June 2024 - Wine-Commits - list.winehq.org

Rémi Bernon : winevulkan: Remove some unnecessary casts.
by Alexandre Julliard June 7, 2024

June 7, 2024

Module: wine Branch: master Commit: 1a4cbba795c09e421a89d852ed2b68f5e48012b4 URL: https://gitlab.winehq.org/wine/wine/-/commit/1a4cbba795c09e421a89d852ed2b68… Author: Rémi Bernon <rbernon(a)codeweavers.com> Date: Fri Jun 7 15:24:35 2024 +0200 winevulkan: Remove some unnecessary casts. Later, for void pointers, the casts would be (const PTR32 *) and fail to assign. --- dlls/winevulkan/make_vulkan | 6 +- dlls/winevulkan/vulkan_thunks.c | 438 ++++++++++++++++++++-------------------- 2 files changed, 223 insertions(+), 221 deletions(-)

1 0

Alistair Leslie-Hughes : odbccp32: Check if a full path was supplied for Driver/Setup/Translator entries.
by Alexandre Julliard June 7, 2024

June 7, 2024

Module: wine Branch: master Commit: be87c0838159898e361fb77710a3c1bf4c7b7c84 URL: https://gitlab.winehq.org/wine/wine/-/commit/be87c0838159898e361fb77710a3c1… Author: Alistair Leslie-Hughes <leslie_alistair(a)hotmail.com> Date: Fri Jun 7 21:13:39 2024 +1000 odbccp32: Check if a full path was supplied for Driver/Setup/Translator entries. --- dlls/odbccp32/odbccp32.c | 28 ++++++++++++++++++++-------- 1 file changed, 20 insertions(+), 8 deletions(-) diff --git a/dlls/odbccp32/odbccp32.c b/dlls/odbccp32/odbccp32.c index 90875121438..62e35e85f75 100644 --- a/dlls/odbccp32/odbccp32.c +++ b/dlls/odbccp32/odbccp32.c @@ -997,16 +997,28 @@ static void write_registry_values(const WCHAR *regkey, const WCHAR *driver, cons if(lstrcmpiW(driverW, entry) == 0 || lstrcmpiW(setupW, entry) == 0 || lstrcmpiW(translator, entry) == 0) { - len = lstrlenW(path) + lstrlenW(slash) + lstrlenW(divider) + 1; - value = malloc(len * sizeof(WCHAR)); - if(!value) + if(GetFileAttributesW(divider) == INVALID_FILE_ATTRIBUTES) { - ERR("Out of memory\n"); - return; + len = lstrlenW(path) + lstrlenW(slash) + lstrlenW(divider) + 1; + value = malloc(len * sizeof(WCHAR)); + if(!value) + { + ERR("Out of memory\n"); + return; + } + + lstrcpyW(value, path); + lstrcatW(value, slash); + } + else + { + value = calloc(1, (lstrlenW(divider)+1) * sizeof(WCHAR)); + if(!value) + { + ERR("Out of memory\n"); + return; + } } - - lstrcpyW(value, path); - lstrcatW(value, slash); lstrcatW(value, divider); } else

1 0

Rémi Bernon : include: Add and fix some WGL prototypes.
by Alexandre Julliard June 7, 2024

June 7, 2024

Module: wine Branch: master Commit: bb772c2f3e08247cf83b84f4f118ff1e8e14e2d9 URL: https://gitlab.winehq.org/wine/wine/-/commit/bb772c2f3e08247cf83b84f4f118ff… Author: Rémi Bernon <rbernon(a)codeweavers.com> Date: Tue Jun 4 20:42:48 2024 +0200 include: Add and fix some WGL prototypes. --- dlls/opengl32/wgl.c | 12 ++++-------- include/wingdi.h | 11 ++++++++++- 2 files changed, 14 insertions(+), 9 deletions(-) diff --git a/dlls/opengl32/wgl.c b/dlls/opengl32/wgl.c index ff97d31b3eb..903e402d0d0 100644 --- a/dlls/opengl32/wgl.c +++ b/dlls/opengl32/wgl.c @@ -439,14 +439,10 @@ BOOL WINAPI wglDescribeLayerPlane(HDC hdc, /*********************************************************************** * wglGetLayerPaletteEntries (OPENGL32.@) */ -int WINAPI wglGetLayerPaletteEntries(HDC hdc, - int iLayerPlane, - int iStart, - int cEntries, - const COLORREF *pcr) { - FIXME("(): stub!\n"); - - return 0; +int WINAPI wglGetLayerPaletteEntries( HDC hdc, int plane, int start, int count, COLORREF *colors ) +{ + FIXME( "hdc %p, plane %d, start %d, count %d, colors %p, stub!\n", hdc, plane, start, count, colors ); + return 0; } /*********************************************************************** diff --git a/include/wingdi.h b/include/wingdi.h index ef1b3e310aa..039436fbbf1 100644 --- a/include/wingdi.h +++ b/include/wingdi.h @@ -4192,6 +4192,14 @@ WINGDIAPI BOOL WINAPI PolyTextOutW(HDC,const POLYTEXTW*,INT); #define WGL_FONT_LINES 0 #define WGL_FONT_POLYGONS 1 +typedef struct _WGLSWAP +{ + HDC hdc; + UINT uiFlags; +} WGLSWAP, *PWGLSWAP, *LPWGLSWAP; + +#define WGL_SWAPMULTIPLE_MAX 16 + /* WGL prototypes */ WGLAPI HGLRC WINAPI wglCreateContext(HDC); WGLAPI HGLRC WINAPI wglCreateLayerContext(HDC,INT); @@ -4200,13 +4208,14 @@ WGLAPI BOOL WINAPI wglDeleteContext(HGLRC); WGLAPI BOOL WINAPI wglDescribeLayerPlane(HDC,INT,INT,UINT,LPLAYERPLANEDESCRIPTOR); WGLAPI HGLRC WINAPI wglGetCurrentContext(void); WGLAPI HDC WINAPI wglGetCurrentDC(void); -WGLAPI INT WINAPI wglGetLayerPaletteEntries(HDC,INT,INT,INT,const COLORREF *); +WGLAPI INT WINAPI wglGetLayerPaletteEntries(HDC,INT,INT,INT,COLORREF *); WGLAPI PROC WINAPI wglGetProcAddress(LPCSTR); WGLAPI BOOL WINAPI wglMakeCurrent(HDC,HGLRC); WGLAPI BOOL WINAPI wglRealizeLayerPalette(HDC,INT,BOOL); WGLAPI INT WINAPI wglSetLayerPaletteEntries(HDC,INT,INT,INT,const COLORREF *); WGLAPI BOOL WINAPI wglShareLists(HGLRC,HGLRC); WGLAPI BOOL WINAPI wglSwapLayerBuffers(HDC,UINT); +WGLAPI DWORD WINAPI wglSwapMultipleBuffers(UINT,const WGLSWAP *); WGLAPI BOOL WINAPI wglUseFontBitmapsA(HDC,DWORD,DWORD,DWORD); WGLAPI BOOL WINAPI wglUseFontBitmapsW(HDC,DWORD,DWORD,DWORD); #define wglUseFontBitmaps WINELIB_NAME_AW(wglUseFontBitmaps)

1 0

Jacob Pfeiffer : wininet: Unify timeout values closer to hInternet.
by Alexandre Julliard June 7, 2024

June 7, 2024

Module: wine Branch: master Commit: bc19f6f66d8d74fc4d2df589532c3b6977645dd4 URL: https://gitlab.winehq.org/wine/wine/-/commit/bc19f6f66d8d74fc4d2df589532c3b… Author: Jacob Pfeiffer <jacob(a)pfeiffer.codes> Date: Thu Jun 6 17:37:40 2024 -0400 wininet: Unify timeout values closer to hInternet. --- dlls/wininet/http.c | 86 +++-------------- dlls/wininet/internet.c | 217 ++++++++++++++++++++++++++++++++++-------- dlls/wininet/internet.h | 12 +-- dlls/wininet/tests/internet.c | 184 +++++++++++++++++++++++++++++++++-- 4 files changed, 370 insertions(+), 129 deletions(-)

1 0

Giovanni Mascellani : d3d11/tests: Add a test for NV12 textures.
by Alexandre Julliard June 7, 2024

June 7, 2024

Module: wine Branch: master Commit: 864aebbfc66ad1a92d348d2ea3a5ad551e84f522 URL: https://gitlab.winehq.org/wine/wine/-/commit/864aebbfc66ad1a92d348d2ea3a5ad… Author: Giovanni Mascellani <gmascellani(a)codeweavers.com> Date: Fri May 31 17:20:14 2024 +0200 d3d11/tests: Add a test for NV12 textures. --- dlls/d3d11/tests/Makefile.in | 2 +- dlls/d3d11/tests/d3d11.c | 314 +++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 315 insertions(+), 1 deletion(-)

1 0

Elizabeth Figura : wined3d: Use separate signature elements for oFog and oPts.
by Alexandre Julliard June 7, 2024

June 7, 2024

Module: wine Branch: master Commit: ce0ccf1a771492d500e29759ec141f0fc3f92918 URL: https://gitlab.winehq.org/wine/wine/-/commit/ce0ccf1a771492d500e29759ec141f… Author: Elizabeth Figura <zfigura(a)codeweavers.com> Date: Thu May 9 19:55:25 2024 -0500 wined3d: Use separate signature elements for oFog and oPts. --- dlls/wined3d/shader.c | 15 ++++++++++----- 1 file changed, 10 insertions(+), 5 deletions(-) diff --git a/dlls/wined3d/shader.c b/dlls/wined3d/shader.c index 09ecc51b608..252046f0a39 100644 --- a/dlls/wined3d/shader.c +++ b/dlls/wined3d/shader.c @@ -1478,7 +1478,7 @@ static HRESULT shader_get_registers_used(struct wined3d_shader *shader, DWORD co case 2: /* oPts */ reg_maps->output_registers |= 1u << 11; - shader_signature_from_usage(&output_signature_elements[11], + shader_signature_from_usage(&output_signature_elements[12], WINED3D_DECL_USAGE_PSIZE, 0, 11, WINED3DSP_WRITEMASK_1); break; } @@ -1822,8 +1822,14 @@ static HRESULT shader_get_registers_used(struct wined3d_shader *shader, DWORD co } else if (reg_maps->output_registers) { - unsigned int count = wined3d_popcount(reg_maps->output_registers); struct wined3d_shader_signature_element *e; + unsigned int count = 0; + + for (i = 0; i < ARRAY_SIZE(output_signature_elements); ++i) + { + if (output_signature_elements[i].semantic_name) + ++count; + } if (!(output_signature->elements = calloc(count, sizeof(*output_signature->elements)))) return E_OUTOFMEMORY; @@ -1832,9 +1838,8 @@ static HRESULT shader_get_registers_used(struct wined3d_shader *shader, DWORD co e = output_signature->elements; for (i = 0; i < ARRAY_SIZE(output_signature_elements); ++i) { - if (!(reg_maps->output_registers & (1u << i))) - continue; - *e++ = output_signature_elements[i]; + if (output_signature_elements[i].semantic_name) + *e++ = output_signature_elements[i]; } }

1 0

Elizabeth Figura : wined3d: Do not create a framebuffer with attachments whose clear is delayed.
by Alexandre Julliard June 7, 2024

June 7, 2024

Module: wine Branch: master Commit: 14b97f8cfd31f34649823aa5ffcfeda0e322c897 URL: https://gitlab.winehq.org/wine/wine/-/commit/14b97f8cfd31f34649823aa5ffcfed… Author: Elizabeth Figura <zfigura(a)codeweavers.com> Date: Mon Apr 29 14:42:53 2024 -0500 wined3d: Do not create a framebuffer with attachments whose clear is delayed. They may not have a VkImage yet, and if we are delaying their clear we will not have ensured that they have one. --- dlls/wined3d/texture.c | 25 +++++++++++-------------- 1 file changed, 11 insertions(+), 14 deletions(-) diff --git a/dlls/wined3d/texture.c b/dlls/wined3d/texture.c index 688dc35de98..eeb2ea3a7cc 100644 --- a/dlls/wined3d/texture.c +++ b/dlls/wined3d/texture.c @@ -6254,12 +6254,13 @@ static void vk_blitter_clear_rendertargets(struct wined3d_context_vk *context_vk const struct wined3d_fb_state *fb, unsigned int rect_count, const RECT *clear_rects, const RECT *draw_rect, uint32_t flags, const struct wined3d_color *colour, float depth, unsigned int stencil) { + unsigned int i, attachment_count, immediate_rt_count = 0, delay_count = 0; VkClearValue clear_values[WINED3D_MAX_RENDER_TARGETS + 1]; VkImageView views[WINED3D_MAX_RENDER_TARGETS + 1]; - unsigned int i, attachment_count, delay_count = 0; struct wined3d_rendertarget_view_vk *rtv_vk; struct wined3d_rendertarget_view *view; const struct wined3d_vk_info *vk_info; + struct wined3d_fb_state immediate_fb; struct wined3d_device_vk *device_vk; VkCommandBuffer vk_command_buffer; VkRenderPassBeginInfo begin_desc; @@ -6313,6 +6314,7 @@ static void vk_blitter_clear_rendertargets(struct wined3d_context_vk *context_vk wined3d_rendertarget_view_validate_location(view, view->resource->draw_binding); wined3d_rendertarget_view_invalidate_location(view, ~view->resource->draw_binding); + immediate_fb.render_targets[immediate_rt_count++] = view; rtv_vk = wined3d_rendertarget_view_vk(view); views[attachment_count] = wined3d_rendertarget_view_vk_get_image_view(rtv_vk, context_vk); wined3d_rendertarget_view_vk_barrier(rtv_vk, context_vk, WINED3D_BIND_RENDER_TARGET); @@ -6349,6 +6351,7 @@ static void vk_blitter_clear_rendertargets(struct wined3d_context_vk *context_vk wined3d_rendertarget_view_validate_location(view, view->resource->draw_binding); wined3d_rendertarget_view_invalidate_location(view, ~view->resource->draw_binding); + immediate_fb.depth_stencil = view; rtv_vk = wined3d_rendertarget_view_vk(view); views[attachment_count] = wined3d_rendertarget_view_vk_get_image_view(rtv_vk, context_vk); wined3d_rendertarget_view_vk_barrier(rtv_vk, context_vk, WINED3D_BIND_DEPTH_STENCIL); @@ -6384,8 +6387,8 @@ static void vk_blitter_clear_rendertargets(struct wined3d_context_vk *context_vk if (delay_count) TRACE_(d3d_perf)("Partial clear: %u immediate, %u delayed.\n", attachment_count, delay_count); - if (!(vk_render_pass = wined3d_context_vk_get_render_pass(context_vk, fb, - rt_count, flags & (WINED3DCLEAR_ZBUFFER | WINED3DCLEAR_STENCIL), flags))) + if (!(vk_render_pass = wined3d_context_vk_get_render_pass(context_vk, &immediate_fb, + immediate_rt_count, flags & (WINED3DCLEAR_ZBUFFER | WINED3DCLEAR_STENCIL), flags))) { ERR("Failed to get render pass.\n"); return; @@ -6441,19 +6444,13 @@ static void vk_blitter_clear_rendertargets(struct wined3d_context_vk *context_vk wined3d_context_vk_destroy_vk_framebuffer(context_vk, vk_framebuffer, context_vk->current_command_buffer.id); - for (i = 0; i < rt_count; ++i) - { - if (!(view = fb->render_targets[i])) - continue; - - wined3d_context_vk_reference_rendertarget_view(context_vk, wined3d_rendertarget_view_vk(view)); - } + for (i = 0; i < immediate_rt_count; ++i) + wined3d_context_vk_reference_rendertarget_view(context_vk, + wined3d_rendertarget_view_vk(immediate_fb.render_targets[i])); if (depth_stencil) - { - view = fb->depth_stencil; - wined3d_context_vk_reference_rendertarget_view(context_vk, wined3d_rendertarget_view_vk(view)); - } + wined3d_context_vk_reference_rendertarget_view(context_vk, + wined3d_rendertarget_view_vk(immediate_fb.depth_stencil)); } static void vk_blitter_clear(struct wined3d_blitter *blitter, struct wined3d_device *device,

1 0

Elizabeth Figura : wined3d: Handle a null vertex declaration in glsl_vertex_pipe_vdecl().
by Alexandre Julliard June 7, 2024

June 7, 2024

Module: wine Branch: master Commit: dc6b34987a7bc0c3c84d0aabb0f1c4f179b98d27 URL: https://gitlab.winehq.org/wine/wine/-/commit/dc6b34987a7bc0c3c84d0aabb0f1c4… Author: Elizabeth Figura <zfigura(a)codeweavers.com> Date: Thu Jun 6 21:30:32 2024 -0500 wined3d: Handle a null vertex declaration in glsl_vertex_pipe_vdecl(). Fixes: db63c8be29fdadf95e4ce2788f5b39e58b0cfe86 --- dlls/wined3d/glsl_shader.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/dlls/wined3d/glsl_shader.c b/dlls/wined3d/glsl_shader.c index b4bfbe1c55b..feea9957069 100644 --- a/dlls/wined3d/glsl_shader.c +++ b/dlls/wined3d/glsl_shader.c @@ -11765,10 +11765,10 @@ static void glsl_vertex_pipe_vdecl(struct wined3d_context *context, const BOOL legacy_clip_planes = needs_legacy_glsl_syntax(gl_info); BOOL transformed = context->stream_info.position_transformed; BOOL wasrhw = context->last_was_rhw; - bool point_size = vdecl->point_size; - bool specular = vdecl->specular; - bool diffuse = vdecl->diffuse; - bool normal = vdecl->normal; + bool point_size = vdecl && vdecl->point_size; + bool specular = vdecl && vdecl->specular; + bool diffuse = vdecl && vdecl->diffuse; + bool normal = vdecl && vdecl->normal; unsigned int i; context->last_was_rhw = transformed;

1 0

Elizabeth Figura : wined3d: Always output normalized fog coordinates from the vertex shader.
by Alexandre Julliard June 7, 2024

June 7, 2024

Module: wine Branch: master Commit: 421059c88052af596cd4654e4cd772d15dc6b5b9 URL: https://gitlab.winehq.org/wine/wine/-/commit/421059c88052af596cd4654e4cd772… Author: Elizabeth Figura <zfigura(a)codeweavers.com> Date: Sun May 5 12:16:52 2024 -0500 wined3d: Always output normalized fog coordinates from the vertex shader. Historically we used a 0-255 scale in cases where the fog coordinate attribute was passed through directly, apparently in order to avoid doing float division on the value passed to glFogCoordfEXT(). This benefit was probably always dubious, and now is simply unnecessary as we no longer use GL FFP fog nor immediate mode draws. Get rid of the GLSL equivalent code to scale fog up to the 0-255 range and then back down to 0-1. --- dlls/wined3d/glsl_shader.c | 7 +++---- dlls/wined3d/utils.c | 5 ----- dlls/wined3d/wined3d_private.h | 1 - 3 files changed, 3 insertions(+), 10 deletions(-) diff --git a/dlls/wined3d/glsl_shader.c b/dlls/wined3d/glsl_shader.c index feea9957069..b45494b9f12 100644 --- a/dlls/wined3d/glsl_shader.c +++ b/dlls/wined3d/glsl_shader.c @@ -9297,7 +9297,7 @@ static GLuint shader_glsl_generate_ffp_vertex_shader(struct shader_glsl_priv *pr break; case WINED3D_FFP_VS_FOG_FOGCOORD: - shader_addline(buffer, "ffp_varying_fogcoord = ffp_attrib_specular.w * 255.0;\n"); + shader_addline(buffer, "ffp_varying_fogcoord = ffp_attrib_specular.w;\n"); break; case WINED3D_FFP_VS_FOG_RANGE: @@ -12240,10 +12240,9 @@ static void glsl_fragment_pipe_fog(struct wined3d_context *context, if (state->render_states[WINED3D_RS_FOGTABLEMODE] == WINED3D_FOG_NONE) { - if (use_vshader) + if (use_vshader || state->render_states[WINED3D_RS_FOGVERTEXMODE] == WINED3D_FOG_NONE + || context->stream_info.position_transformed) new_source = FOGSOURCE_VS; - else if (state->render_states[WINED3D_RS_FOGVERTEXMODE] == WINED3D_FOG_NONE || context->stream_info.position_transformed) - new_source = FOGSOURCE_COORD; else new_source = FOGSOURCE_FFP; } diff --git a/dlls/wined3d/utils.c b/dlls/wined3d/utils.c index 8c90908ec76..779af782b6c 100644 --- a/dlls/wined3d/utils.c +++ b/dlls/wined3d/utils.c @@ -5834,11 +5834,6 @@ void get_fog_start_end(const struct wined3d_context *context, const struct wined *end = 0.0f; break; - case FOGSOURCE_COORD: - *start = 255.0f; - *end = 0.0f; - break; - case FOGSOURCE_FFP: tmpvalue.d = state->render_states[WINED3D_RS_FOGSTART]; *start = tmpvalue.f; diff --git a/dlls/wined3d/wined3d_private.h b/dlls/wined3d/wined3d_private.h index abdb97f9d9f..f49a604422d 100644 --- a/dlls/wined3d/wined3d_private.h +++ b/dlls/wined3d/wined3d_private.h @@ -1827,7 +1827,6 @@ void dispatch_compute(struct wined3d_device *device, const struct wined3d_state enum fogsource { FOGSOURCE_FFP, FOGSOURCE_VS, - FOGSOURCE_COORD, }; /* Direct3D terminology with little modifications. We do not have an issued

1 0

Alexandre Julliard : msvcrt: Implement asm sqrt functions on ARM platforms.
by Alexandre Julliard June 7, 2024

June 7, 2024

Module: wine Branch: master Commit: 2fc773199080dedbe64a33ef66be02a4f44bcd08 URL: https://gitlab.winehq.org/wine/wine/-/commit/2fc773199080dedbe64a33ef66be02… Author: Alexandre Julliard <julliard(a)winehq.org> Date: Fri Jun 7 10:21:34 2024 +0200 msvcrt: Implement asm sqrt functions on ARM platforms. --- dlls/msvcrt/math.c | 67 ++++++++++++++++++++++++++++++------------------------ 1 file changed, 37 insertions(+), 30 deletions(-) diff --git a/dlls/msvcrt/math.c b/dlls/msvcrt/math.c index cbe17463b03..738d4138c99 100644 --- a/dlls/msvcrt/math.c +++ b/dlls/msvcrt/math.c @@ -77,14 +77,12 @@ void msvcrt_init_math( void *module ) #endif } -#if defined(__i386__) || defined(__x86_64__) static inline double ret_nan( BOOL update_sw ) { double x = 1.0; if (!update_sw) return -NAN; return (x - x) / (x - x); } -#endif #define SET_X87_CW(MASK) \ "subl $4, %esp\n\t" \ @@ -260,7 +258,7 @@ float CDECL MSVCRT_atanf( float x ) } #endif -#ifdef __x86_64__ +#ifndef __i386__ extern short CDECL _fdclass(float x); static BOOL sqrtf_validate( float *x ) @@ -278,10 +276,21 @@ static BOOL sqrtf_validate( float *x ) return TRUE; } -float CDECL sse2_sqrtf(float); -__ASM_GLOBAL_FUNC( sse2_sqrtf, - "sqrtss %xmm0, %xmm0\n\t" - "ret" ) +#ifdef __arm64ec__ +static float __attribute__((naked)) CDECL asm_sqrtf(float x) +{ + asm( "fsqrt s0,s0; ret" ); +} +#elif defined __aarch64__ +float CDECL asm_sqrtf(float); +__ASM_GLOBAL_FUNC( asm_sqrtf, "fsqrt s0,s0; ret" ) +#elif defined __arm__ +float CDECL asm_sqrtf(float); +__ASM_GLOBAL_FUNC( asm_sqrtf, "vsqrt s0,s0; bx lr" ) +#elif defined __x86_64__ +float CDECL asm_sqrtf(float); +__ASM_GLOBAL_FUNC( asm_sqrtf, "sqrtss %xmm0, %xmm0; ret" ) +#endif #endif /********************************************************************* @@ -289,11 +298,11 @@ __ASM_GLOBAL_FUNC( sse2_sqrtf, */ float CDECL MSVCRT_sqrtf( float x ) { -#ifdef __x86_64__ +#ifndef __i386__ if (!sqrtf_validate(&x)) return x; - return sse2_sqrtf(x); + return asm_sqrtf(x); #else return sqrtf( x ); #endif @@ -382,7 +391,6 @@ double CDECL MSVCRT_exp( double x ) } #endif -#if defined(__x86_64__) || defined(__i386__) extern short CDECL _dclass(double x); static BOOL sqrt_validate( double *x, BOOL update_sw ) @@ -410,15 +418,23 @@ static BOOL sqrt_validate( double *x, BOOL update_sw ) return TRUE; } -double CDECL sse2_sqrt(double); -__ASM_GLOBAL_FUNC( sse2_sqrt, - "sqrtsd %xmm0, %xmm0\n\t" - "ret" ) -#endif - -#ifdef __i386__ -double CDECL x87_sqrt(double); -__ASM_GLOBAL_FUNC( x87_sqrt, +#ifdef __arm64ec__ +static double __attribute__((naked)) CDECL asm_sqrt(double x) +{ + asm( "fsqrt d0,d0; ret" ); +} +#elif defined __aarch64__ +double CDECL asm_sqrt(double); +__ASM_GLOBAL_FUNC( asm_sqrt, "fsqrt d0,d0; ret" ) +#elif defined __arm__ +double CDECL asm_sqrt(double); +__ASM_GLOBAL_FUNC( asm_sqrt, "vsqrt d0,d0; bx lr" ) +#elif defined __x86_64__ +double CDECL asm_sqrt(double); +__ASM_GLOBAL_FUNC( asm_sqrt, "sqrtsd %xmm0, %xmm0; ret" ) +#elif defined __i386__ +double CDECL asm_sqrt(double); +__ASM_GLOBAL_FUNC( asm_sqrt, "fldl 4(%esp)\n\t" SET_X87_CW(0xc00) "fsqrt\n\t" @@ -431,19 +447,10 @@ __ASM_GLOBAL_FUNC( x87_sqrt, */ double CDECL MSVCRT_sqrt( double x ) { -#ifdef __x86_64__ if (!sqrt_validate(&x, TRUE)) return x; - return sse2_sqrt(x); -#elif defined( __i386__ ) - if (!sqrt_validate(&x, TRUE)) - return x; - - return x87_sqrt(x); -#else - return sqrt( x ); -#endif + return asm_sqrt(x); } /********************************************************************* @@ -2561,7 +2568,7 @@ void __cdecl __libm_sse2_sqrt_precise(void) __asm__ __volatile__( "movq %0,%%xmm0" : : "m" (d) ); return; } - __asm__ __volatile__( "call " __ASM_NAME( "sse2_sqrt" ) ); + __asm__ __volatile__( "call " __ASM_NAME( "asm_sqrt" ) ); } #endif /* __i386__ */

1 0