I find its text is abnormal.
So please open attachment question.txt.
Thank you.
tips:
normal should be normally in question.txt line 14.
normal should be normally in question.txt line 29.
I don't know - I wrote that demo nearly 14 years ago (and about 10 years ago released the patched version you're running), and I think it was the first test ROM I ever wrote.
My best guess is that it's due to timing differences in the actual operations - turning off background rendering (test #1) takes effect immediately, but changing the pattern table address (test #2) or the nametable address (test #3) takes longer because of pipelining within the PPU. Looking back at my old IRC logs (from October 2000), it looks like I arrived at those numbers through experimentation, namely by having kevtris run it on his CopyNES and tweaking it until it worked.
I figure it out.
Column w and column x aren't hidden in test #2 and #3.
PPU is drawing tiles fetched from column u and column v in test #2 and #3 at that time.
As you say, it is pipelining.
PPU already fetched 2 tiles data from tick 321 to 336 at previous scanline.
There's a screenshot for displaying dynamic effect. I think it can help you to understand this demo clearly.