Comment improvements, rename a dirty-flag

Common/CPUDetect.cpp

@@ -287,8 +287,7 @@ void CPUInfo::Detect() {
 		// -- Is the RTM bit set in CPUID? (>>11)
 		// -- No need to check HLE bit because legacy processors ignore HLE hints
 		// -- See https://software.intel.com/en-us/articles/how-to-detect-new-instruction-support-in-the-4th-generation-intel-core-processor-family
-		if (max_std_fn >= 7)
-		{
+		if (max_std_fn >= 7) {
 			do_cpuid(cpu_id, 0x00000007);
 			// careful; we can't enable AVX2 unless the XSAVE/XGETBV checks above passed
 			if ((cpu_id[1] >> 5) & 1)

GPU/Common/FragmentShaderGenerator.cpp

@@ -1094,17 +1094,19 @@ bool GenerateFragmentShader(const FShaderID &id, char *buffer, const ShaderLangu
 		case STENCIL_VALUE_INCR_4BIT:
 		case STENCIL_VALUE_DECR_4BIT:
 			// We're adding/subtracting, just by the smallest value in 4-bit.
+			// We have to set the blend mode to match elsewhere.
 			snprintf(replacedAlpha, sizeof(replacedAlpha), "%f", 1.0 / 15.0);
 			break;
 
 		case STENCIL_VALUE_INCR_8BIT:
 		case STENCIL_VALUE_DECR_8BIT:
 			// We're adding/subtracting, just by the smallest value in 8-bit.
+			// We have to set the blend mode to match elsewhere.
 			snprintf(replacedAlpha, sizeof(replacedAlpha), "%f", 1.0 / 255.0);
 			break;
 
 		case STENCIL_VALUE_KEEP:
-			// Do nothing. We'll mask out the alpha using color mask.
+			// Do nothing. We'll mask out the alpha using color mask (could also be done with a blend mode).
 			break;
 		}
 	}

GPU/Common/FramebufferManagerCommon.cpp

@@ -429,6 +429,9 @@ VirtualFramebuffer *FramebufferManagerCommon::DoSetRenderFrameBuffer(Framebuffer
 			} else if (PSP_CoreParameter().compat.flags().FramebufferAllowLargeVerticalOffset &&
 				params.fb_address > v->fb_address && v->fb_stride > 0 && (params.fb_address - v->fb_address) % v->FbStrideInBytes() == 0 &&
 				params.fb_address != 0x04088000 && v->fb_address != 0x04000000) {  // Heuristic to avoid merging the main framebuffers.
+
+				// Breath of Fire III relies on this to render the second framebuffer inside the first one.
+
 				y_offset = (params.fb_address - v->fb_address) / v->FbStrideInBytes();
 				if (y_offset <= v->bufferHeight) {  // note: v->height is misdetected as 256 instead of 272 here in tokimeki. Note that 272 is just the height of the upper part, it's supersampling vertically.
 					large_offset_vfb = v;
@@ -1037,11 +1040,11 @@ void FramebufferManagerCommon::NotifyRenderFramebufferCreated(VirtualFramebuffer
 
 void FramebufferManagerCommon::NotifyRenderFramebufferUpdated(VirtualFramebuffer *vfb) {
 	if (gstate_c.curRTWidth != vfb->width || gstate_c.curRTHeight != vfb->height) {
-		gstate_c.Dirty(DIRTY_PROJTHROUGHMATRIX | DIRTY_VIEWPORTSCISSOR_STATE);
+		gstate_c.Dirty(DIRTY_FRAMEBUFFER_DIM | DIRTY_VIEWPORTSCISSOR_STATE);
 	}
 	if (gstate_c.curRTRenderWidth != vfb->renderWidth || gstate_c.curRTRenderHeight != vfb->renderHeight) {
 		gstate_c.Dirty(DIRTY_PROJMATRIX);
-		gstate_c.Dirty(DIRTY_PROJTHROUGHMATRIX);
+		gstate_c.Dirty(DIRTY_FRAMEBUFFER_DIM);
 	}
 }
 

GPU/Common/ShaderCommon.h

@@ -44,7 +44,7 @@ enum DebugShaderStringType {
 // more code than before. TODO: Can probably cut the number of these down without too much slowdown.
 enum : uint64_t {
 	DIRTY_PROJMATRIX = 1ULL << 0,
-	DIRTY_PROJTHROUGHMATRIX = 1ULL << 1,
+	DIRTY_FRAMEBUFFER_DIM = 1ULL << 1,
 	DIRTY_FOGCOLOR = 1ULL << 2,
 	DIRTY_FOGCOEF = 1ULL << 3,
 	DIRTY_TEXENV = 1ULL << 4,

GPU/Common/ShaderUniforms.cpp

@@ -76,7 +76,7 @@ void BaseUpdateUniforms(UB_VS_FS_Base *ub, uint64_t dirtyUniforms, bool useBuffe
 		ub->rotation = useBufferedRendering ? 0 : (float)g_display.rotation;
 	}
 
-	if (dirtyUniforms & DIRTY_PROJTHROUGHMATRIX) {
+	if (dirtyUniforms & DIRTY_FRAMEBUFFER_DIM) {
 		ub->xywh[0] = (float)gstate_c.curRTOffsetX;
 		ub->xywh[1] = (float)gstate_c.curRTOffsetY;
 		ub->xywh[2] = (float)gstate_c.curRTWidth;
@@ -96,7 +96,6 @@ void BaseUpdateUniforms(UB_VS_FS_Base *ub, uint64_t dirtyUniforms, bool useBuffe
 	}
 
 	if (dirtyUniforms & DIRTY_VIEWPORT_UNIFORMS) {
-		// TODO: This should be a couple of SIMD instructions.
 		Vec4F32 vpScale = Vec4F32::LoadF24x4(&gstate.viewportxscale);
 		Vec4F32 vpOffset = Vec4F32::LoadF24x4(&gstate.viewportxcenter);
 		vpScale.Store(ub->vpScale);

GPU/Common/ShaderUniforms.h

@@ -10,7 +10,7 @@
 
 enum : uint64_t {
 	DIRTY_BASE_UNIFORMS =
-	DIRTY_WORLDMATRIX | DIRTY_PROJTHROUGHMATRIX | DIRTY_VIEWMATRIX | DIRTY_TEXMATRIX | DIRTY_ALPHACOLORREF |
+	DIRTY_WORLDMATRIX | DIRTY_FRAMEBUFFER_DIM | DIRTY_VIEWMATRIX | DIRTY_TEXMATRIX | DIRTY_ALPHACOLORREF |
 	DIRTY_PROJMATRIX | DIRTY_FOGCOLOR | DIRTY_FOGCOEF | DIRTY_TEXENV | DIRTY_TEX_ALPHA_MUL | DIRTY_STENCILREPLACEVALUE |
 	DIRTY_ALPHACOLORMASK | DIRTY_SHADERBLEND | DIRTY_COLORWRITEMASK | DIRTY_UVSCALEOFFSET | DIRTY_TEXCLAMP | DIRTY_MATAMBIENTALPHA |
 	DIRTY_DEPAL | DIRTY_VIEWPORT_UNIFORMS | DIRTY_RASTER_OFFSET,

GPU/Common/VertexShaderGenerator.cpp

@@ -407,7 +407,7 @@ bool GenerateVertexShader(const VShaderID &id, char *buffer, const ShaderLanguag
 
 		WRITE(p, "uniform vec4 u_xywh;\n");
 		WRITE(p, "uniform float u_NaN;\n");
-		*uniformMask |= DIRTY_PROJTHROUGHMATRIX;
+		*uniformMask |= DIRTY_FRAMEBUFFER_DIM;
 
 		WRITE(p, "uniform vec2 u_minZmaxZ;\n");
 		*uniformMask |= DIRTY_RASTER_OFFSET;  // this flag is shared with raster offset.

GPU/GLES/ShaderManagerGLES.cpp

@@ -406,7 +406,7 @@ void LinkedShader::UpdateUniforms(const ShaderID &vsid, const ShaderLanguageDesc
 		memcpy(&matrix, gstate.projMatrix, 16 * sizeof(float));
 		render_->SetUniformM4x4(&u_proj, matrix.m);
 	}
-	if (dirty & DIRTY_PROJTHROUGHMATRIX) {
+	if (dirty & DIRTY_FRAMEBUFFER_DIM) {
 		float xywh[4];
 		xywh[0] = (float)gstate_c.curRTOffsetX;
 		xywh[1] = (float)gstate_c.curRTOffsetY;

GPU/GPUState.h

@@ -654,7 +654,7 @@ public:
 		if (xoff != curRTOffsetX || yoff != curRTOffsetY) {
 			curRTOffsetX = xoff;
 			curRTOffsetY = yoff;
-			Dirty(DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_PROJTHROUGHMATRIX);
+			Dirty(DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_FRAMEBUFFER_DIM);
 		}
 	}
 	int curRTOffsetX;

GPU/Math3D.h

@@ -1176,6 +1176,7 @@ inline void ConvertMatrix4x3To4x4Transposed(float *m4x4, const float *m4x3) {
 // 4567
 // 89AB
 // Don't see a way to SIMD that. Should be pretty fast anyway.
+// And on NEON it just works!
 inline void ConvertMatrix4x3To3x4Transposed(float *m3x4, const float *m4x3) {
 #if PPSSPP_ARCH(ARM_NEON)
 	// vld3q is a perfect match here!