- wine-gitlab - winehq.org

Re: [PATCH v4 0/1] MR203: vkd3d-shader/hlsl: Fold constant integral abs().
by Ethan Lee (＠flibitijibibo) 18 May '23

18 May '23

On Thu May 18 20:30:01 2023 +0000, Giovanni Mascellani wrote: > I think that what we want is: > ``` > dst->value.u[k].i = src->value.u[k].i == INT_MIN ? INT_MIN : abs(src->value.u[k].i); > ``` If I'm understanding correctly, the folding pass makes a temporary node that the fold functions write into - the fold function can either successfully inline an op into a constant that can replace the op node, or it can't and the temporary node is discarded, leaving the original node list alone. It sounds like the intended change to the current revision is to always assign the full constant value even if it turns out to be invalid - that is, the caller can either listen to the return value and discard the dst node, or use it at their own risk... does that make sense? -- https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/203#note_33144

1 0

Re: [PATCH v4 0/1] MR203: vkd3d-shader/hlsl: Fold constant integral abs().
by Giovanni Mascellani (＠giomasce) 18 May '23

18 May '23

On Thu May 18 20:28:51 2023 +0000, Zebediah Figura wrote: > I don't understand, why would it break the result? I think that what we want is: ``` dst->value.u[k].i = src->value.u[k].i == INT_MIN ? INT_MIN : abs(src->value.u[k].i); ``` -- https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/203#note_33143

1 0

Re: [PATCH v4 0/1] MR203: vkd3d-shader/hlsl: Fold constant integral abs().
by Zebediah Figura (＠zfigura) 18 May '23

18 May '23

On Thu May 18 20:23:30 2023 +0000, Ethan Lee wrote: > This change seems like it would break the result - do we want it to be > more like > ``` > bool success = true; > ... > case HLSL_TYPE_INT: > /* Always evaluate this expression... */ > dst->value.u[k].i = abs(src->value.u[k].i); > /* ... but if we try abs(INT_MIN), mark this fold as > invalid */ > if (src->value.u[k].i == INT_MIN) > { > success = false; > } > break; > ... > return success; > ``` I don't understand, why would it break the result? -- https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/203#note_33142

1 0

Re: [PATCH v4 0/1] MR203: vkd3d-shader/hlsl: Fold constant integral abs().
by Ethan Lee (＠flibitijibibo) 18 May '23

18 May '23

On Thu May 18 20:22:26 2023 +0000, Francisco Casas wrote: > Indeed, I didn't have the big picture in mind. returning false keeps the > original instruction. > However, I also think that we should go for the change in case we ever > call hlsl_fold_constant_exprs() from somewhere else. This change seems like it would break the result - do we want it to be more like ``` bool success = true; ... case HLSL_TYPE_INT: /* Always evaluate this expression... */ dst->value.u[k].i = abs(src->value.u[k].i); /* ... but if we try abs(INT_MIN), mark this fold as invalid */ if (src->value.u[k].i == INT_MIN) { success = false; } break; ... return success; ``` -- https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/203#note_33141

1 0

Re: [PATCH v4 0/1] MR203: vkd3d-shader/hlsl: Fold constant integral abs().
by Francisco Casas (＠fcasas) 18 May '23

18 May '23

On Thu May 18 20:19:08 2023 +0000, Zebediah Figura wrote: > > I might have misread, but doesn't the return value apply to the full > vector and not its components? I thought returning false made it so the > whole abs op was left alone if any component was INT_MIN, like an > all-or-nothing deal. > That is the case, yes. That said, I believe we want to make that change anyway. Indeed, I didn't have the big picture in mind. returning false keeps the original instruction. However, I also think that we should go for the change in case we ever call hlsl_fold_constant_exprs() from somewhere else. -- https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/203#note_33140

1 0

Re: [PATCH v4 0/1] MR203: vkd3d-shader/hlsl: Fold constant integral abs().
by Zebediah Figura (＠zfigura) 18 May '23

18 May '23

> I might have misread, but doesn't the return value apply to the full vector and not its components? I thought returning false made it so the whole abs op was left alone if any component was INT_MIN, like an all-or-nothing deal. That is the case, yes. That said, I believe we want to make that change anyway. -- https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/203#note_33139

1 0

[PATCH 0/1] MR2852: winemac: Add missing event to dbgstr_event().
by Brendan Shanks (＠bshanks) 18 May '23

18 May '23

-- https://gitlab.winehq.org/wine/wine/-/merge_requests/2852

3 2

Re: [PATCH v4 0/1] MR203: vkd3d-shader/hlsl: Fold constant integral abs().
by Ethan Lee (＠flibitijibibo) 18 May '23

18 May '23

On Thu May 18 19:54:01 2023 +0000, Francisco Casas wrote: > This won't get the absolute value of components that come after the > component that is INT_MIN. > Consider an int vector such as `(-1, INT_MIN, -1, -1)`; This would end > up as `(1, INT_MIN, -1, -1)`. > I think it would be better to guard the assignment: > ```c > case HLSL_TYPE_INT: > /* abs(INT_MIN) is undefined, leave this expression > alone */ > if (src->value.u[k].i != INT_MIN) > dst->value.u[k].i = abs(src->value.u[k].i); > break; > ``` I might have misread, but doesn't the return value apply to the full vector and not its components? I thought returning false made it so the whole abs op was left alone if any component was INT_MIN, like an all-or-nothing deal. -- https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/203#note_33136

1 0

[PATCH v5 0/1] MR2346: winsqlite3: Add stub DLL.
by Tareque Md Hanif (＠tmdh) 18 May '23

18 May '23

Required to run winget package manager. -- v5: winsqlite3: Add stub DLL. https://gitlab.winehq.org/wine/wine/-/merge_requests/2346

3 4

Re: [PATCH v4 0/1] MR203: vkd3d-shader/hlsl: Fold constant integral abs().
by Francisco Casas (＠fcasas) 18 May '23

18 May '23

Francisco Casas (@fcasas) commented about libs/vkd3d-shader/hlsl_constant_ops.c: > + { > + switch (type) > + { > + case HLSL_TYPE_FLOAT: > + case HLSL_TYPE_HALF: > + dst->value.u[k].f = fabsf(src->value.u[k].f); > + break; > + > + case HLSL_TYPE_DOUBLE: > + dst->value.u[k].d = fabs(src->value.u[k].d); > + break; > + > + case HLSL_TYPE_INT: > + /* abs(INT_MIN) is undefined, leave this expression alone */ > + if (src->value.u[k].i == INT_MIN) > + return false; This won't get the absolute value of components that come after the component that is INT_MIN. Consider an int vector such as `(-1, INT_MIN, -1, -1)`; This would end up as `(1, INT_MIN, -1, -1)`. I think it would be better to guard the assignment: ```c case HLSL_TYPE_INT: /* abs(INT_MIN) is undefined, leave this expression alone */ if (src->value.u[k].i != INT_MIN) dst->value.u[k].i = abs(src->value.u[k].i); break; ``` -- https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/203#note_33133

1 0

2025

2024

2023

2022

wine-gitlab