headless: Add some new and passing tests #16046
Conversation
GPU/Common/VertexDecoderArm.cpp
Outdated
| LDRB(tempReg3, srcReg, dec_->posoff + 2); | ||
| static const ARMReg tr[3] = { tempReg1, tempReg2, tempReg3 }; | ||
| static const ARMReg fr[3] = { fpScratchReg, fpScratchReg2, fpScratchReg3 }; | ||
| // Intentionally just sets all to zero. |
There was a problem hiding this comment.
I think a better comment here is what you have below:
// 8-bit positions in throughmode always decode to 0, depth included.
GPU/Common/VertexDecoderArm64.cpp
Outdated
| fp.SCVTF(fpScratchReg, tempReg1); | ||
| fp.SCVTF(fpScratchReg2, tempReg2); | ||
| fp.SCVTF(fpScratchReg3, tempReg3); | ||
| // Intentionally just sets all to zero. |
| @@ -872,22 +872,14 @@ void VertexDecoderJitCache::Jit_NormalFloat() { | |||
| STMIA(scratchReg, false, 3, tempReg1, tempReg2, tempReg3); | |||
| } | |||
|
|
|||
| // Through expands into floats, always. Might want to look at changing this. | |||
| void VertexDecoderJitCache::Jit_PosS8Through() { | |||
There was a problem hiding this comment.
Should this be renamed PosInvalid too?
There was a problem hiding this comment.
Well, "invalid" I used for when no position is specified at all, and the verts are all culled.
However, when a position is specified, as s8 in through, it is not culled. It decodes to zero. So it's kinda invalid, but different from "invalid position in throughmode (no bits set)" which has different behavior.
That is to say:
(vertType & GE_VTYPE_POS_MASK) == 0-> invalid, no drawing occurs.(vertType & GE_VTYPE_POS_MASK) == GE_VTYPE_POS_8BIT && (vertType & GE_VTYPE_THROUGH) != 0->
"valid", but decodes to zero.(vertType & GE_VTYPE_POS_MASK) == GE_VTYPE_POS_8BIT && (vertType & GE_VTYPE_THROUGH) == 0-> valid, and decodes to sensible values.
(this behavior is all demonstrated in the "invalidprim" test.)
-[Unknown]
There was a problem hiding this comment.
As to Arm/Arm64/etc. - I figured might as well keep these jitted, although they're hopefully rarely being hit. I didn't think it was worth bothering to newly jit invalid, though.
-[Unknown]
There was a problem hiding this comment.
Ok, cool. Thanks for the clarifications!
Also always cull no-position verts, hardware too. Matches tests.
unknownbrackets
force-pushed
the
tests-update
branch
from
c71e0a6 to
ca248e1
Compare
|
Feel free to merge if it's ready :) |
Also small fix to an existing one to make it pass as well.
-[Unknown]