diff --git a/CMakeLists.txt b/CMakeLists.txt
index 83f56c3aa1..a794d97f18 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -892,6 +892,8 @@ add_library(native STATIC
 	ext/native/math/fast/fast_matrix.c
 	ext/native/math/fast/fast_matrix_neon.S
 	ext/native/math/fast/fast_matrix_sse.c
+	ext/native/math/dataconv.cpp
+	ext/native/math/dataconv.h
 	ext/native/math/curves.cpp
 	ext/native/math/curves.h
 	ext/native/math/expression_parser.cpp
diff --git a/Core/HLE/HLE.cpp b/Core/HLE/HLE.cpp
index 5c6610a780..e36a8c4109 100644
--- a/Core/HLE/HLE.cpp
+++ b/Core/HLE/HLE.cpp
@@ -519,8 +519,7 @@ void CallSyscall(MIPSOpcode op)
 {
 	PROFILE_THIS_SCOPE("syscall");
 	double start = 0.0;  // need to initialize to fix the race condition where coreCollectDebugStats is enabled in the middle of this func.
-	if (coreCollectDebugStats)
-	{
+	if (coreCollectDebugStats) {
 		time_update();
 		start = time_now_d();
 	}
@@ -544,8 +543,7 @@ void CallSyscall(MIPSOpcode op)
 		ERROR_LOG_REPORT(HLE, "Unimplemented HLE function %s", info->name ? info->name : "(\?\?\?)");
 	}
 
-	if (coreCollectDebugStats)
-	{
+	if (coreCollectDebugStats) {
 		time_update();
 		u32 callno = (op >> 6) & 0xFFFFF; //20 bits
 		int funcnum = callno & 0xFFF;
diff --git a/GPU/Common/ShaderCommon.h b/GPU/Common/ShaderCommon.h
index ac44251b05..5f17e236dc 100644
--- a/GPU/Common/ShaderCommon.h
+++ b/GPU/Common/ShaderCommon.h
@@ -94,6 +94,7 @@ enum : uint64_t {
 	DIRTY_BONE_UNIFORMS = 0xFF000000ULL,
 
 	DIRTY_ALL_UNIFORMS = 0x1FFFFFFFFULL,
+	DIRTY_ALL_LIGHTS = DIRTY_LIGHT0 | DIRTY_LIGHT1 | DIRTY_LIGHT2 | DIRTY_LIGHT3,
 
 	// Other dirty elements that aren't uniforms!
 	DIRTY_FRAMEBUF = 1ULL << 40,
diff --git a/GPU/Common/ShaderUniforms.cpp b/GPU/Common/ShaderUniforms.cpp
index e92448c7ed..4e3a3deab4 100644
--- a/GPU/Common/ShaderUniforms.cpp
+++ b/GPU/Common/ShaderUniforms.cpp
@@ -219,7 +219,6 @@ void LightUpdateUniforms(UB_VS_Lights *ub, uint64_t dirtyUniforms) {
 	if (dirtyUniforms & DIRTY_MATEMISSIVE) {
 		Uint8x3ToFloat4(ub->materialEmissive, gstate.materialemissive);
 	}
-
 	for (int i = 0; i < 4; i++) {
 		if (dirtyUniforms & (DIRTY_LIGHT0 << i)) {
 			if (gstate.isDirectionalLight(i)) {
diff --git a/GPU/Math3D.h b/GPU/Math3D.h
index 53c7a1e8a0..8d24f0b437 100644
--- a/GPU/Math3D.h
+++ b/GPU/Math3D.h
@@ -822,8 +822,7 @@ typedef Math3D::Vec3Packed<float> Vec3Packedf;
 typedef Math3D::Vec4<float> Vec4f;
 
 
-inline void Vec3ByMatrix43(float vecOut[3], const float v[3], const float m[12])
-{
+inline void Vec3ByMatrix43(float vecOut[3], const float v[3], const float m[12]) {
 	vecOut[0] = v[0] * m[0] + v[1] * m[3] + v[2] * m[6] + m[9];
 	vecOut[1] = v[0] * m[1] + v[1] * m[4] + v[2] * m[7] + m[10];
 	vecOut[2] = v[0] * m[2] + v[1] * m[5] + v[2] * m[8] + m[11];
@@ -895,6 +894,14 @@ inline void ConvertMatrix4x3To4x4Transposed(float *m4x4, const float *m4x3) {
 	m4x4[15] = 1.0f;
 }
 
+// 0369
+// 147A
+// 258B
+// ->>-
+// 0123
+// 4567
+// 89AB
+// Don't see a way to SIMD that. Should be pretty fast anyway.
 inline void ConvertMatrix4x3To3x4Transposed(float *m4x4, const float *m4x3) {
 	m4x4[0] = m4x3[0];
 	m4x4[1] = m4x3[3];
diff --git a/GPU/Vulkan/DrawEngineVulkan.cpp b/GPU/Vulkan/DrawEngineVulkan.cpp
index ce76ae0057..d973b4228e 100644
--- a/GPU/Vulkan/DrawEngineVulkan.cpp
+++ b/GPU/Vulkan/DrawEngineVulkan.cpp
@@ -670,8 +670,8 @@ void DrawEngineVulkan::DoFlush() {
 	VulkanVertexShader *vshader = nullptr;
 	VulkanFragmentShader *fshader = nullptr;
 
-	uint32_t ibOffset = 0;
-	uint32_t vbOffset = 0;
+	uint32_t ibOffset;
+	uint32_t vbOffset;
 	
 	if (useHWTransform) {
 		// We don't detect clears in this path, so here we can switch framebuffers if necessary.
@@ -880,7 +880,7 @@ void DrawEngineVulkan::DoFlush() {
 				return;
 			}
 			if (pipeline != lastPipeline_) {
-				vkCmdBindPipeline(cmd, VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline->pipeline);  // TODO: Avoid if same as last draw.
+				vkCmdBindPipeline(cmd, VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline->pipeline);
 				lastPipeline_ = pipeline;
 			}
 			ApplyDrawStateLate(cmd, false, 0);
diff --git a/GPU/Vulkan/GPU_Vulkan.cpp b/GPU/Vulkan/GPU_Vulkan.cpp
index e79b39e87b..bed17ed1d6 100644
--- a/GPU/Vulkan/GPU_Vulkan.cpp
+++ b/GPU/Vulkan/GPU_Vulkan.cpp
@@ -482,8 +482,6 @@ void GPU_Vulkan::Execute_Prim(u32 op, u32 diff) {
 
 	// This also makes skipping drawing very effective.
 	framebufferManager_->SetRenderFrameBuffer(gstate_c.IsDirty(DIRTY_FRAMEBUF), gstate_c.skipDrawReason);
-	if (!draw_->GetNativeObject(Draw::NativeObject::CURRENT_RENDERPASS))
-		Crash();
 
 	if (gstate_c.skipDrawReason & (SKIPDRAW_SKIPFRAME | SKIPDRAW_NON_DISPLAYED_FB)) {
 		drawEngine_.SetupVertexDecoder(gstate.vertType);  // Do we still need to do this?
diff --git a/ext/native/Android.mk b/ext/native/Android.mk
index 089f49d192..4ff5c8005d 100644
--- a/ext/native/Android.mk
+++ b/ext/native/Android.mk
@@ -51,6 +51,7 @@ LOCAL_SRC_FILES :=\
     input/input_state.cpp \
     math/fast/fast_math.c \
     math/fast/fast_matrix.c \
+    math/dataconv.cpp \
     math/math_util.cpp \
     math/curves.cpp \
     math/expression_parser.cpp \
diff --git a/ext/native/math/dataconv.cpp b/ext/native/math/dataconv.cpp
new file mode 100644
index 0000000000..2fcc339c90
--- /dev/null
+++ b/ext/native/math/dataconv.cpp
@@ -0,0 +1,3 @@
+#include "dataconv.h"
+
+alignas(16) const float one_over_255_x4[4] = { 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f, };
diff --git a/ext/native/math/dataconv.h b/ext/native/math/dataconv.h
index 3940293e59..8c93fffed0 100644
--- a/ext/native/math/dataconv.h
+++ b/ext/native/math/dataconv.h
@@ -3,26 +3,65 @@
 #include <cstdint>
 #include <cstring>
 
+#include "Common/Common.h"
+#include "ppsspp_config.h"
+
 #ifdef _M_SSE
-#include <emmintrin.h>>
+#include <emmintrin.h>
+#endif
+#if PPSSPP_PLATFORM(ARM_NEON)
+#include <arm_neon.h>
 #endif
 
+extern const float one_over_255_x4[4];
+
 // Utilities useful for filling in std140-layout uniform buffers, and similar.
+// NEON intrinsics: http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0491f/BABDCGGF.html
 
 // LSBs in f[0], etc.
 // Could be SSE optimized.
 inline void Uint8x4ToFloat4(float f[4], uint32_t u) {
+#ifdef _M_SSE
+	__m128i zero = _mm_setzero_si128();
+	__m128i value = _mm_set1_epi32(u);
+	__m128i value32 = _mm_unpacklo_epi16(_mm_unpacklo_epi8(value, zero), zero);
+	__m128 fvalues = _mm_mul_ps(_mm_cvtepi32_ps(value32), _mm_load_ps(one_over_255_x4));
+	_mm_storeu_ps(f, fvalues);
+#elif PPSSPP_PLATFORM(ARM_NEON)
+	const float32x4_t one_over = vdupq_n_f32(1.0f/255.0f);
+	const uint8x8_t value = vld1_lane_u32(u);
+	const uint16x8_t value16 = vmovl_s8(value);
+	const uint32x4_t value32 = vmovl_s16(vget_low_s16(value16));
+	const float32x4_t valueFloat = vmulq_f32(vcvtq_f32_u32(value32), one_over);
+	vst1q_u32((uint32_t *)dest, valueFloat);
+#else
 	f[0] = ((u >> 0) & 0xFF) * (1.0f / 255.0f);
 	f[1] = ((u >> 8) & 0xFF) * (1.0f / 255.0f);
 	f[2] = ((u >> 16) & 0xFF) * (1.0f / 255.0f);
 	f[3] = ((u >> 24) & 0xFF) * (1.0f / 255.0f);
+#endif
 }
 
-inline void Uint8x3ToFloat4(float f[4], uint32_t u) {
+inline void Uint8x3ToFloat4_AlphaUint8(float f[4], uint32_t u, uint8_t alpha) {
+#if defined(_M_SSE) || PPSSPP_PLATFORM(ARM_NEON)
+	Uint8x4ToFloat4(f, (u & 0xFFFFFF) | (alpha << 24));
+#else
 	f[0] = ((u >> 0) & 0xFF) * (1.0f / 255.0f);
 	f[1] = ((u >> 8) & 0xFF) * (1.0f / 255.0f);
 	f[2] = ((u >> 16) & 0xFF) * (1.0f / 255.0f);
-	f[3] = 0.0f;
+	f[3] = alpha * (1.0f / 255.0f);
+#endif
+}
+
+inline void Uint8x3ToFloat4(float f[4], uint32_t u) {
+#if defined(_M_SSE) || PPSSPP_PLATFORM(ARM_NEON)
+	Uint8x4ToFloat4(f, u & 0xFFFFFF);
+#else
+	f[0] = ((u >> 0) & 0xFF) * (1.0f / 255.0f);
+	f[1] = ((u >> 8) & 0xFF) * (1.0f / 255.0f);
+	f[2] = ((u >> 16) & 0xFF) * (1.0f / 255.0f);
+	f[3] = ((u >> 24) & 0xFF) * (1.0f / 255.0f);
+#endif
 }
 
 inline void Uint8x3ToInt4(int i[4], uint32_t u) {
@@ -46,13 +85,6 @@ inline void Uint8x3ToFloat4_Alpha(float f[4], uint32_t u, float alpha) {
 	f[3] = alpha;
 }
 
-inline void Uint8x3ToFloat4_AlphaUint8(float f[4], uint32_t u, uint8_t alpha) {
-	f[0] = ((u >> 0) & 0xFF) * (1.0f / 255.0f);
-	f[1] = ((u >> 8) & 0xFF) * (1.0f / 255.0f);
-	f[2] = ((u >> 16) & 0xFF) * (1.0f / 255.0f);
-	f[3] = alpha * (1.0f / 255.0f);
-}
-
 inline void Uint8x1ToFloat4(float f[4], uint32_t u) {
 	f[0] = ((u >> 0) & 0xFF) * (1.0f / 255.0f);
 	f[1] = 0.0f;
@@ -63,40 +95,54 @@ inline void Uint8x1ToFloat4(float f[4], uint32_t u) {
 // These are just for readability.
 
 inline void CopyFloat2(float dest[2], const float src[2]) {
-	memcpy(dest, src, sizeof(float) * 2);
+	dest[0] = src[0];
+	dest[1] = src[1];
 }
 
 inline void CopyFloat3(float dest[3], const float src[3]) {
-	memcpy(dest, src, sizeof(float) * 3);
+	dest[0] = src[0];
+	dest[1] = src[1];
+	dest[2] = src[2];
 }
 
 inline void CopyFloat1To4(float dest[4], const float src) {
+#ifdef _M_SSE
+	_mm_storeu_ps(dest, _mm_set_ss(src));
+#else
 	dest[0] = src;
 	dest[1] = 0.0f;
 	dest[2] = 0.0f;
 	dest[3] = 0.0f;
+#endif
 }
 
 inline void CopyFloat2To4(float dest[4], const float src[2]) {
-	memcpy(dest, src, sizeof(float) * 2);
+	dest[0] = src[0];
+	dest[1] = src[1];
 	dest[2] = 0.0f;
 	dest[3] = 0.0f;
 }
 
 inline void CopyFloat3To4(float dest[4], const float src[3]) {
-	memcpy(dest, src, sizeof(float) * 3);
+	dest[0] = src[0];
+	dest[1] = src[1];
+	dest[2] = src[2];
 	dest[3] = 0.0f;
 }
 
-inline void CopyFloat4(float dest[4], const float src[4]) {
-	memcpy(dest, src, sizeof(float) * 4);
-}
-
 inline void CopyMatrix4x4(float dest[16], const float src[16]) {
 	memcpy(dest, src, sizeof(float) * 16);
 }
 
 inline void ExpandFloat24x3ToFloat4(float dest[4], uint32_t src[3]) {
+#ifdef _M_SSE
+	__m128i values = _mm_slli_epi32(_mm_load_si128((const __m128i *)src), 8);
+	_mm_storeu_si128((__m128i *)dest, values);
+#elif PPSSPP_PLATFORM(ARM_NEON)
+	const uint32x4_t values = vshlq_n_u32(vld1q_u32(&gstate.texscaleu), 8);
+	vst1q_u32((uint32_t *)dest, values);
+#else
 	uint32_t temp[4] = { src[0] << 8, src[1] << 8, src[2] << 8, 0 };
 	memcpy(dest, temp, sizeof(float) * 4);
+#endif
 }
diff --git a/ext/native/native.vcxproj b/ext/native/native.vcxproj
index 34f5e2aeb8..cbee2ad055 100644
--- a/ext/native/native.vcxproj
+++ b/ext/native/native.vcxproj
@@ -696,6 +696,7 @@
     <ClCompile Include="gfx_es2\draw_text_android.cpp" />
     <ClCompile Include="gfx_es2\draw_text_qt.cpp" />
     <ClCompile Include="gfx_es2\draw_text_win.cpp" />
+    <ClCompile Include="math\dataconv.cpp" />
     <ClCompile Include="thin3d\d3d11_loader.cpp" />
     <ClCompile Include="thin3d\thin3d_d3d11.cpp" />
     <ClCompile Include="util\text\wrap_text.cpp" />
diff --git a/ext/native/native.vcxproj.filters b/ext/native/native.vcxproj.filters
index ee8f08f807..417a918213 100644
--- a/ext/native/native.vcxproj.filters
+++ b/ext/native/native.vcxproj.filters
@@ -784,6 +784,9 @@
     <ClCompile Include="gfx_es2\draw_text_android.cpp">
       <Filter>gfx</Filter>
     </ClCompile>
+    <ClCompile Include="math\dataconv.cpp">
+      <Filter>math</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <Filter Include="gfx">