softgpu: Avoid flush on END.

We only, but always, flush when exiting list interp in FinishDeferred.
It's not necessary at END, and hurts for simple signals that don't stop
list processing.

GPU/GPUCommon.cpp

@@ -1348,7 +1348,8 @@ void GPUCommon::Execute_Ret(u32 op, u32 diff) {
 }
 
 void GPUCommon::Execute_End(u32 op, u32 diff) {
-	Flush();
+	if (flushOnParams_)
+		Flush();
 
 	const u32 prev = Memory::ReadUnchecked_U32(currentList->pc - 4);
 	UpdatePC(currentList->pc, currentList->pc);
@@ -2408,7 +2409,8 @@ void GPUCommon::FastLoadBoneMatrix(u32 target) {
 	}
 
 	if (!g_Config.bSoftwareSkinning) {
-		Flush();
+		if (flushOnParams_)
+			Flush();
 		gstate_c.Dirty(uniformsToDirty);
 	} else {
 		gstate_c.deferredVertTypeDirty |= uniformsToDirty;

GPU/GPUCommon.h

@@ -318,6 +318,7 @@ class GPUCommon : public GPUInterface, public GPUDebugInterface {
 	TextureCacheCommon *textureCache_ = nullptr;
 	DrawEngineCommon *drawEngineCommon_ = nullptr;
 	ShaderManagerCommon *shaderManager_ = nullptr;
+	bool flushOnParams_ = true;
 
 	GraphicsContext *gfxCtx_;
 	Draw::DrawContext *draw_;

GPU/Software/SoftGpu.cpp

@@ -438,6 +438,9 @@ SoftGPU::SoftGPU(GraphicsContext *gfxCtx, Draw::DrawContext *draw)
 	// Push the initial CLUT buffer in case it's all zero (we push only on change.)
 	drawEngine_->transformUnit.NotifyClutUpdate(clut);
 
+	// No need to flush for simple parameter changes.
+	flushOnParams_ = false;
+
 	if (gfxCtx && draw) {
 		presentation_ = new PresentationCommon(draw_);
 		presentation_->SetLanguage(draw_->GetShaderLanguageDesc().shaderLanguage);