March 2016 - Wine-devel - list.winehq.org

Re: [PATCH 1/4] xmllite/tests: A test for Write* behaviour on initial writer state
by Marvin 23 Mar '16

23 Mar '16

Hi, While running your changed tests on Windows, I think I found new failures. Being a bot and all I'm not very good at pattern recognition, so I might be wrong, but could you please double-check? Full results can be found at https://testbot.winehq.org/JobDetails.pl?Key=21605 Your paranoid android. === build (build) === Patch failed to apply

1 0

Re: [v2] bcrypt: Added BCryptHash function.
by Hans Leidekker 23 Mar '16

23 Mar '16

On Tue, 2016-03-22 at 07:50 -0700, Patrick Armstrong wrote: > +NTSTATUS WINAPI BCryptHash( BCRYPT_ALG_HANDLE algorithm, UCHAR *secret, ULONG secretlen, > + UCHAR *input, ULONG inputlen, UCHAR *output, ULONG outputlen ) > +{ > + NTSTATUS status; > + BCRYPT_HASH_HANDLE handle; > + > + TRACE( "%p, %p, %u, %p, %u, %p, %u\n", algorithm, secret, secretlen, > + input, inputlen, output, outputlen ); > + > + status = BCryptCreateHash( algorithm, &handle, NULL, 0, secret, secretlen, 0); > + if (status != STATUS_SUCCESS) > + { > + BCryptDestroyHash( handle ); > + return status; > + } BCryptCreateHash doesn't return a handle when it fails. > NTSTATUS WINAPI BCryptCloseAlgorithmProvider(BCRYPT_ALG_HANDLE, ULONG); > NTSTATUS WINAPI BCryptCreateHash(BCRYPT_ALG_HANDLE, BCRYPT_HASH_HANDLE *, PUCHAR, ULONG, PUCHAR, ULONG, ULONG); > NTSTATUS WINAPI BCryptDestroyHash(BCRYPT_HASH_HANDLE); > +NTSTATUS WINAPI BCryptHash(BCRYPT_ALG_HANDLE, PUCHAR, ULONG, PUCHAR, ULONG, PUCHAR, ULONG); > NTSTATUS WINAPI BCryptEnumAlgorithms(ULONG, ULONG *, BCRYPT_ALGORITHM_IDENTIFIER **, ULONG); > NTSTATUS WINAPI BCryptFinishHash(BCRYPT_HASH_HANDLE, PUCHAR, ULONG, ULONG); > NTSTATUS WINAPI BCryptGenRandom(BCRYPT_ALG_HANDLE, PUCHAR, ULONG, ULONG); While you're at it, please keep these sorted.

1 0

Re: [v5 2/8] d3dx9: Introduce preshaders in effect.
by Matteo Bruni 23 Mar '16

23 Mar '16

2016-03-21 14:55 GMT+01:00 Paul Gofman <gofmanp(a)gmail.com>: > Signed-off-by: Paul Gofman <gofmanp(a)gmail.com> > @@ -5435,9 +5416,30 @@ static HRESULT d3dx9_parse_array_selector(struct d3dx9_base_effect *base, struct > } > TRACE("Unknown DWORD: 0x%.8x.\n", *(DWORD *)(ptr + string_size)); > > - FIXME("Parse preshader.\n"); > + d3dx_create_param_eval(base, (DWORD *)(ptr + string_size) + 1, object->size - (string_size + 1), > + D3DXPT_INT, &param->param_eval); Does this do the right thing if ptr + string_size is aligned to a DWORD? In other words, does the bytecode contain 4 bytes of padding if string_size % 4 == 0?

2 2

Re: mshtml: Fixed remaining xmlhttprequest test failures.
by Marvin 23 Mar '16

23 Mar '16

Hi, While running your changed tests on Windows, I think I found new failures. Being a bot and all I'm not very good at pattern recognition, so I might be wrong, but could you please double-check? Full results can be found at https://testbot.winehq.org/JobDetails.pl?Key=21603 Your paranoid android. === build (build) === Patch failed to apply

1 0

Re: [tools 2/2] Support for disabling and enabling warnings, handle warning number 2368
by Nikolay Sivov 23 Mar '16

23 Mar '16

On 20.03.2016 13:12, Bolotov Sergey wrote: > diff --git a/tools/widl/write_msft.c b/tools/widl/write_msft.c > index 8b65ead..137bb2d 100644 > --- a/tools/widl/write_msft.c > +++ b/tools/widl/write_msft.c > @@ -51,6 +51,7 @@ > #include "header.h" > #include "hash.h" > #include "typetree.h" > +#include "parser.h" > > enum MSFT_segment_index { > MSFT_SEG_TYPEINFO = 0, /* type information */ > @@ -504,7 +505,7 @@ static int ctl2_alloc_guid( > offset = ctl2_find_guid(typelib, hash_key, &guid->guid); > if (offset != -1) > { > - if (pedantic) > + if (is_warning_enabled(2368)) > warning("duplicate uuid {%08x-%04x-%04x-%02x%02x-%02x%02x%02x%02x%02x%02x}\n", > guid->guid.Data1, guid->guid.Data2, guid->guid.Data3, > guid->guid.Data4[0], guid->guid.Data4[1], guid->guid.Data4[2], guid->guid.Data4[3], > Duplicate UUID is not a warning 2368, it's an error 2270, I don't think you can suppress it, so we should really abort compilation at this point. Also if we're doing this, it looks cleaner to have is_warning..() check within warning(), so you can check there if it's enabled or not, and it's possible 'pedantic' should take precedence. Could you share a minimal idl you're using to test that?

2 3

Re: [v5 8/8] d3dx9: Implement setting named shader constants in effect.
by Matteo Bruni 22 Mar '16

22 Mar '16

2016-03-21 14:55 GMT+01:00 Paul Gofman <gofmanp(a)gmail.com>: > +HRESULT d3dx_set_shader_const_fxlc(struct ID3DXEffectImpl *effect, struct d3dx_pass *pass, > + struct d3dx_parameter *param, unsigned int vs) Using a BOOL for the vs flag seems nicer. > diff --git a/dlls/d3dx9_36/preshader.c b/dlls/d3dx9_36/preshader.c > index fec9de4..1770046 100644 > --- a/dlls/d3dx9_36/preshader.c > +++ b/dlls/d3dx9_36/preshader.c > @@ -242,7 +242,7 @@ static void regstore_set_values(struct d3dx_regstore *rs, unsigned int table, vo > static unsigned int regstore_is_val_set_reg(struct d3dx_regstore *rs, unsigned int table, unsigned int reg_idx) > { > return rs->table_value_set[table][reg_idx / PRES_BITMASK_BLOCK_SIZE] & > - (1ul << (reg_idx % PRES_BITMASK_BLOCK_SIZE)); > + (1u << (reg_idx % PRES_BITMASK_BLOCK_SIZE)); When you end up changing stuff introduced in a previous patch in the series you know there is something wrong ;) I guess it's a rebase glitch (by any chance, was this a remnant of a try with the uint_ptr idea you mentioned earlier?) > +HRESULT d3dx_param_eval_set_shader_constants(struct IDirect3DDevice9 *device, struct d3dx_param_eval *peval) > +{ > + HRESULT hr, res; > + struct d3dx_preshader *pres = &peval->pres; > + struct d3dx_regstore *rs = &pres->regs; > + > + TRACE("device %p, peval %p, param_type %u.\n", device, peval, peval->param_type); > + > + if (FAILED(hr = set_constants(rs, &pres->inputs))) > + return hr; > + if (FAILED(hr = execute_preshader(pres))) > + return hr; > + if (FAILED(hr = set_constants(rs, &peval->shader_inputs))) > + return hr; > + > + #define SET_TABLE(table, func_suff, type) \ > + { \ > + unsigned int is, n; \ > + \ > + is = 0; \ > + while (is < rs->table_sizes[table]) \ > + {\ > + n = 0; \ > + while (is < rs->table_sizes[table] && !regstore_is_val_set_reg(rs, table, is)) \ > + is++; \ > + while (is + n < rs->table_sizes[table] && regstore_is_val_set_reg(rs, table, is + n)) \ > + n++; \ > + if (!n) \ > + continue; \ > + TRACE("setting %u consts at %u.\n", n, is); \ > + hr = IDirect3DDevice9_##func_suff(device, is, \ > + (const type *)rs->tables[table] + is * table_info[table].reg_component_count, n); \ > + if (FAILED(hr)) \ > + { \ > + ERR(#func_suff " failed, hr %#x.\n", hr); \ > + res = hr; \ > + } \ > + is += n; \ > + } \ > + regstore_reset_table(rs, table); \ > + } > + res = D3D_OK; > + if (peval->param_type == D3DXPT_VERTEXSHADER) > + { > + SET_TABLE(PRES_REGTAB_OCONST, SetVertexShaderConstantF, float); > + SET_TABLE(PRES_REGTAB_OICONST, SetVertexShaderConstantI, int); > + SET_TABLE(PRES_REGTAB_OBCONST, SetVertexShaderConstantB, BOOL); > + } > + else if (peval->param_type == D3DXPT_PIXELSHADER) > + { > + SET_TABLE(PRES_REGTAB_OCONST, SetPixelShaderConstantF, float); > + SET_TABLE(PRES_REGTAB_OICONST, SetPixelShaderConstantI, int); > + SET_TABLE(PRES_REGTAB_OBCONST, SetPixelShaderConstantB, BOOL); > + } > + else > + { > + FIXME("Unexpected parameter type %u.\n", peval->param_type); > + return D3DERR_INVALIDCALL; > + } > + #undef SET_TABLE This is certainly hard to handle nicely because of the multiple similar d3d9 functions with slightly different signatures but I think it's better to avoid macros (especially such large ones) as much as possible. There are multiple ways to achieve that. Perhaps the easiest one is to turn SET_TABLE into a function taking an array of enum PRES_REG_TABLES (which, btw, should be lowercase) and the shader type as additional parameters, loop over the tables and call the correct d3d9 function by using two levels of switches (or ifs). The type isn't really necessary, you can calculate the offset with table_info[].component_size.

2 2

Re: [v5 1/8] d3dx9: Add test for preshader in effect.
by Matteo Bruni 22 Mar '16

22 Mar '16

2016-03-21 14:55 GMT+01:00 Paul Gofman <gofmanp(a)gmail.com>: > +struct test_struct > +{ > + float3 v1; > + float fv; > + float4 v2; > +}; > + > +test_struct ts1[1] = {{{9, 10, 11}, 12, {13, 14, 15, 16}}}; > +test_struct ts2[2] = {{{0, 0, 0}, 0, {0, 0, 0, 0}}, {{1, 2, 3}, 4, {5, 6, 7, 8}}}; Is it also possible to create a struct with an array field? And an array of those structs? > + static const struct > + { > + const char *comment; > + BOOL todo[4]; > + unsigned int result[4]; > + } > + test_effect_preshader_op_results[] = > + { > + {"1 / op", { TRUE, TRUE, TRUE, TRUE}, {0x7f800000, 0xff800000, 0xbee8ba2e, 0x00200000}}, > +#if __x86_64__ > + {"rsq", { TRUE, TRUE, TRUE, TRUE}, {0x7f800000, 0x7f800000, 0x3f2c985d, 0x1f800000}}, > +#else > + {"rsq", { TRUE, TRUE, TRUE, TRUE}, {0x7f800000, 0x7f800000, 0x3f2c985c, 0x1f800001}}, > +#endif > + {"mul", { TRUE, TRUE, TRUE, TRUE}, {0x00000000, 0x80000000, 0x40d33334, 0x7f800000}}, > + {"add", {FALSE, TRUE, TRUE, TRUE}, {0x3f800000, 0x40000000, 0xc0a66666, 0x7f7fffff}}, > + {"lt", {FALSE, FALSE, TRUE, FALSE}, {0x3f800000, 0x3f800000, 0x00000000, 0x00000000}}, > + {"ge", { TRUE, TRUE, FALSE, TRUE}, {0x00000000, 0x00000000, 0x3f800000, 0x3f800000}}, > + {"neg", { TRUE, TRUE, TRUE, TRUE}, {0x80000000, 0x00000000, 0x400ccccd, 0xff7fffff}}, > + {"rcp", { TRUE, TRUE, TRUE, TRUE}, {0x7f800000, 0xff800000, 0xbee8ba2e, 0x00200000}}, > + {"frac", {FALSE, FALSE, TRUE, FALSE}, {0x00000000, 0x00000000, 0x3f4ccccc, 0x00000000}}, > + {"min", {FALSE, TRUE, TRUE, TRUE}, {0x00000000, 0x80000000, 0xc0400000, 0x40800000}}, > + {"max", { TRUE, TRUE, TRUE, TRUE}, {0x3f800000, 0x40000000, 0xc00ccccd, 0x7f7fffff}}, > +#if __x86_64__ > + {"sin", {FALSE, TRUE, TRUE, TRUE}, {0x00000000, 0x80000000, 0xbf4ef99e, 0xbf0599b3}}, > + {"cos", { TRUE, TRUE, TRUE, TRUE}, {0x3f800000, 0x3f800000, 0xbf16a803, 0x3f5a5f96}}, > +#else > + {"sin", {FALSE, TRUE, TRUE, TRUE}, {0x00000000, 0x80000000, 0xbf4ef99e, 0x3f792dc4}}, > + {"cos", { TRUE, TRUE, TRUE, TRUE}, {0x3f800000, 0x3f800000, 0xbf16a803, 0xbe6acefc}}, > +#endif > + {"den mul",{ TRUE, TRUE, TRUE, TRUE}, {0x7f800000, 0xff800000, 0xbb94f209, 0x000051ec}}, > + {"dot", {FALSE, TRUE, TRUE, FALSE}, {0x00000000, 0x7f800000, 0x41f00000, 0x00000000}}, > +#if __x86_64__ > + {"prec", { TRUE, TRUE, TRUE, FALSE}, {0x2b8cbccc, 0x2c0cbccc, 0xac531800, 0x00000000}} > +#else > + {"prec", { TRUE, TRUE, FALSE, FALSE}, {0x2b8cbccc, 0x2c0cbccc, 0x00000000, 0x00000000}} > +#endif > + }; I don't like those #ifdefs much although apparently I don't have nicer suggestions.

2 2

Re: [v5 5/8] d3dx9: Implement fxlc constants (expressions) in effect.
by Matteo Bruni 22 Mar '16

22 Mar '16

2016-03-21 14:55 GMT+01:00 Paul Gofman <gofmanp(a)gmail.com>: > +static unsigned int regstore_is_val_set_reg(struct d3dx_regstore *rs, unsigned int table, unsigned int reg_idx) > +{ > + return rs->table_value_set[table][reg_idx / PRES_BITMASK_BLOCK_SIZE] & > + (1ul << (reg_idx % PRES_BITMASK_BLOCK_SIZE)); > +} Actually you don't want "ul" here, just "u". > +static void regstore_set_float(struct d3dx_regstore *rs, unsigned int table, unsigned int offset, float v) > +{ > + BYTE *p; > + unsigned int reg_idx; > + > + p = (BYTE *)rs->tables[table] + table_info[table].component_size * offset; > + switch (table_info[table].type) > + { > + case PRES_VT_FLOAT : *(float *)p = v; break; > + case PRES_VT_DOUBLE: *(double *)p = v; break; > + case PRES_VT_INT : *(int *)p = roundf(v); break; > + case PRES_VT_BOOL : *(BOOL *)p = !!v; break; > + } > + reg_idx = OFFSET2REG(table, offset); > + rs->table_value_set[table][reg_idx / PRES_BITMASK_BLOCK_SIZE] |= > + 1ul << (reg_idx % PRES_BITMASK_BLOCK_SIZE); Same here. > +static HRESULT set_constants_param(struct d3dx_regstore *rs, struct d3dx_const_tab *const_tab, > + D3DXHANDLE hc, struct d3dx_parameter *param) > +{ > + ID3DXConstantTable *ctab = const_tab->ctab; > + D3DXCONSTANT_DESC desc; > + unsigned int const_count, param_count, i, j, n, table, start_offset; > + unsigned int nminor, nmajor, major_stride, inp_ind; Nitpick, also rename nmajor/nminor to major_size or just major (resp. minor). inp_ind also feels like it has room for improvement, although I don't have any good suggestion. > +#define ARGS_ARRAY_SIZE 8 > +static HRESULT execute_preshader(struct d3dx_preshader *pres) > +{ > + unsigned int i, j, k; > + float args[ARGS_ARRAY_SIZE]; Doesn't this need to be double to preserve the precision of the immediate operands / literals? I guess that means the regstore should also be able to return doubles.

1 0

Re: [v5 3/8] d3dx9: Implement preshader parsing.
by Matteo Bruni 22 Mar '16

22 Mar '16

2016-03-21 14:55 GMT+01:00 Paul Gofman <gofmanp(a)gmail.com>: > Signed-off-by: Paul Gofman <gofmanp(a)gmail.com> > +enum PRES_VALUE_TYPE Please use lowercase for type names. > +#define OFFSET2REG(table, offset) ((offset) / table_info[table].reg_component_count) I think this could be a function instead, that's usually preferred to macros when possible. Not sure about the name, maybe get_reg_offset(). > +static HRESULT regstore_alloc_table(struct d3dx_regstore *rs, unsigned int table) > +{ > + unsigned int sz; Nitpick, I would just spell out "size", it's pretty short anyway. > +static void regstore_set_values(struct d3dx_regstore *rs, unsigned int table, void *data, > + unsigned int start_offset, unsigned int count) > +{ > + unsigned int reg_idx; > + > + if (!count) > + return; > + > + memcpy((BYTE *)rs->tables[table] + start_offset * table_info[table].component_size, > + data, count * table_info[table].component_size); > + for (reg_idx = OFFSET2REG(table, start_offset); reg_idx <= OFFSET2REG(table, start_offset + count - 1); > + ++reg_idx) > + rs->table_value_set[table][reg_idx / PRES_BITMASK_BLOCK_SIZE] |= > + 1u << (reg_idx % PRES_BITMASK_BLOCK_SIZE); This might be optimized, enabling each bit one by one doesn't seem very efficient. > +} > + > +static void dump_bytecode(void *data, unsigned int size) > +{ > + unsigned int *bytecode = (unsigned int *)data; > + unsigned int i, j, n; > + > + size /= 4; Nitpick, sizeof(*bytecode).

1 0

Re: [v4 2/6] d3dx9: add preshaders to effect.
by Matteo Bruni 22 Mar '16

22 Mar '16

2016-03-17 12:59 GMT+01:00 Paul Gofman <gofmanp(a)gmail.com>: > +enum PRES_REG_TABLES > +{ > + PRES_REGTAB_NONE, > + PRES_REGTAB_IMMED, /* immediate double constants from CLIT */ > + PRES_REGTAB_CONST, > + PRES_REGTAB_VERTEX, /* not used */ > + PRES_REGTAB_OCONST, > + PRES_REGTAB_OBCONST, > + PRES_REGTAB_OICONST, > + PRES_REGTAB_REG, > + PRES_REGTAB_MAX, > + PRES_REGTAB_FIRST = PRES_REGTAB_IMMED, > + PRES_REGTAB_LAST = PRES_REGTAB_REG, > +}; You can drop PRES_REGTAB_NONE unless it's going to be used. > +++ b/dlls/d3dx9_36/preshader.c > @@ -0,0 +1,627 @@ > +/* > + * FX preshader parsing & execution, setting shader constants > + * from effect parameters. > + * It looks like that kind of general file description is omitted in newer source files. Maybe drop it? I don't particularly care though. > + /* TODO: use double precision for x64 arch */ > + {sizeof(float), 4, PRES_VT_FLOAT } /* PRES_REGTAB_REG */ Nitpick, just mention "64 bit" or something, we don't want to make all the non-Intel archs angry ;) > +#define PRES_VS_BITS_PER_WORD (sizeof(unsigned int) * 8) What does the _VS_ part of the name mean? I'm also a bit thrown off by the _WORD part, mostly because in Win32 WORD is a 16 bit while DWORD is 32 bit and unsigned int is also 32 bit. Maybe it's just me, I don't know. > +static void dump_shader_bytecode(void *data, unsigned int size) > +{ > + unsigned int *words = (unsigned int *)data; Same here. I guess you could call it "bytecode" or such. > +static unsigned int *find_bytecode_comment(unsigned int *ptr, unsigned int fourcc) > +{ > + while ((*ptr & 0xFFFF) == 0xFFFE) A couple leftover uppercase hexadecimal constants. > + hr = D3DXGetShaderConstantTable(byte_code, &ctab); > + if (FAILED(hr) || !ctab) > + { > + TRACE("Could not get CTAB data, hr %#x.\n", hr); > + /* returning OK is not a typo */ > + return D3D_OK; > + } Sorry, back to that comment again. I guess it would be better to spell out why that's the intended return value instead (e.g. "Not returning an error, shaders without a CTAB are valid."). > +static void dump_arg(struct d3dx_regstore *rs, struct d3dx_pres_operand *arg, int component_count) > +{ > + static const char *xyzw_str = "xyzw"; > + unsigned int i, table; > + > + table = arg->table; > + if (table == PRES_REGTAB_IMMED) > + { > + TRACE("("); > + for (i = 0; i < component_count; ++i) > + TRACE(i < component_count - 1 ? "%lf, " : "%lf", > + ((double *)rs->tables[PRES_REGTAB_IMMED])[arg->offset + i]); %lf is not right, %f already expects a double. I'd just use %.16e (which has the added benefit of showing something different from 0 in the "precision" test from the previous patch). Another option would be to use the same format used by MS when disassembling shaders, although this seems to be a case where it's worth diverging. > +static void dump_ins(struct d3dx_regstore *rs, struct d3dx_pres_ins *ins) > +{ > + unsigned int i; > + > + TRACE(" %s ", pres_op_info[ins->op].mnem); > + dump_arg(rs, &ins->output, pres_op_info[ins->op].func_all_comps ? 1 : ins->component_count); > + for (i = 0; i < pres_op_info[ins->op].input_count; ++i) > + { > + TRACE(","); With a blank after the comma it would be prettier. There is still some inconsistent formatting (e.g. "== NULL" vs "!", uppercase hexadecimals, cnt vs count). A possible idea to split this patch further: what if you move the addition of (most of) the contents of the various functions in preshader.c to a separate patch? That is, make a patch which adds the various calls to d3dx_create_param_eval(), d3dx_free_param_eval() and similar (i.e. most if not all the changes to effect.c) but only introduces a skeleton implementation of those functions in preshader.c. Those are actually fleshed out only in the next patch. Many of the new preshader-related definitions can probably also go into the second patch. The traces of the disassembled preshader can also potentially go to a separate patch (leaving only dump_shader_bytecode() in the initial implementation - btw, I think it should be called dump_preshader_bytecode() given the renaming of the other stuff).

2 2