April 2024 - Wine-commits - list.winehq.org

Conor McCarthy : vkd3d-shader/dxil: Implement DX intrinsics Texture2DMSGetSamplePosition and RenderTargetGetSamplePosition.
by Alexandre Julliard 17 Apr '24

17 Apr '24

Module: vkd3d Branch: master Commit: fe5e821cca2c263f2ab29ac6b2b31f8a33431ad0 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/fe5e821cca2c263f2ab29ac6b2b31… Author: Conor McCarthy <cmccarthy(a)codeweavers.com> Date: Thu Jun 15 11:53:48 2023 +1000 vkd3d-shader/dxil: Implement DX intrinsics Texture2DMSGetSamplePosition and RenderTargetGetSamplePosition. --- libs/vkd3d-shader/dxil.c | 58 +++++++++++++++++++++++++++++++ tests/hlsl/get-sample-pos.shader_test | 2 +- tests/hlsl/rt-get-sample-info.shader_test | 4 +-- 3 files changed, 61 insertions(+), 3 deletions(-) diff --git a/libs/vkd3d-shader/dxil.c b/libs/vkd3d-shader/dxil.c index c03e20a7..031d4b01 100644 --- a/libs/vkd3d-shader/dxil.c +++ b/libs/vkd3d-shader/dxil.c @@ -401,6 +401,8 @@ enum dx_intrinsic_opcode DX_GET_DIMENSIONS = 72, DX_TEXTURE_GATHER = 73, DX_TEXTURE_GATHER_CMP = 74, + DX_TEX2DMS_GET_SAMPLE_POS = 75, + DX_RT_GET_SAMPLE_POS = 76, DX_ATOMIC_BINOP = 78, DX_ATOMIC_CMP_XCHG = 79, DX_BARRIER = 80, @@ -2576,6 +2578,26 @@ static bool sm6_value_validate_is_texture_handle(const struct sm6_value *value, return true; } +static bool sm6_value_validate_is_texture_2dms_handle(const struct sm6_value *value, enum dx_intrinsic_opcode op, + struct sm6_parser *sm6) +{ + enum dxil_resource_kind kind; + + if (!sm6_value_validate_is_handle(value, sm6)) + return false; + + kind = value->u.handle.d->kind; + if (!resource_kind_is_multisampled(kind)) + { + WARN("Resource kind %u for op %u is not a 2DMS texture.\n", kind, op); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_RESOURCE_HANDLE, + "Resource kind %u for texture operation %u is not a 2DMS texture.", kind, op); + return false; + } + + return true; +} + static bool sm6_value_validate_is_sampler_handle(const struct sm6_value *value, enum dx_intrinsic_opcode op, struct sm6_parser *sm6) { @@ -5191,6 +5213,40 @@ static void sm6_parser_emit_dx_buffer_store(struct sm6_parser *sm6, enum dx_intr dst_param->reg = resource->u.handle.reg; } +static void sm6_parser_emit_dx_get_sample_pos(struct sm6_parser *sm6, enum dx_intrinsic_opcode op, + const struct sm6_value **operands, struct function_emission_state *state) +{ + struct vkd3d_shader_instruction *ins = state->ins; + struct vkd3d_shader_src_param *src_params; + const struct sm6_value *resource = NULL; + + if (op == DX_TEX2DMS_GET_SAMPLE_POS) + { + resource = operands[0]; + if (!sm6_value_validate_is_texture_2dms_handle(resource, op, sm6)) + return; + } + + vsir_instruction_init(ins, &sm6->p.location, VKD3DSIH_SAMPLE_POS); + + if (!(src_params = instruction_src_params_alloc(ins, 2, sm6))) + return; + if (op == DX_TEX2DMS_GET_SAMPLE_POS) + { + src_param_init_vector_from_reg(&src_params[0], &resource->u.handle.reg); + src_param_init_from_value(&src_params[1], operands[1]); + } + else + { + src_param_init_vector(&src_params[0], 2); + vsir_register_init(&src_params[0].reg, VKD3DSPR_RASTERIZER, VKD3D_DATA_FLOAT, 0); + src_params[0].reg.dimension = VSIR_DIMENSION_VEC4; + src_param_init_from_value(&src_params[1], operands[0]); + } + + instruction_dst_param_init_ssa_vector(ins, 2, sm6); +} + static unsigned int sm6_value_get_texel_offset(const struct sm6_value *value) { return sm6_value_is_undef(value) ? 0 : sm6_value_get_constant_uint(value); @@ -5639,6 +5695,7 @@ static const struct sm6_dx_opcode_info sm6_dx_op_table[] = [DX_ROUND_PI ] = {"g", "R", sm6_parser_emit_dx_unary}, [DX_ROUND_Z ] = {"g", "R", sm6_parser_emit_dx_unary}, [DX_RSQRT ] = {"g", "R", sm6_parser_emit_dx_unary}, + [DX_RT_GET_SAMPLE_POS ] = {"o", "i", sm6_parser_emit_dx_get_sample_pos}, [DX_SAMPLE ] = {"o", "HHffffiiif", sm6_parser_emit_dx_sample}, [DX_SAMPLE_B ] = {"o", "HHffffiiiff", sm6_parser_emit_dx_sample}, [DX_SAMPLE_C ] = {"o", "HHffffiiiff", sm6_parser_emit_dx_sample}, @@ -5651,6 +5708,7 @@ static const struct sm6_dx_opcode_info sm6_dx_op_table[] = [DX_SQRT ] = {"g", "R", sm6_parser_emit_dx_unary}, [DX_STORE_OUTPUT ] = {"v", "ii8o", sm6_parser_emit_dx_store_output}, [DX_TAN ] = {"g", "R", sm6_parser_emit_dx_unary}, + [DX_TEX2DMS_GET_SAMPLE_POS ] = {"o", "Hi", sm6_parser_emit_dx_get_sample_pos}, [DX_TEXTURE_GATHER ] = {"o", "HHffffiic", sm6_parser_emit_dx_texture_gather}, [DX_TEXTURE_GATHER_CMP ] = {"o", "HHffffiicf", sm6_parser_emit_dx_texture_gather}, [DX_TEXTURE_LOAD ] = {"o", "HiiiiCCC", sm6_parser_emit_dx_texture_load}, diff --git a/tests/hlsl/get-sample-pos.shader_test b/tests/hlsl/get-sample-pos.shader_test index 23dbba67..69b9101b 100644 --- a/tests/hlsl/get-sample-pos.shader_test +++ b/tests/hlsl/get-sample-pos.shader_test @@ -15,5 +15,5 @@ float4 main() : sv_target } [test] -todo draw quad +todo(sm<6 | glsl) draw quad probe all rgba (-0.125, -0.375, 0.375, -0.125) diff --git a/tests/hlsl/rt-get-sample-info.shader_test b/tests/hlsl/rt-get-sample-info.shader_test index 52353f29..c8721748 100644 --- a/tests/hlsl/rt-get-sample-info.shader_test +++ b/tests/hlsl/rt-get-sample-info.shader_test @@ -27,8 +27,8 @@ float4 main() : sv_target [test] uniform 0 uint 0 -todo draw quad +todo(sm<6 | glsl) draw quad probe all rgba (-0.125, -0.375, 0.375, -0.125) uniform 0 uint 2 -todo draw quad +todo(sm<6 | glsl) draw quad probe all rgba (-0.375, 0.125, 0.125, 0.375)

1 0

Conor McCarthy : vkd3d-shader/dxil: Implement DX intrinsic RenderTargetGetSampleCount.
by Alexandre Julliard 17 Apr '24

17 Apr '24

Module: vkd3d Branch: master Commit: 99996ed5b57f2d5603409d6f413fee15db8b7b99 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/99996ed5b57f2d5603409d6f413fe… Author: Conor McCarthy <cmccarthy(a)codeweavers.com> Date: Thu Jun 15 11:59:34 2023 +1000 vkd3d-shader/dxil: Implement DX intrinsic RenderTargetGetSampleCount. --- libs/vkd3d-shader/dxil.c | 21 +++++++++++++++++++++ tests/hlsl/rt-get-sample-info.shader_test | 2 +- 2 files changed, 22 insertions(+), 1 deletion(-) diff --git a/libs/vkd3d-shader/dxil.c b/libs/vkd3d-shader/dxil.c index 031d4b01..2553d903 100644 --- a/libs/vkd3d-shader/dxil.c +++ b/libs/vkd3d-shader/dxil.c @@ -403,6 +403,7 @@ enum dx_intrinsic_opcode DX_TEXTURE_GATHER_CMP = 74, DX_TEX2DMS_GET_SAMPLE_POS = 75, DX_RT_GET_SAMPLE_POS = 76, + DX_RT_GET_SAMPLE_COUNT = 77, DX_ATOMIC_BINOP = 78, DX_ATOMIC_CMP_XCHG = 79, DX_BARRIER = 80, @@ -5213,6 +5214,25 @@ static void sm6_parser_emit_dx_buffer_store(struct sm6_parser *sm6, enum dx_intr dst_param->reg = resource->u.handle.reg; } +static void sm6_parser_emit_dx_get_sample_count(struct sm6_parser *sm6, enum dx_intrinsic_opcode op, + const struct sm6_value **operands, struct function_emission_state *state) +{ + struct vkd3d_shader_instruction *ins = state->ins; + struct vkd3d_shader_src_param *src_param; + + vsir_instruction_init(ins, &sm6->p.location, VKD3DSIH_SAMPLE_INFO); + ins->flags = VKD3DSI_SAMPLE_INFO_UINT; + + if (!(src_param = instruction_src_params_alloc(ins, 1, sm6))) + return; + vsir_register_init(&src_param->reg, VKD3DSPR_RASTERIZER, VKD3D_DATA_FLOAT, 0); + src_param->reg.dimension = VSIR_DIMENSION_VEC4; + src_param_init(src_param); + + instruction_dst_param_init_ssa_scalar(ins, sm6); + ins->dst->reg.data_type = VKD3D_DATA_FLOAT; +} + static void sm6_parser_emit_dx_get_sample_pos(struct sm6_parser *sm6, enum dx_intrinsic_opcode op, const struct sm6_value **operands, struct function_emission_state *state) { @@ -5695,6 +5715,7 @@ static const struct sm6_dx_opcode_info sm6_dx_op_table[] = [DX_ROUND_PI ] = {"g", "R", sm6_parser_emit_dx_unary}, [DX_ROUND_Z ] = {"g", "R", sm6_parser_emit_dx_unary}, [DX_RSQRT ] = {"g", "R", sm6_parser_emit_dx_unary}, + [DX_RT_GET_SAMPLE_COUNT ] = {"i", "", sm6_parser_emit_dx_get_sample_count}, [DX_RT_GET_SAMPLE_POS ] = {"o", "i", sm6_parser_emit_dx_get_sample_pos}, [DX_SAMPLE ] = {"o", "HHffffiiif", sm6_parser_emit_dx_sample}, [DX_SAMPLE_B ] = {"o", "HHffffiiiff", sm6_parser_emit_dx_sample}, diff --git a/tests/hlsl/rt-get-sample-info.shader_test b/tests/hlsl/rt-get-sample-info.shader_test index c8721748..eea4f1a4 100644 --- a/tests/hlsl/rt-get-sample-info.shader_test +++ b/tests/hlsl/rt-get-sample-info.shader_test @@ -13,7 +13,7 @@ float4 main() : sv_target } [test] -todo draw quad +todo(sm<6 | glsl) draw quad probe all rgba (4.0, 0.0, 0.0, 0.0)

1 0

Alexandre Julliard : rpcrt4: Export NdrpClientCall2.
by Alexandre Julliard 17 Apr '24

17 Apr '24

Module: wine Branch: master Commit: 2fff2c6b4a5bde69f9438e11191a24fb5439d8b9 URL: https://gitlab.winehq.org/wine/wine/-/commit/2fff2c6b4a5bde69f9438e11191a24… Author: Alexandre Julliard <julliard(a)winehq.org> Date: Wed Apr 17 12:40:59 2024 +0200 rpcrt4: Export NdrpClientCall2. --- dlls/rpcrt4/cproxy.c | 31 ++++++++++++++++--------------- dlls/rpcrt4/ndr_stubless.c | 32 ++++++++++++++++---------------- dlls/rpcrt4/ndr_stubless.h | 2 -- dlls/rpcrt4/rpcrt4.spec | 1 + 4 files changed, 33 insertions(+), 33 deletions(-)

1 0

Alexandre Julliard : rpcrt4: Implement NdrClientCall2 in assembly on all platforms.
by Alexandre Julliard 17 Apr '24

17 Apr '24

Module: wine Branch: master Commit: e994c8ef741ce28ac6971ea5ff6a131c483f0fd7 URL: https://gitlab.winehq.org/wine/wine/-/commit/e994c8ef741ce28ac6971ea5ff6a13… Author: Alexandre Julliard <julliard(a)winehq.org> Date: Wed Apr 17 12:46:10 2024 +0200 rpcrt4: Implement NdrClientCall2 in assembly on all platforms. --- dlls/rpcrt4/ndr_stubless.c | 85 +++++++++++++++++++++++++++++++++------------- 1 file changed, 62 insertions(+), 23 deletions(-) diff --git a/dlls/rpcrt4/ndr_stubless.c b/dlls/rpcrt4/ndr_stubless.c index 1cdb575b9e6..f87817f7fd1 100644 --- a/dlls/rpcrt4/ndr_stubless.c +++ b/dlls/rpcrt4/ndr_stubless.c @@ -1000,8 +1000,48 @@ LONG_PTR WINAPI NdrpClientCall2( PMIDL_STUB_DESC pStubDesc, PFORMAT_STRING pForm return RetVal; } -#ifdef __x86_64__ - +#ifdef __aarch64__ +__ASM_GLOBAL_FUNC( NdrClientCall2, + "stp x29, x30, [sp, #-0x40]!\n\t" + ".seh_save_fplr_x 0x40\n\t" + ".seh_endprologue\n\t" + "stp x2, x3, [sp, #0x10]\n\t" + "stp x4, x5, [sp, #0x20]\n\t" + "stp x6, x7, [sp, #0x30]\n\t" + "add x2, sp, #0x10\n\t" /* stack */ + "mov x3, #0\n\t" /* fpu_stack */ + "bl NdrpClientCall2\n\t" + "ldp x29, x30, [sp], #0x40\n\t" + "ret" ) +#elif defined(__arm64ec__) +CLIENT_CALL_RETURN __attribute__((naked)) NdrClientCall2( PMIDL_STUB_DESC desc, PFORMAT_STRING fmt, ... ) +{ + asm( ".seh_proc \"#NdrClientCall2\"\n\t" + "stp x29, x30, [sp, #-0x10]!\n\t" + ".seh_save_fplr_x 0x10\n\t" + ".seh_endprologue\n\t" + "stp x2, x3, [x4, #-0x10]!\n\t" + "mov x2, x4\n\t" /* stack */ + "mov x3, #0\n\t" /* fpu_stack */ + "bl \"#NdrpClientCall2\"\n\t" + "ldp x29, x30, [sp], #0x10\n\t" + "ret\n\t" + ".seh_endproc" ); +} +#elif defined(__arm__) +__ASM_GLOBAL_FUNC( NdrClientCall2, + "push {r2-r3}\n\t" + ".seh_save_regs {r2,r3}\n\t" + "push {fp,lr}\n\t" + ".seh_save_regs_w {fp,lr}\n\t" + ".seh_endprologue\n\t" + "add r2, sp, #8\n\t" /* stack */ + "mov r3, #0\n\t" /* fpu_stack */ + "bl NdrpClientCall2\n\t" + "pop {fp,lr}\n\t" + "add sp, #8\n\t" + "bx lr" ) +#elif defined(__x86_64__) __ASM_GLOBAL_FUNC( NdrClientCall2, "subq $0x28,%rsp\n\t" __ASM_SEH(".seh_stackalloc 0x28\n\t") @@ -1009,30 +1049,29 @@ __ASM_GLOBAL_FUNC( NdrClientCall2, __ASM_CFI(".cfi_adjust_cfa_offset 0x28\n\t") "movq %r8,0x40(%rsp)\n\t" "movq %r9,0x48(%rsp)\n\t" - "leaq 0x40(%rsp),%r8\n\t" - "xorq %r9,%r9\n\t" + "leaq 0x40(%rsp),%r8\n\t" /* stack */ + "xorq %r9,%r9\n\t" /* fpu_stack */ "call " __ASM_NAME("NdrpClientCall2") "\n\t" "addq $0x28,%rsp\n\t" __ASM_CFI(".cfi_adjust_cfa_offset -0x28\n\t") - "ret" ); - -#else /* __x86_64__ */ - -/*********************************************************************** - * NdrClientCall2 [RPCRT4.@] - */ -CLIENT_CALL_RETURN WINAPIV NdrClientCall2( PMIDL_STUB_DESC desc, PFORMAT_STRING format, ... ) -{ - va_list args; - LONG_PTR ret; - - va_start( args, format ); - ret = NdrpClientCall2( desc, format, va_arg( args, void ** ), NULL ); - va_end( args ); - return *(CLIENT_CALL_RETURN *)&ret; -} - -#endif /* __x86_64__ */ + "ret" ) +#elif defined(__i386__) +__ASM_GLOBAL_FUNC( NdrClientCall2, + "pushl %ebp\n\t" + __ASM_CFI(".cfi_adjust_cfa_offset 4\n\t") + __ASM_CFI(".cfi_rel_offset %ebp,0\n\t") + "movl %esp,%ebp\n\t" + __ASM_CFI(".cfi_def_cfa_register %ebp\n\t") + "push $0\n\t" /* fpu_stack */ + "push 16(%ebp)\n\t" /* stack */ + "push 12(%ebp)\n\t" /* format */ + "push 8(%ebp)\n\t" /* desc */ + "call " __ASM_STDCALL("NdrpClientCall2",16) "\n\t" + "leave\n\t" + __ASM_CFI(".cfi_def_cfa %esp,4\n\t") + __ASM_CFI(".cfi_same_value %ebp\n\t") + "ret" ) +#endif /* Calls a function with the specified arguments, restoring the stack * properly afterwards as we don't know the calling convention of the

1 0

Alexandre Julliard : rpcrt4: Implement NdrClientCall3 in assembly on all platforms.
by Alexandre Julliard 17 Apr '24

17 Apr '24

Module: wine Branch: master Commit: 19548ad5124593daaa43ca351b5c9f02ee2e2bd2 URL: https://gitlab.winehq.org/wine/wine/-/commit/19548ad5124593daaa43ca351b5c9f… Author: Alexandre Julliard <julliard(a)winehq.org> Date: Wed Apr 17 13:04:13 2024 +0200 rpcrt4: Implement NdrClientCall3 in assembly on all platforms. --- dlls/rpcrt4/ndr_stubless.c | 54 +++++++++++++++++++++++++++------------------- 1 file changed, 32 insertions(+), 22 deletions(-) diff --git a/dlls/rpcrt4/ndr_stubless.c b/dlls/rpcrt4/ndr_stubless.c index f87817f7fd1..0642cf77751 100644 --- a/dlls/rpcrt4/ndr_stubless.c +++ b/dlls/rpcrt4/ndr_stubless.c @@ -2291,37 +2291,47 @@ LONG_PTR CDECL ndr64_client_call( MIDL_STUBLESS_PROXY_INFO *info, return 0; } -#ifdef __x86_64__ - +#ifdef __aarch64__ +__ASM_GLOBAL_FUNC( NdrClientCall3, + "stp x29, x30, [sp, #-0x40]!\n\t" + ".seh_save_fplr_x 0x40\n\t" + ".seh_endprologue\n\t" + "str x3, [sp, #0x18]\n\t" + "stp x4, x5, [sp, #0x20]\n\t" + "stp x6, x7, [sp, #0x30]\n\t" + "add x3, sp, #0x18\n\t" /* stack */ + "mov x4, #0\n\t" /* fpu_stack */ + "bl ndr64_client_call\n\t" + "ldp x29, x30, [sp], #0x40\n\t" + "ret" ) +#elif defined(__arm64ec__) +CLIENT_CALL_RETURN __attribute__((naked)) NdrClientCall3( MIDL_STUBLESS_PROXY_INFO *info, ULONG proc, void *retval, ... ) +{ + asm( ".seh_proc \"#NdrClientCall3\"\n\t" + "stp x29, x30, [sp, #-0x10]!\n\t" + ".seh_save_fplr_x 0x10\n\t" + ".seh_endprologue\n\t" + "str x3, [x4, #-0x8]!\n\t" + "mov x3, x4\n\t" /* stack */ + "mov x4, #0\n\t" /* fpu_stack */ + "bl \"#ndr64_client_call\"\n\t" + "ldp x29, x30, [sp], #0x10\n\t" + "ret\n\t" + ".seh_endproc" ); +} +#elif defined(__x86_64__) __ASM_GLOBAL_FUNC( NdrClientCall3, "subq $0x28,%rsp\n\t" __ASM_SEH(".seh_stackalloc 0x28\n\t") __ASM_SEH(".seh_endprologue\n\t") __ASM_CFI(".cfi_adjust_cfa_offset 0x28\n\t") "movq %r9,0x48(%rsp)\n\t" - "leaq 0x48(%rsp),%r9\n\t" - "movq $0,0x20(%rsp)\n\t" + "leaq 0x48(%rsp),%r9\n\t" /* stack */ + "movq $0,0x20(%rsp)\n\t" /* fpu_stack */ "call " __ASM_NAME("ndr64_client_call") "\n\t" "addq $0x28,%rsp\n\t" __ASM_CFI(".cfi_adjust_cfa_offset -0x28\n\t") - "ret" ); - -#elif defined(_WIN64) - -/*********************************************************************** - * NdrClientCall3 [RPCRT4.@] - */ -CLIENT_CALL_RETURN WINAPIV NdrClientCall3( MIDL_STUBLESS_PROXY_INFO *info, ULONG proc, void *retval, ... ) -{ - va_list args; - LONG_PTR ret; - - va_start( args, retval ); - ret = ndr64_client_call( info, proc, retval, va_arg( args, void ** ), NULL ); - va_end( args ); - return *(CLIENT_CALL_RETURN *)&ret; -} - + "ret" ) #endif LONG_PTR CDECL ndr64_async_client_call( MIDL_STUBLESS_PROXY_INFO *info,

1 0

Alexandre Julliard : rpcrt4: Implement NdrAsyncClientCall in assembly on all platforms.
by Alexandre Julliard 17 Apr '24

17 Apr '24

Module: wine Branch: master Commit: 8980dc0fd5d54492fe5ff14ed2e248367d839858 URL: https://gitlab.winehq.org/wine/wine/-/commit/8980dc0fd5d54492fe5ff14ed2e248… Author: Alexandre Julliard <julliard(a)winehq.org> Date: Wed Apr 17 13:08:04 2024 +0200 rpcrt4: Implement NdrAsyncClientCall in assembly on all platforms. --- dlls/rpcrt4/ndr_stubless.c | 84 ++++++++++++++++++++++++++++++++++------------ dlls/rpcrt4/ndr_stubless.h | 2 -- 2 files changed, 62 insertions(+), 24 deletions(-) diff --git a/dlls/rpcrt4/ndr_stubless.c b/dlls/rpcrt4/ndr_stubless.c index 0642cf77751..666a1901022 100644 --- a/dlls/rpcrt4/ndr_stubless.c +++ b/dlls/rpcrt4/ndr_stubless.c @@ -1949,8 +1949,48 @@ cleanup: return status; } -#ifdef __x86_64__ - +#ifdef __aarch64__ +__ASM_GLOBAL_FUNC( NdrAsyncClientCall, + "stp x29, x30, [sp, #-0x40]!\n\t" + ".seh_save_fplr_x 0x40\n\t" + ".seh_endprologue\n\t" + "stp x2, x3, [sp, #0x10]\n\t" + "stp x4, x5, [sp, #0x20]\n\t" + "stp x6, x7, [sp, #0x30]\n\t" + "add x2, sp, #0x10\n\t" /* stack */ + "mov x3, #0\n\t" /* fpu_stack */ + "bl ndr_async_client_call\n\t" + "ldp x29, x30, [sp], #0x40\n\t" + "ret" ) +#elif defined(__arm64ec__) +CLIENT_CALL_RETURN __attribute__((naked)) NdrAsyncClientCall( PMIDL_STUB_DESC desc, PFORMAT_STRING fmt, ... ) +{ + asm( ".seh_proc \"#NdrAsyncClientCall\"\n\t" + "stp x29, x30, [sp, #-0x10]!\n\t" + ".seh_save_fplr_x 0x10\n\t" + ".seh_endprologue\n\t" + "stp x2, x3, [x4, #-0x10]!\n\t" + "mov x2, x4\n\t" /* stack */ + "mov x3, #0\n\t" /* fpu_stack */ + "bl \"#ndr_async_client_call\"\n\t" + "ldp x29, x30, [sp], #0x10\n\t" + "ret\n\t" + ".seh_endproc" ); +} +#elif defined(__arm__) +__ASM_GLOBAL_FUNC( NdrAsyncClientCall, + "push {r2-r3}\n\t" + ".seh_save_regs {r2,r3}\n\t" + "push {fp,lr}\n\t" + ".seh_save_regs_w {fp,lr}\n\t" + ".seh_endprologue\n\t" + "add r2, sp, #8\n\t" /* stack */ + "mov r3, #0\n\t" /* fpu_stack */ + "bl ndr_async_client_call\n\t" + "pop {fp,lr}\n\t" + "add sp, #8\n\t" + "bx lr" ) +#elif defined(__x86_64__) __ASM_GLOBAL_FUNC( NdrAsyncClientCall, "subq $0x28,%rsp\n\t" __ASM_SEH(".seh_stackalloc 0x28\n\t") @@ -1958,29 +1998,29 @@ __ASM_GLOBAL_FUNC( NdrAsyncClientCall, __ASM_CFI(".cfi_adjust_cfa_offset 0x28\n\t") "movq %r8,0x40(%rsp)\n\t" "movq %r9,0x48(%rsp)\n\t" - "leaq 0x40(%rsp),%r8\n\t" + "leaq 0x40(%rsp),%r8\n\t" /* stack */ + "xorq %r9,%r9\n\t" /* fpu_stack */ "call " __ASM_NAME("ndr_async_client_call") "\n\t" "addq $0x28,%rsp\n\t" __ASM_CFI(".cfi_adjust_cfa_offset -0x28\n\t") - "ret" ); - -#else /* __x86_64__ */ - -/*********************************************************************** - * NdrAsyncClientCall [RPCRT4.@] - */ -CLIENT_CALL_RETURN WINAPIV NdrAsyncClientCall( PMIDL_STUB_DESC desc, PFORMAT_STRING format, ... ) -{ - va_list args; - LONG_PTR ret; - - va_start( args, format ); - ret = ndr_async_client_call( desc, format, va_arg( args, void ** )); - va_end( args ); - return *(CLIENT_CALL_RETURN *)&ret; -} - -#endif /* __x86_64__ */ + "ret" ) +#elif defined(__i386__) +__ASM_GLOBAL_FUNC( NdrAsyncClientCall, + "pushl %ebp\n\t" + __ASM_CFI(".cfi_adjust_cfa_offset 4\n\t") + __ASM_CFI(".cfi_rel_offset %ebp,0\n\t") + "movl %esp,%ebp\n\t" + __ASM_CFI(".cfi_def_cfa_register %ebp\n\t") + "push $0\n\t" /* fpu_stack */ + "push 16(%ebp)\n\t" /* stack */ + "push 12(%ebp)\n\t" /* format */ + "push 8(%ebp)\n\t" /* desc */ + "call " __ASM_NAME("ndr_async_client_call") "\n\t" + "leave\n\t" + __ASM_CFI(".cfi_def_cfa %esp,4\n\t") + __ASM_CFI(".cfi_same_value %ebp\n\t") + "ret" ) +#endif RPCRTAPI LONG RPC_ENTRY NdrAsyncStubCall(struct IRpcStubBuffer* pThis, struct IRpcChannelBuffer* pChannel, PRPC_MESSAGE pRpcMsg, diff --git a/dlls/rpcrt4/ndr_stubless.h b/dlls/rpcrt4/ndr_stubless.h index 78c42050f31..0bb04a50b72 100644 --- a/dlls/rpcrt4/ndr_stubless.h +++ b/dlls/rpcrt4/ndr_stubless.h @@ -253,8 +253,6 @@ enum stubless_phase STUBLESS_FREE }; -LONG_PTR CDECL ndr_async_client_call( PMIDL_STUB_DESC pStubDesc, PFORMAT_STRING pFormat, - void **stack_top ); void client_do_args( PMIDL_STUB_MESSAGE pStubMsg, PFORMAT_STRING pFormat, enum stubless_phase phase, void **fpu_args, unsigned short number_of_params, unsigned char *pRetVal ); PFORMAT_STRING convert_old_args( PMIDL_STUB_MESSAGE pStubMsg, PFORMAT_STRING pFormat,

1 0

Alexandre Julliard : rpcrt4: Implement Ndr64AsyncClientCall in assembly on all platforms.
by Alexandre Julliard 17 Apr '24

17 Apr '24

Module: wine Branch: master Commit: b2564ffcd928ff2da6cfea3bbaeffdb5f33e1118 URL: https://gitlab.winehq.org/wine/wine/-/commit/b2564ffcd928ff2da6cfea3bbaeffd… Author: Alexandre Julliard <julliard(a)winehq.org> Date: Wed Apr 17 13:08:16 2024 +0200 rpcrt4: Implement Ndr64AsyncClientCall in assembly on all platforms. --- dlls/rpcrt4/ndr_stubless.c | 54 +++++++++++++++++++++++++++------------------- 1 file changed, 32 insertions(+), 22 deletions(-) diff --git a/dlls/rpcrt4/ndr_stubless.c b/dlls/rpcrt4/ndr_stubless.c index 666a1901022..96dff0e3410 100644 --- a/dlls/rpcrt4/ndr_stubless.c +++ b/dlls/rpcrt4/ndr_stubless.c @@ -2403,35 +2403,45 @@ LONG_PTR CDECL ndr64_async_client_call( MIDL_STUBLESS_PROXY_INFO *info, return 0; } -#ifdef __x86_64__ - +#ifdef __aarch64__ +__ASM_GLOBAL_FUNC( Ndr64AsyncClientCall, + "stp x29, x30, [sp, #-0x40]!\n\t" + ".seh_save_fplr_x 0x40\n\t" + ".seh_endprologue\n\t" + "str x3, [sp, #0x18]\n\t" + "stp x4, x5, [sp, #0x20]\n\t" + "stp x6, x7, [sp, #0x30]\n\t" + "add x3, sp, #0x18\n\t" /* stack */ + "mov x4, #0\n\t" /* fpu_stack */ + "bl ndr64_async_client_call\n\t" + "ldp x29, x30, [sp], #0x40\n\t" + "ret" ) +#elif defined(__arm64ec__) +CLIENT_CALL_RETURN __attribute__((naked)) Ndr64AsyncClientCall( MIDL_STUBLESS_PROXY_INFO *info, ULONG proc, void *retval, ... ) +{ + asm( ".seh_proc \"#Ndr64AsyncClientCall\"\n\t" + "stp x29, x30, [sp, #-0x10]!\n\t" + ".seh_save_fplr_x 0x10\n\t" + ".seh_endprologue\n\t" + "str x3, [x4, #-0x8]!\n\t" + "mov x3, x4\n\t" /* stack */ + "mov x4, #0\n\t" /* fpu_stack */ + "bl \"#ndr64_async_client_call\"\n\t" + "ldp x29, x30, [sp], #0x10\n\t" + "ret\n\t" + ".seh_endproc" ); +} +#elif defined(__x86_64__) __ASM_GLOBAL_FUNC( Ndr64AsyncClientCall, "subq $0x28,%rsp\n\t" __ASM_SEH(".seh_stackalloc 0x28\n\t") __ASM_SEH(".seh_endprologue\n\t") __ASM_CFI(".cfi_adjust_cfa_offset 0x28\n\t") "movq %r9,0x48(%rsp)\n\t" - "leaq 0x48(%rsp),%r9\n\t" - "movq $0,0x20(%rsp)\n\t" + "leaq 0x48(%rsp),%r9\n\t" /* stack */ + "movq $0,0x20(%rsp)\n\t" /* fpu_stack */ "call " __ASM_NAME("ndr64_async_client_call") "\n\t" "addq $0x28,%rsp\n\t" __ASM_CFI(".cfi_adjust_cfa_offset -0x28\n\t") - "ret" ); - -#elif defined(_WIN64) - -/*********************************************************************** - * Ndr64AsyncClientCall [RPCRT4.@] - */ -CLIENT_CALL_RETURN WINAPIV Ndr64AsyncClientCall( MIDL_STUBLESS_PROXY_INFO *info, ULONG proc, void *retval, ... ) -{ - va_list args; - LONG_PTR ret; - - va_start( args, retval ); - ret = ndr64_async_client_call( info, proc, retval, va_arg( args, void ** ), NULL ); - va_end( args ); - return *(CLIENT_CALL_RETURN *)&ret; -} - + "ret" ) #endif

1 0

Giovanni Mascellani : vkd3d-shader/ir: Swap selection branches if the if branch is empty.
by Alexandre Julliard 16 Apr '24

16 Apr '24

Module: vkd3d Branch: master Commit: d6ba8b88faeb48ab74a83350f253c89b5eaa9d6a URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/d6ba8b88faeb48ab74a83350f253c… Author: Giovanni Mascellani <gmascellani(a)codeweavers.com> Date: Fri Mar 15 13:06:27 2024 +0100 vkd3d-shader/ir: Swap selection branches if the if branch is empty. --- libs/vkd3d-shader/ir.c | 11 +++++++++++ 1 file changed, 11 insertions(+) diff --git a/libs/vkd3d-shader/ir.c b/libs/vkd3d-shader/ir.c index d972e8e1..5ff54879 100644 --- a/libs/vkd3d-shader/ir.c +++ b/libs/vkd3d-shader/ir.c @@ -4268,6 +4268,17 @@ static enum vkd3d_result vsir_cfg_move_breaks_out_of_selections(struct vsir_cfg --cfg->loop_intervals[else_target].target_count; } + /* If a branch becomes empty, make it the else branch, so we save a block. */ + if (selection->u.selection.if_body.count == 0) + { + struct vsir_cfg_structure_list tmp; + + selection->u.selection.invert_condition = !selection->u.selection.invert_condition; + tmp = selection->u.selection.if_body; + selection->u.selection.if_body = selection->u.selection.else_body; + selection->u.selection.else_body = tmp; + } + return VKD3D_OK; }

1 0

Giovanni Mascellani : vkd3d-shader/ir: Only emit trampolines when needed.
by Alexandre Julliard 16 Apr '24

16 Apr '24

Module: vkd3d Branch: master Commit: df4731dc17e3feb6c2c9b9561ecaae45e0ed51cf URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/df4731dc17e3feb6c2c9b9561ecaa… Author: Giovanni Mascellani <gmascellani(a)codeweavers.com> Date: Mon Mar 18 12:38:51 2024 +0100 vkd3d-shader/ir: Only emit trampolines when needed. --- libs/vkd3d-shader/ir.c | 54 ++++++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 52 insertions(+), 2 deletions(-) diff --git a/libs/vkd3d-shader/ir.c b/libs/vkd3d-shader/ir.c index 5ff54879..bd26ab63 100644 --- a/libs/vkd3d-shader/ir.c +++ b/libs/vkd3d-shader/ir.c @@ -2986,6 +2986,8 @@ struct vsir_cfg_structure { struct vsir_cfg_structure_list body; unsigned idx; + bool needs_trampoline; + struct vsir_cfg_structure *outer_loop; } loop; struct vsir_cfg_structure_selection { @@ -3257,7 +3259,8 @@ static void vsir_cfg_structure_dump(struct vsir_cfg *cfg, struct vsir_cfg_struct vsir_cfg_structure_list_dump(cfg, &structure->u.loop.body); - TRACE("%s} # %u\n", cfg->debug_buffer.buffer, structure->u.loop.idx); + TRACE("%s} # %u%s\n", cfg->debug_buffer.buffer, structure->u.loop.idx, + structure->u.loop.needs_trampoline ? ", tramp" : ""); break; case STRUCTURE_TYPE_SELECTION: @@ -4503,6 +4506,51 @@ static void vsir_cfg_count_targets(struct vsir_cfg *cfg, struct vsir_cfg_structu } } +/* Trampolines are code gadgets used to emulate multilevel jumps (which are not natively supported + * by SPIR-V). A trampoline is inserted just after a loop and checks whether control has reached the + * intended site (i.e., we just jumped out of the target block) or if other levels of jumping are + * needed. For each jump a trampoline is required for all the loops between the jump itself and the + * target loop, excluding the target loop itself. */ +static void vsir_cfg_mark_trampolines(struct vsir_cfg *cfg, struct vsir_cfg_structure_list *list, + struct vsir_cfg_structure *loop) +{ + size_t i; + + for (i = 0; i < list->count; ++i) + { + struct vsir_cfg_structure *structure = &list->structures[i]; + + switch (structure->type) + { + case STRUCTURE_TYPE_BLOCK: + break; + + case STRUCTURE_TYPE_LOOP: + structure->u.loop.outer_loop = loop; + vsir_cfg_mark_trampolines(cfg, &structure->u.loop.body, structure); + break; + + case STRUCTURE_TYPE_SELECTION: + vsir_cfg_mark_trampolines(cfg, &structure->u.selection.if_body, loop); + vsir_cfg_mark_trampolines(cfg, &structure->u.selection.else_body, loop); + break; + + case STRUCTURE_TYPE_JUMP: + { + struct vsir_cfg_structure *l; + if (structure->u.jump.type != JUMP_BREAK && structure->u.jump.type != JUMP_CONTINUE) + break; + for (l = loop; l && l->u.loop.idx != structure->u.jump.target; l = l->u.loop.outer_loop) + { + assert(l->type == STRUCTURE_TYPE_LOOP); + l->u.loop.needs_trampoline = true; + } + break; + } + } + } +} + static enum vkd3d_result vsir_cfg_optimize(struct vsir_cfg *cfg) { enum vkd3d_result ret; @@ -4511,6 +4559,8 @@ static enum vkd3d_result vsir_cfg_optimize(struct vsir_cfg *cfg) ret = vsir_cfg_optimize_recurse(cfg, &cfg->structured_program); + vsir_cfg_mark_trampolines(cfg, &cfg->structured_program, NULL); + if (TRACE_ON()) vsir_cfg_dump_structured_program(cfg); @@ -4559,7 +4609,7 @@ static enum vkd3d_result vsir_cfg_structure_list_emit_loop(struct vsir_cfg *cfg, /* Add a trampoline to implement multilevel jumping depending on the stored * jump_target value. */ - if (loop_idx != UINT_MAX) + if (loop->needs_trampoline) { /* If the multilevel jump is a `continue' and the target is the loop we're inside * right now, then we can finally do the `continue'. */

1 0

Giovanni Mascellani : vkd3d-shader/ir: Only emit launchers when needed.
by Alexandre Julliard 16 Apr '24

16 Apr '24

Module: vkd3d Branch: master Commit: a359c42795e968b5fbb9c80d23416e18f285193a URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/a359c42795e968b5fbb9c80d23416… Author: Giovanni Mascellani <gmascellani(a)codeweavers.com> Date: Mon Mar 18 12:50:02 2024 +0100 vkd3d-shader/ir: Only emit launchers when needed. --- libs/vkd3d-shader/ir.c | 53 +++++++++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 50 insertions(+), 3 deletions(-) diff --git a/libs/vkd3d-shader/ir.c b/libs/vkd3d-shader/ir.c index bd26ab63..07be84e6 100644 --- a/libs/vkd3d-shader/ir.c +++ b/libs/vkd3d-shader/ir.c @@ -3010,6 +3010,7 @@ struct vsir_cfg_structure unsigned int target; struct vkd3d_shader_src_param *condition; bool invert_condition; + bool needs_launcher; } jump; } u; }; @@ -3304,8 +3305,9 @@ static void vsir_cfg_structure_dump(struct vsir_cfg *cfg, struct vsir_cfg_struct vkd3d_unreachable(); } - TRACE("%s%s%s %u\n", cfg->debug_buffer.buffer, type_str, - structure->u.jump.condition ? "c" : "", structure->u.jump.target); + TRACE("%s%s%s %u%s\n", cfg->debug_buffer.buffer, type_str, + structure->u.jump.condition ? "c" : "", structure->u.jump.target, + structure->u.jump.needs_launcher ? " # launch" : ""); break; } @@ -4551,6 +4553,45 @@ static void vsir_cfg_mark_trampolines(struct vsir_cfg *cfg, struct vsir_cfg_stru } } +/* Launchers are the counterpart of trampolines. A launcher is inserted just before a jump, and + * writes in a well-known variable what is the target of the jump. Trampolines will then read that + * variable to decide how to redirect the jump to its intended target. A launcher is needed each + * time the innermost loop containing the jump itself has a trampoline (independently of whether the + * jump is targeting that loop or not). */ +static void vsir_cfg_mark_launchers(struct vsir_cfg *cfg, struct vsir_cfg_structure_list *list, + struct vsir_cfg_structure *loop) +{ + size_t i; + + for (i = 0; i < list->count; ++i) + { + struct vsir_cfg_structure *structure = &list->structures[i]; + + switch (structure->type) + { + case STRUCTURE_TYPE_BLOCK: + break; + + case STRUCTURE_TYPE_LOOP: + vsir_cfg_mark_launchers(cfg, &structure->u.loop.body, structure); + break; + + case STRUCTURE_TYPE_SELECTION: + vsir_cfg_mark_launchers(cfg, &structure->u.selection.if_body, loop); + vsir_cfg_mark_launchers(cfg, &structure->u.selection.else_body, loop); + break; + + case STRUCTURE_TYPE_JUMP: + if (structure->u.jump.type != JUMP_BREAK && structure->u.jump.type != JUMP_CONTINUE) + break; + assert(loop && loop->type == STRUCTURE_TYPE_LOOP); + if (loop->u.loop.needs_trampoline) + structure->u.jump.needs_launcher = true; + break; + } + } +} + static enum vkd3d_result vsir_cfg_optimize(struct vsir_cfg *cfg) { enum vkd3d_result ret; @@ -4559,7 +4600,13 @@ static enum vkd3d_result vsir_cfg_optimize(struct vsir_cfg *cfg) ret = vsir_cfg_optimize_recurse(cfg, &cfg->structured_program); + /* Trampolines and launchers cannot be marked with the same pass, + * because a jump might have to be marked as launcher even when it + * targets its innermost loop, if other jumps in the same loop + * need a trampoline anyway. So launchers can be discovered only + * once all the trampolines are known. */ vsir_cfg_mark_trampolines(cfg, &cfg->structured_program, NULL); + vsir_cfg_mark_launchers(cfg, &cfg->structured_program, NULL); if (TRACE_ON()) vsir_cfg_dump_structured_program(cfg); @@ -4746,7 +4793,7 @@ static enum vkd3d_result vsir_cfg_structure_list_emit_jump(struct vsir_cfg *cfg, if (!reserve_instructions(&target->instructions, &target->ins_capacity, target->ins_count + 2)) return VKD3D_ERROR_OUT_OF_MEMORY; - if (opcode == VKD3DSIH_BREAK || opcode == VKD3DSIH_BREAKP) + if (jump->needs_launcher) { if (!vsir_instruction_init_with_params(cfg->program, &target->instructions[target->ins_count], &no_loc, VKD3DSIH_MOV, 1, 1))

1 0