride + texel_index_within_tile * 2;
 
-        // TODO: compoent order not verified
+        // TODO: component order not verified
 
         if (disable_alpha) {
             // Show intensity as red, alpha as green
-            return { *source_ptr, *(source_ptr+1), 0, 255 };
+            return { source_ptr[0], source_ptr[1], 0, 255 };
         } else {
-            return { *source_ptr, *source_ptr, *source_ptr, *(source_ptr+1)};
+            return { source_ptr[0], source_ptr[0], source_ptr[0], source_ptr[1]};
         }
     }
 
@@ -423,9 +423,9 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
     {
         const u8* source_ptr = source + coarse_x * block_height / 2 + coarse_y * info.stride + texel_index_within_tile / 2;
 
-        // TODO: compoent order not verified
+        // TODO: component order not verified
 
-        u8 i = (*source_ptr)&0xF;
+        u8 i = (*source_ptr) & 0xF;
         u8 a = ((*source_ptr) & 0xF0) >> 4;
         a |= a << 4;
         i |= i << 4;
@@ -442,11 +442,11 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
     {
         const u8* source_ptr = source + coarse_x * block_height / 2 + coarse_y * info.stride + texel_index_within_tile / 2;
 
-        // TODO: Order?
+        // TODO: component order not verified
+
         u8 a = (coarse_x % 2) ? ((*source_ptr)&0xF) : (((*source_ptr) & 0xF0) >> 4);
         a |= a << 4;
 
-        // TODO: Better control this...
         if (disable_alpha) {
             return { *source_ptr, *source_ptr, *source_ptr, 255 };
         } else {
-- 
cgit v1.2.3


From fe186d3a598837ba7337f06399dfb8ae7930a070 Mon Sep 17 00:00:00 2001
From: Yuri Kunde Schlesner <yuriks@yuriks.net>
Date: Tue, 23 Dec 2014 12:27:56 -0200
Subject: GPU: Bitwise texture swizzling

Replace the loop-based texture address swizzling code by a bit-twiddling
implementation, providing a very small speed up. Also simplify
addressing code.
---
 src/video_core/debug_utils/debug_utils.cpp | 51 ++++++++++++++----------------
 1 file changed, 24 insertions(+), 27 deletions(-)

(limited to 'src/video_core/debug_utils/debug_utils.cpp')

diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
index 328386b7e..5921185a6 100644
--- a/src/video_core/debug_utils/debug_utils.cpp
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -304,7 +304,6 @@ std::unique_ptr<PicaTrace> FinishPicaTracing()
 }
 
 const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const TextureInfo& info, bool disable_alpha) {
-
     // Images are split into 8x8 tiles. Each tile is composed of four 4x4 subtiles each
     // of which is composed of four 2x2 subtiles each of which is composed of four texels.
     // Each structure is embedded into the next-bigger one in a diagonal pattern, e.g.
@@ -323,41 +322,39 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
     // 02 03 06 07 18 19 22 23
     // 00 01 04 05 16 17 20 21
 
-    // TODO(neobrain): Not sure if this swizzling pattern is used for all textures.
-    // To be flexible in case different but similar patterns are used, we keep this
-    // somewhat inefficient code around for now.
-    int texel_index_within_tile = 0;
-    for (int block_size_index = 0; block_size_index < 3; ++block_size_index) {
-        int sub_tile_width = 1 << block_size_index;
-        int sub_tile_height = 1 << block_size_index;
-
-        int sub_tile_index = (x & sub_tile_width) << block_size_index;
-        sub_tile_index += 2 * ((y & sub_tile_height) << block_size_index);
-        texel_index_within_tile += sub_tile_index;
-    }
+    const unsigned int block_width = 8;
+    const unsigned int block_height = 8;
+
+    const unsigned int coarse_x = x & ~7;
+    const unsigned int coarse_y = y & ~7;
 
-    const int block_width = 8;
-    const int block_height = 8;
+    // Interleave the lower 3 bits of each coordinate to get the intra-block offsets, which are
+    // arranged in a Z-order curve. More details on the bit manipulation at:
+    // https://fgiesen.wordpress.com/2009/12/13/decoding-morton-codes/
+    unsigned int i = (x | (y << 8)) & 0x0707; // ---- -210
+    i = (i ^ (i << 2)) & 0x1313;              // ---2 --10
+    i = (i ^ (i << 1)) & 0x1515;              // ---2 -1-0
+    i = (i | (i >> 7)) & 0x3F;
 
-    int coarse_x = (x / block_width) * block_width;
-    int coarse_y = (y / block_height) * block_height;
+    source += coarse_y * info.stride;
+    const unsigned int offset = coarse_x * block_height + i;
 
     switch (info.format) {
     case Regs::TextureFormat::RGBA8:
     {
-        const u8* source_ptr = source + coarse_x * block_height * 4 + coarse_y * info.stride + texel_index_within_tile * 4;
+        const u8* source_ptr = source + offset * 4;
         return { source_ptr[3], source_ptr[2], source_ptr[1], disable_alpha ? (u8)255 : source_ptr[0] };
     }
 
     case Regs::TextureFormat::RGB8:
     {
-        const u8* source_ptr = source + coarse_x * block_height * 3 + coarse_y * info.stride + texel_index_within_tile * 3;
+        const u8* source_ptr = source + offset * 3;
         return { source_ptr[2], source_ptr[1], source_ptr[0], 255 };
     }
 
     case Regs::TextureFormat::RGBA5551:
     {
-        const u16 source_ptr = *(const u16*)(source + coarse_x * block_height * 2 + coarse_y * info.stride + texel_index_within_tile * 2);
+        const u16 source_ptr = *(const u16*)(source + offset * 2);
         u8 r = (source_ptr >> 11) & 0x1F;
         u8 g = ((source_ptr) >> 6) & 0x1F;
         u8 b = (source_ptr >> 1) & 0x1F;
@@ -367,7 +364,7 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
 
     case Regs::TextureFormat::RGB565:
     {
-        const u16 source_ptr = *(const u16*)(source + coarse_x * block_height * 2 + coarse_y * info.stride + texel_index_within_tile * 2);
+        const u16 source_ptr = *(const u16*)(source + offset * 2);
         u8 r = (source_ptr >> 11) & 0x1F;
         u8 g = ((source_ptr) >> 5) & 0x3F;
         u8 b = (source_ptr) & 0x1F;
@@ -376,7 +373,7 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
 
     case Regs::TextureFormat::RGBA4:
     {
-        const u8* source_ptr = source + coarse_x * block_height * 2 + coarse_y * info.stride + texel_index_within_tile * 2;
+        const u8* source_ptr = source + offset * 2;
         u8 r = source_ptr[1] >> 4;
         u8 g = source_ptr[1] & 0xFF;
         u8 b = source_ptr[0] >> 4;
@@ -390,7 +387,7 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
 
     case Regs::TextureFormat::IA8:
     {
-        const u8* source_ptr = source + coarse_x * block_height * 2 + coarse_y * info.stride + texel_index_within_tile * 2;
+        const u8* source_ptr = source + offset * 2;
 
         // TODO: component order not verified
 
@@ -404,13 +401,13 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
 
     case Regs::TextureFormat::I8:
     {
-        const u8* source_ptr = source + coarse_x * block_height + coarse_y * info.stride + texel_index_within_tile;
+        const u8* source_ptr = source + offset;
         return { *source_ptr, *source_ptr, *source_ptr, 255 };
     }
 
     case Regs::TextureFormat::A8:
     {
-        const u8* source_ptr = source + coarse_x * block_height + coarse_y * info.stride + texel_index_within_tile;
+        const u8* source_ptr = source + offset;
 
         if (disable_alpha) {
             return { *source_ptr, *source_ptr, *source_ptr, 255 };
@@ -421,7 +418,7 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
 
     case Regs::TextureFormat::IA4:
     {
-        const u8* source_ptr = source + coarse_x * block_height / 2 + coarse_y * info.stride + texel_index_within_tile / 2;
+        const u8* source_ptr = source + offset / 2;
 
         // TODO: component order not verified
 
@@ -440,7 +437,7 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
 
     case Regs::TextureFormat::A4:
     {
-        const u8* source_ptr = source + coarse_x * block_height / 2 + coarse_y * info.stride + texel_index_within_tile / 2;
+        const u8* source_ptr = source + offset / 2;
 
         // TODO: component order not verified
 
-- 
cgit v1.2.3