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

Zebediah Figura : tests: Use a UAV barrier between ClearUnorderedAccessViewFloat() and a compute shader.
by Alexandre Julliard 13 Mar '24

13 Mar '24

Module: vkd3d Branch: master Commit: a686fa7750b44462c18c62757aa9b897f09d5056 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/a686fa7750b44462c18c62757aa9b… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Tue Mar 12 19:13:18 2024 -0500 tests: Use a UAV barrier between ClearUnorderedAccessViewFloat() and a compute shader. In an attempt to fix test failures on radv. --- tests/d3d12.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/tests/d3d12.c b/tests/d3d12.c index e3771b7f..15aed9f3 100644 --- a/tests/d3d12.c +++ b/tests/d3d12.c @@ -23171,6 +23171,7 @@ static void test_cs_uav_store(void) ID3D12GraphicsCommandList_ClearUnorderedAccessViewFloat(command_list, gpu_descriptor_handle, cpu_descriptor_handle, resource, zero, 0, NULL); + uav_barrier(command_list, resource); ID3D12GraphicsCommandList_SetPipelineState(command_list, pipeline_state); ID3D12GraphicsCommandList_SetComputeRootSignature(command_list, root_signature); @@ -23200,6 +23201,7 @@ static void test_cs_uav_store(void) ID3D12GraphicsCommandList_ClearUnorderedAccessViewFloat(command_list, gpu_descriptor_handle, cpu_descriptor_handle, resource, zero, 0, NULL); + uav_barrier(command_list, resource); ID3D12GraphicsCommandList_SetPipelineState(command_list, pipeline_state); ID3D12GraphicsCommandList_SetComputeRootSignature(command_list, root_signature); ID3D12GraphicsCommandList_SetComputeRootConstantBufferView(command_list, 1,

1 0

Alexandre Julliard : ntdll: Implement RtlUnwindEx on ARM64EC.
by Alexandre Julliard 13 Mar '24

13 Mar '24

Module: wine Branch: master Commit: 83949ee2693076f8fc058c2515ffef5e86825091 URL: https://gitlab.winehq.org/wine/wine/-/commit/83949ee2693076f8fc058c2515ffef… Author: Alexandre Julliard <julliard(a)winehq.org> Date: Wed Mar 13 09:55:32 2024 +0100 ntdll: Implement RtlUnwindEx on ARM64EC. --- dlls/ntdll/signal_arm64ec.c | 190 +++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 189 insertions(+), 1 deletion(-) diff --git a/dlls/ntdll/signal_arm64ec.c b/dlls/ntdll/signal_arm64ec.c index 643ec3331de..1e62f7a3118 100644 --- a/dlls/ntdll/signal_arm64ec.c +++ b/dlls/ntdll/signal_arm64ec.c @@ -1542,6 +1542,59 @@ static NTSTATUS virtual_unwind( ULONG type, DISPATCHER_CONTEXT_ARM64EC *dispatch } +/********************************************************************** + * unwind_exception_handler + * + * Handler for exceptions happening while calling an unwind handler. + */ +EXCEPTION_DISPOSITION WINAPI unwind_exception_handler( EXCEPTION_RECORD *record, void *frame, + CONTEXT *context, DISPATCHER_CONTEXT_ARM64EC *dispatch ) +{ + DISPATCHER_CONTEXT_ARM64EC *orig_dispatch = ((DISPATCHER_CONTEXT_ARM64EC **)frame)[-2]; + + /* copy the original dispatcher into the current one, except for the TargetPc */ + dispatch->ControlPc = orig_dispatch->ControlPc; + dispatch->ImageBase = orig_dispatch->ImageBase; + dispatch->FunctionEntry = orig_dispatch->FunctionEntry; + dispatch->EstablisherFrame = orig_dispatch->EstablisherFrame; + dispatch->LanguageHandler = orig_dispatch->LanguageHandler; + dispatch->HandlerData = orig_dispatch->HandlerData; + dispatch->HistoryTable = orig_dispatch->HistoryTable; + dispatch->ScopeIndex = orig_dispatch->ScopeIndex; + dispatch->ControlPcIsUnwound = orig_dispatch->ControlPcIsUnwound; + *dispatch->ContextRecord = *orig_dispatch->ContextRecord; + memcpy( dispatch->NonVolatileRegisters, orig_dispatch->NonVolatileRegisters, + sizeof(DISPATCHER_CONTEXT_NONVOLREG_ARM64) ); + TRACE( "detected collided unwind\n" ); + return ExceptionCollidedUnwind; +} + + +/********************************************************************** + * call_unwind_handler + */ +static DWORD __attribute__((naked)) call_unwind_handler( EXCEPTION_RECORD *rec, ULONG_PTR frame, + CONTEXT *context, void *dispatch, + PEXCEPTION_ROUTINE handler ) +{ + asm( ".seh_proc call_unwind_handler\n\t" + "stp x29, x30, [sp, #-32]!\n\t" + ".seh_save_fplr_x 32\n\t" + ".seh_endprologue\n\t" + ".seh_handler unwind_exception_handler, @except\n\t" + "str x3, [sp, #16]\n\t" /* frame[-2] = dispatch */ + "mov x11, x4\n\t" /* handler */ + "adr x10, $iexit_thunk$cdecl$i8$i8i8i8i8\n\t" + "adrp x16, __os_arm64x_dispatch_icall\n\t" + "ldr x16, [x16, #:lo12:__os_arm64x_dispatch_icall]\n\t" + "blr x16\n\t" + "blr x11\n\t" + "ldp x29, x30, [sp], #32\n\t" + "ret\n\t" + ".seh_endproc" ); +} + + /*********************************************************************** * call_seh_handler */ @@ -1991,7 +2044,142 @@ void CDECL RtlRestoreContext( CONTEXT *context, EXCEPTION_RECORD *rec ) void WINAPI RtlUnwindEx( PVOID end_frame, PVOID target_ip, EXCEPTION_RECORD *rec, PVOID retval, CONTEXT *context, UNWIND_HISTORY_TABLE *table ) { - FIXME( "not implemented\n" ); + EXCEPTION_REGISTRATION_RECORD *teb_frame = NtCurrentTeb()->Tib.ExceptionList; + DISPATCHER_CONTEXT_NONVOLREG_ARM64 nonvol_regs; + EXCEPTION_RECORD record; + DISPATCHER_CONTEXT_ARM64EC dispatch; + ARM64EC_NT_CONTEXT new_context; + NTSTATUS status; + ULONG_PTR frame; + DWORD i, res; + BOOL is_target_ec = RtlIsEcCode( (ULONG_PTR)target_ip ); + + RtlCaptureContext( context ); + new_context.AMD64_Context = *context; + + /* build an exception record, if we do not have one */ + if (!rec) + { + record.ExceptionCode = STATUS_UNWIND; + record.ExceptionFlags = 0; + record.ExceptionRecord = NULL; + record.ExceptionAddress = (void *)context->Rip; + record.NumberParameters = 0; + rec = &record; + } + + rec->ExceptionFlags |= EXCEPTION_UNWINDING | (end_frame ? 0 : EXCEPTION_EXIT_UNWIND); + + TRACE( "code=%lx flags=%lx end_frame=%p target_ip=%p\n", + rec->ExceptionCode, rec->ExceptionFlags, end_frame, target_ip ); + for (i = 0; i < min( EXCEPTION_MAXIMUM_PARAMETERS, rec->NumberParameters ); i++) + TRACE( " info[%ld]=%016I64x\n", i, rec->ExceptionInformation[i] ); + TRACE_CONTEXT( context ); + + dispatch.TargetIp = (ULONG64)target_ip; + dispatch.ContextRecord = context; + dispatch.HistoryTable = table; + dispatch.NonVolatileRegisters = nonvol_regs.Buffer; + + for (;;) + { + status = virtual_unwind( UNW_FLAG_UHANDLER, &dispatch, &new_context ); + if (status != STATUS_SUCCESS) raise_status( status, rec ); + + unwind_done: + if (!dispatch.EstablisherFrame) break; + + if (!is_valid_arm64ec_frame( dispatch.EstablisherFrame )) + { + ERR( "invalid frame %I64x (%p-%p)\n", dispatch.EstablisherFrame, + NtCurrentTeb()->Tib.StackLimit, NtCurrentTeb()->Tib.StackBase ); + rec->ExceptionFlags |= EXCEPTION_STACK_INVALID; + break; + } + + if (dispatch.LanguageHandler) + { + if (end_frame && (dispatch.EstablisherFrame > (ULONG64)end_frame)) + { + ERR( "invalid end frame %I64x/%p\n", dispatch.EstablisherFrame, end_frame ); + raise_status( STATUS_INVALID_UNWIND_TARGET, rec ); + } + if (dispatch.EstablisherFrame == (ULONG64)end_frame) rec->ExceptionFlags |= EXCEPTION_TARGET_UNWIND; + + TRACE( "calling handler %p (rec=%p, frame=%I64x context=%p, dispatch=%p)\n", + dispatch.LanguageHandler, rec, dispatch.EstablisherFrame, + dispatch.ContextRecord, &dispatch ); + res = call_unwind_handler( rec, dispatch.EstablisherFrame, dispatch.ContextRecord, + &dispatch, dispatch.LanguageHandler ); + TRACE( "handler %p returned %lx\n", dispatch.LanguageHandler, res ); + + switch (res) + { + case ExceptionContinueSearch: + rec->ExceptionFlags &= ~EXCEPTION_COLLIDED_UNWIND; + break; + case ExceptionCollidedUnwind: + new_context.AMD64_Context = *context; + RtlVirtualUnwind( UNW_FLAG_NHANDLER, dispatch.ImageBase, dispatch.ControlPc, + dispatch.FunctionEntry, &new_context.AMD64_Context, + &dispatch.HandlerData, &frame, NULL ); + rec->ExceptionFlags |= EXCEPTION_COLLIDED_UNWIND; + goto unwind_done; + default: + raise_status( STATUS_INVALID_DISPOSITION, rec ); + break; + } + } + else /* hack: call builtin handlers registered in the tib list */ + { + while (is_valid_arm64ec_frame( (ULONG_PTR)teb_frame ) && + (ULONG64)teb_frame < new_context.Sp && + (ULONG64)teb_frame < (ULONG64)end_frame) + { + TRACE( "calling TEB handler %p (rec=%p, frame=%p context=%p, dispatch=%p)\n", + teb_frame->Handler, rec, teb_frame, dispatch.ContextRecord, &dispatch ); + res = call_unwind_handler( rec, (ULONG_PTR)teb_frame, dispatch.ContextRecord, &dispatch, + (PEXCEPTION_ROUTINE)teb_frame->Handler ); + TRACE( "handler at %p returned %lu\n", teb_frame->Handler, res ); + teb_frame = __wine_pop_frame( teb_frame ); + + switch (res) + { + case ExceptionContinueSearch: + rec->ExceptionFlags &= ~EXCEPTION_COLLIDED_UNWIND; + break; + case ExceptionCollidedUnwind: + new_context.AMD64_Context = *context; + RtlVirtualUnwind( UNW_FLAG_NHANDLER, dispatch.ImageBase, dispatch.ControlPc, + dispatch.FunctionEntry, &new_context.AMD64_Context, + &dispatch.HandlerData, &frame, NULL ); + rec->ExceptionFlags |= EXCEPTION_COLLIDED_UNWIND; + goto unwind_done; + default: + raise_status( STATUS_INVALID_DISPOSITION, rec ); + break; + } + } + if ((ULONG64)teb_frame == (ULONG64)end_frame && (ULONG64)end_frame < new_context.Sp) break; + } + + if (dispatch.EstablisherFrame == (ULONG64)end_frame) + { + if (is_target_ec) break; + if (!RtlIsEcCode( dispatch.ControlPc )) break; + /* we just crossed into x64 code, unwind one more frame */ + } + *context = new_context.AMD64_Context; + } + + if (rec->ExceptionCode != STATUS_UNWIND_CONSOLIDATE) + context->Rip = (ULONG64)target_ip; + else if (rec->ExceptionInformation[10] == -1) + rec->ExceptionInformation[10] = (ULONG_PTR)&nonvol_regs; + + if (is_target_ec) context->Rcx = (ULONG64)retval; + else context->Rax = (ULONG64)retval; + RtlRestoreContext( context, rec ); }

1 0

Alexandre Julliard : ntdll: Move __C_specific_handler implementation to unwind.c.
by Alexandre Julliard 13 Mar '24

13 Mar '24

Module: wine Branch: master Commit: 7b5267cf63f225519ef4ef53c111402adc325116 URL: https://gitlab.winehq.org/wine/wine/-/commit/7b5267cf63f225519ef4ef53c11140… Author: Alexandre Julliard <julliard(a)winehq.org> Date: Wed Mar 13 10:44:21 2024 +0100 ntdll: Move __C_specific_handler implementation to unwind.c. --- dlls/ntdll/signal_arm.c | 107 +-------------------- dlls/ntdll/signal_arm64.c | 146 +++------------------------- dlls/ntdll/signal_arm64ec.c | 13 --- dlls/ntdll/signal_x86_64.c | 90 ----------------- dlls/ntdll/unwind.c | 229 ++++++++++++++++++++++++++++++++++++++++++++ 5 files changed, 245 insertions(+), 340 deletions(-)

1 0

Alexandre Julliard : ntdll: Implement __C_specific_handler on ARM64EC.
by Alexandre Julliard 13 Mar '24

13 Mar '24

Module: wine Branch: master Commit: c47d6ef1f782fe305851ac339c8424162ca7c559 URL: https://gitlab.winehq.org/wine/wine/-/commit/c47d6ef1f782fe305851ac339c8424… Author: Alexandre Julliard <julliard(a)winehq.org> Date: Wed Mar 13 11:03:01 2024 +0100 ntdll: Implement __C_specific_handler on ARM64EC. --- dlls/ntdll/signal_arm64ec.c | 34 ++++++++++++++++++++++++++++++++++ dlls/ntdll/unwind.c | 6 ++++++ 2 files changed, 40 insertions(+) diff --git a/dlls/ntdll/signal_arm64ec.c b/dlls/ntdll/signal_arm64ec.c index 1007df91f85..55ed12a609a 100644 --- a/dlls/ntdll/signal_arm64ec.c +++ b/dlls/ntdll/signal_arm64ec.c @@ -2333,6 +2333,40 @@ void __attribute__((naked)) __chkstk_arm64ec(void) } +/******************************************************************* + * __C_ExecuteExceptionFilter + */ +LONG __attribute__((naked)) __C_ExecuteExceptionFilter( EXCEPTION_POINTERS *ptrs, void *frame, + PEXCEPTION_FILTER filter, BYTE *nonvolatile ) +{ + asm( ".seh_proc _C_ExecuteExceptionFilter\n\t" + "stp x29, x30, [sp, #-80]!\n\t" + ".seh_save_fplr_x 80\n\t" + "stp x19, x20, [sp, #16]\n\t" + ".seh_save_regp x19, 16\n\t" + "stp x21, x22, [sp, #32]\n\t" + ".seh_save_regp x21, 32\n\t" + "stp x25, x26, [sp, #48]\n\t" + ".seh_save_regp x25, 48\n\t" + "str x27, [sp, #64]\n\t" + ".seh_save_reg x27, 64\n\t" + ".seh_endprologue\n\t" + "ldp x19, x20, [x3, #0]\n\t" /* nonvolatile regs */ + "ldp x21, x22, [x3, #16]\n\t" + "ldp x25, x26, [x3, #48]\n\t" + "ldr x27, [x3, #64]\n\t" + "ldr x1, [x3, #80]\n\t" /* x29 = frame */ + "blr x2\n\t" /* filter */ + "ldp x19, x20, [sp, #16]\n\t" + "ldp x21, x22, [sp, #32]\n\t" + "ldp x25, x26, [sp, #48]\n\t" + "ldr x27, [sp, #64]\n\t" + "ldp x29, x30, [sp], #80\n\t" + "ret\n\t" + ".seh_endproc" ); +} + + /*********************************************************************** * RtlRaiseException (NTDLL.@) */ diff --git a/dlls/ntdll/unwind.c b/dlls/ntdll/unwind.c index 3960d13143d..06779247da1 100644 --- a/dlls/ntdll/unwind.c +++ b/dlls/ntdll/unwind.c @@ -2204,6 +2204,12 @@ EXCEPTION_DISPOSITION WINAPI __C_specific_handler( EXCEPTION_RECORD *rec, void * ULONG_PTR pc = dispatch->ControlPc; unsigned int i; +#ifdef __arm64ec__ + if (RtlIsEcCode( pc )) + return __C_specific_handler_arm64( rec, frame, (ARM64_NT_CONTEXT *)context, + (DISPATCHER_CONTEXT_ARM64 *)dispatch ); +#endif + TRACE( "%p %p %p %p pc %Ix\n", rec, frame, context, dispatch, pc ); if (TRACE_ON(unwind)) DUMP_SCOPE_TABLE( base, table );

1 0

Alexandre Julliard : ntdll: Always use SEH support on ARM64.
by Alexandre Julliard 13 Mar '24

13 Mar '24

Module: wine Branch: master Commit: 8c1f96f7abd2307c3fc3e998fe16927cc4aaccb4 URL: https://gitlab.winehq.org/wine/wine/-/commit/8c1f96f7abd2307c3fc3e998fe1692… Author: Alexandre Julliard <julliard(a)winehq.org> Date: Wed Mar 13 11:24:26 2024 +0100 ntdll: Always use SEH support on ARM64. --- dlls/ntdll/relay.c | 6 +++--- dlls/ntdll/signal_arm64.c | 25 +++++++++++-------------- dlls/ntdll/signal_arm64ec.c | 6 +++--- 3 files changed, 17 insertions(+), 20 deletions(-) diff --git a/dlls/ntdll/relay.c b/dlls/ntdll/relay.c index ecb76eaedbd..959dde9a5f5 100644 --- a/dlls/ntdll/relay.c +++ b/dlls/ntdll/relay.c @@ -646,10 +646,10 @@ void WINAPI relay_trace_exit( struct relay_descr *descr, unsigned int idx, extern LONGLONG CDECL call_entry_point( void *func, int nb_args, const INT_PTR *args ); __ASM_GLOBAL_FUNC( call_entry_point, "stp x29, x30, [SP,#-16]!\n\t" - __ASM_SEH(".seh_save_fplr_x 16\n\t") + ".seh_save_fplr_x 16\n\t" "mov x29, SP\n\t" - __ASM_SEH(".seh_set_fp\n\t") - __ASM_SEH(".seh_endprologue\n\t") + ".seh_set_fp\n\t" + ".seh_endprologue\n\t" "ldr x8, [x2, #-32]\n\t" "mov x9, x0\n\t" "cbz w1, 2f\n\t" diff --git a/dlls/ntdll/signal_arm64.c b/dlls/ntdll/signal_arm64.c index 6bd99df3ff1..2568cf1ae22 100644 --- a/dlls/ntdll/signal_arm64.c +++ b/dlls/ntdll/signal_arm64.c @@ -305,8 +305,8 @@ NTSTATUS call_seh_handlers( EXCEPTION_RECORD *rec, CONTEXT *orig_context ) * KiUserExceptionDispatcher (NTDLL.@) */ __ASM_GLOBAL_FUNC( KiUserExceptionDispatcher, - __ASM_SEH(".seh_context\n\t") - __ASM_SEH(".seh_endprologue\n\t") + ".seh_context\n\t" + ".seh_endprologue\n\t" "adr x16, " __ASM_NAME("pWow64PrepareForException") "\n\t" "ldr x16, [x16]\n\t" "cbz x16, 1f\n\t" @@ -323,10 +323,10 @@ __ASM_GLOBAL_FUNC( KiUserExceptionDispatcher, * KiUserApcDispatcher (NTDLL.@) */ __ASM_GLOBAL_FUNC( KiUserApcDispatcher, - __ASM_SEH(".seh_context\n\t") + ".seh_context\n\t" "nop\n\t" - __ASM_SEH(".seh_stackalloc 0x30\n\t") - __ASM_SEH(".seh_endprologue\n\t") + ".seh_stackalloc 0x30\n\t" + ".seh_endprologue\n\t" "ldp x16, x0, [sp]\n\t" /* func, arg1 */ "ldp x1, x2, [sp, #0x10]\n\t" /* arg2, arg3 */ "add x3, sp, #0x30\n\t" /* context (FIXME) */ @@ -660,13 +660,10 @@ __ASM_GLOBAL_FUNC( __C_ExecuteExceptionFilter, */ __ASM_GLOBAL_FUNC( RtlRaiseException, "sub sp, sp, #0x3b0\n\t" /* 0x390 (context) + 0x20 */ + ".seh_stackalloc 0x3b0\n\t" "stp x29, x30, [sp]\n\t" - __ASM_SEH(".seh_stackalloc 0x3b0\n\t") - __ASM_SEH(".seh_save_fplr 0\n\t") - __ASM_SEH(".seh_endprologue\n\t") - __ASM_CFI(".cfi_def_cfa x29, 944\n\t") - __ASM_CFI(".cfi_offset x30, -936\n\t") - __ASM_CFI(".cfi_offset x29, -944\n\t") + ".seh_save_fplr 0\n\t" + ".seh_endprologue\n\t" "mov x29, sp\n\t" "str x0, [sp, #0x10]\n\t" "add x0, sp, #0x20\n\t" @@ -741,15 +738,15 @@ void __cdecl NTDLL_longjmp( _JUMP_BUFFER *buf, int retval ) */ __ASM_GLOBAL_FUNC( RtlUserThreadStart, "stp x29, x30, [sp, #-16]!\n\t" - __ASM_SEH(".seh_save_fplr_x 16\n\t") - __ASM_SEH(".seh_endprologue\n\t") + ".seh_save_fplr_x 16\n\t" + ".seh_endprologue\n\t" "adr x8, " __ASM_NAME("pBaseThreadInitThunk") "\n\t" "ldr x8, [x8]\n\t" "mov x2, x1\n\t" "mov x1, x0\n\t" "mov x0, #0\n\t" "blr x8\n\t" - __ASM_SEH(".seh_handler " __ASM_NAME("call_unhandled_exception_handler") ", @except") ) + ".seh_handler call_unhandled_exception_handler, @except" ) /****************************************************************** * LdrInitializeThunk (NTDLL.@) diff --git a/dlls/ntdll/signal_arm64ec.c b/dlls/ntdll/signal_arm64ec.c index 55ed12a609a..8ec2575c786 100644 --- a/dlls/ntdll/signal_arm64ec.c +++ b/dlls/ntdll/signal_arm64ec.c @@ -1758,10 +1758,10 @@ static void __attribute__((used)) dispatch_apc( void (CALLBACK *func)(ULONG_PTR, NtContinue( &context.AMD64_Context, alertable ); } __ASM_GLOBAL_FUNC( "#KiUserApcDispatcher", - __ASM_SEH(".seh_context\n\t") + ".seh_context\n\t" "nop\n\t" - __ASM_SEH(".seh_stackalloc 0x30\n\t") - __ASM_SEH(".seh_endprologue\n\t") + ".seh_stackalloc 0x30\n\t" + ".seh_endprologue\n\t" "ldp x0, x1, [sp]\n\t" /* func, arg1 */ "ldp x2, x3, [sp, #0x10]\n\t" /* arg2, arg3 */ "ldr w4, [sp, #0x20]\n\t" /* alertable */

1 0

Giovanni Mascellani : vkd3d-shader/ir: Compute loop as intervals of the block order.
by Alexandre Julliard 12 Mar '24

12 Mar '24

Module: vkd3d Branch: master Commit: 1d606123401640fcf0e9c1cb35caf72755327b43 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/1d606123401640fcf0e9c1cb35caf… Author: Giovanni Mascellani <gmascellani(a)codeweavers.com> Date: Sun Feb 4 20:21:20 2024 +0100 vkd3d-shader/ir: Compute loop as intervals of the block order. --- libs/vkd3d-shader/ir.c | 53 ++++++++++++++++++++++++++++++++++++++++++-------- 1 file changed, 45 insertions(+), 8 deletions(-) diff --git a/libs/vkd3d-shader/ir.c b/libs/vkd3d-shader/ir.c index 14aea329..67f5b5fd 100644 --- a/libs/vkd3d-shader/ir.c +++ b/libs/vkd3d-shader/ir.c @@ -3143,6 +3143,16 @@ struct vsir_cfg size_t *loops_by_header; struct vsir_block_list order; + struct cfg_loop_interval + { + /* `begin' is the position of the first block of the loop in + * the topological sort; `end' is the position of the first + * block after the loop. In other words, `begin' is where a + * `continue' instruction would jump and `end' is where a + * `break' instruction would jump. */ + unsigned int begin, end; + } *loop_intervals; + size_t loop_interval_count, loop_interval_capacity; }; static void vsir_cfg_cleanup(struct vsir_cfg *cfg) @@ -3160,11 +3170,29 @@ static void vsir_cfg_cleanup(struct vsir_cfg *cfg) vkd3d_free(cfg->blocks); vkd3d_free(cfg->loops); vkd3d_free(cfg->loops_by_header); + vkd3d_free(cfg->loop_intervals); if (TRACE_ON()) vkd3d_string_buffer_cleanup(&cfg->debug_buffer); } +static enum vkd3d_result vsir_cfg_add_loop_interval(struct vsir_cfg *cfg, unsigned int begin, + unsigned int end) +{ + struct cfg_loop_interval *interval; + + if (!vkd3d_array_reserve((void **)&cfg->loop_intervals, &cfg->loop_interval_capacity, + cfg->loop_interval_count + 1, sizeof(*cfg->loop_intervals))) + return VKD3D_ERROR_OUT_OF_MEMORY; + + interval = &cfg->loop_intervals[cfg->loop_interval_count++]; + + interval->begin = begin; + interval->end = end; + + return VKD3D_OK; +} + static bool vsir_block_dominates(struct vsir_block *b1, struct vsir_block *b2) { return bitmap_is_set(b1->dominates, b2->label - 1); @@ -3399,14 +3427,14 @@ static void vsir_cfg_compute_dominators(struct vsir_cfg *cfg) * block without passing through the header block) belong to the same * loop. * - * If the input CFG is reducible, each two loops are either disjoint - * or one is a strict subset of the other, provided that each block - * has at most one incoming back edge. If this condition does not - * hold, a synthetic block can be introduced as the only back edge - * block for the given header block, with all the previous back edge - * now being forward edges to the synthetic block. This is not - * currently implemented (but it is rarely found in practice - * anyway). */ + * If the input CFG is reducible its loops are properly nested (i.e., + * each two loops are either disjoint or one is contained in the + * other), provided that each block has at most one incoming back + * edge. If this condition does not hold, a synthetic block can be + * introduced as the only back edge block for the given header block, + * with all the previous back edge now being forward edges to the + * synthetic block. This is not currently implemented (but it is + * rarely found in practice anyway). */ static enum vkd3d_result vsir_cfg_scan_loop(struct vsir_block_list *loop, struct vsir_block *block, struct vsir_block *header) { @@ -3499,6 +3527,7 @@ struct vsir_cfg_node_sorter { struct vsir_block_list *loop; unsigned int seen_count; + unsigned int begin; } *stack; size_t stack_count, stack_capacity; struct vsir_block_list available_blocks; @@ -3525,6 +3554,7 @@ static enum vkd3d_result vsir_cfg_node_sorter_make_node_available(struct vsir_cf item = &sorter->stack[sorter->stack_count++]; item->loop = loop; item->seen_count = 0; + item->begin = sorter->cfg->order.count; return VKD3D_OK; } @@ -3649,6 +3679,10 @@ static enum vkd3d_result vsir_cfg_sort_nodes(struct vsir_cfg *cfg) if (inner_stack_item->seen_count != inner_stack_item->loop->count) break; + if ((ret = vsir_cfg_add_loop_interval(cfg, inner_stack_item->begin, + cfg->order.count)) < 0) + goto fail; + new_seen_count = inner_stack_item->loop->count; --sorter.stack_count; } @@ -3697,6 +3731,9 @@ static enum vkd3d_result vsir_cfg_sort_nodes(struct vsir_cfg *cfg) TRACE("%s\n", cfg->debug_buffer.buffer); vkd3d_string_buffer_clear(&cfg->debug_buffer); + + for (i = 0; i < cfg->loop_interval_count; ++i) + TRACE("Loop interval %u - %u\n", cfg->loop_intervals[i].begin, cfg->loop_intervals[i].end); } return VKD3D_OK;

1 0

Giovanni Mascellani : vkd3d-shader/ir: Generate synthetic intervals for forward edges.
by Alexandre Julliard 12 Mar '24

12 Mar '24

Module: vkd3d Branch: master Commit: 070c4af8b3d3d6910344f1786d11709f9b4a1909 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/070c4af8b3d3d6910344f1786d117… Author: Giovanni Mascellani <gmascellani(a)codeweavers.com> Date: Mon Feb 5 11:15:24 2024 +0100 vkd3d-shader/ir: Generate synthetic intervals for forward edges. --- libs/vkd3d-shader/ir.c | 159 +++++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 153 insertions(+), 6 deletions(-) diff --git a/libs/vkd3d-shader/ir.c b/libs/vkd3d-shader/ir.c index 67f5b5fd..88d54e56 100644 --- a/libs/vkd3d-shader/ir.c +++ b/libs/vkd3d-shader/ir.c @@ -3068,7 +3068,7 @@ static void vsir_block_list_remove_index(struct vsir_block_list *list, size_t id struct vsir_block { - unsigned int label; + unsigned int label, order_pos; /* `begin' points to the instruction immediately following the * LABEL that introduces the block. `end' points to the terminator * instruction (either BRANCH or RET). They can coincide, meaning @@ -3151,6 +3151,27 @@ struct vsir_cfg * `continue' instruction would jump and `end' is where a * `break' instruction would jump. */ unsigned int begin, end; + /* Each loop interval can be natural or synthetic. Natural + * intervals are added to represent loops given by CFG back + * edges. Synthetic intervals do not correspond to loops in + * the input CFG, but are added to leverage their `break' + * instruction in order to execute forward edges. + * + * For a synthetic loop interval it's not really important + * which one is the `begin' block, since we don't need to + * execute `continue' for them. So we have some leeway for + * moving it provided that these conditions are met: 1. the + * interval must contain all `break' instructions that target + * it, which in practice means that `begin' can be moved + * backward and not forward; 2. intervals must remain properly + * nested (for each pair of intervals, either one contains the + * other or they are disjoint). + * + * Subject to these conditions, we try to reuse the same loop + * as much as possible (if many forward edges target the same + * block), but we still try to keep `begin' as forward as + * possible, to keep the loop scope as small as possible. */ + bool synthetic; } *loop_intervals; size_t loop_interval_count, loop_interval_capacity; }; @@ -3177,7 +3198,7 @@ static void vsir_cfg_cleanup(struct vsir_cfg *cfg) } static enum vkd3d_result vsir_cfg_add_loop_interval(struct vsir_cfg *cfg, unsigned int begin, - unsigned int end) + unsigned int end, bool synthetic) { struct cfg_loop_interval *interval; @@ -3189,6 +3210,7 @@ static enum vkd3d_result vsir_cfg_add_loop_interval(struct vsir_cfg *cfg, unsign interval->begin = begin; interval->end = end; + interval->synthetic = synthetic; return VKD3D_OK; } @@ -3661,6 +3683,7 @@ static enum vkd3d_result vsir_cfg_sort_nodes(struct vsir_cfg *cfg) } vsir_block_list_remove_index(&sorter.available_blocks, i); + block->order_pos = cfg->order.count; if ((ret = vsir_block_list_add_checked(&cfg->order, block)) < 0) goto fail; @@ -3680,7 +3703,7 @@ static enum vkd3d_result vsir_cfg_sort_nodes(struct vsir_cfg *cfg) break; if ((ret = vsir_cfg_add_loop_interval(cfg, inner_stack_item->begin, - cfg->order.count)) < 0) + cfg->order.count, false)) < 0) goto fail; new_seen_count = inner_stack_item->loop->count; @@ -3731,9 +3754,6 @@ static enum vkd3d_result vsir_cfg_sort_nodes(struct vsir_cfg *cfg) TRACE("%s\n", cfg->debug_buffer.buffer); vkd3d_string_buffer_clear(&cfg->debug_buffer); - - for (i = 0; i < cfg->loop_interval_count; ++i) - TRACE("Loop interval %u - %u\n", cfg->loop_intervals[i].begin, cfg->loop_intervals[i].end); } return VKD3D_OK; @@ -3746,6 +3766,127 @@ fail: return ret; } +static enum vkd3d_result vsir_cfg_generate_synthetic_loop_intervals(struct vsir_cfg *cfg) +{ + enum vkd3d_result ret; + size_t i, j, k; + + for (i = 0; i < cfg->block_count; ++i) + { + struct vsir_block *block = &cfg->blocks[i]; + + if (block->label == 0) + continue; + + for (j = 0; j < block->successors.count; ++j) + { + struct vsir_block *successor = block->successors.blocks[j]; + struct cfg_loop_interval *extend = NULL; + unsigned int begin; + enum + { + ACTION_DO_NOTHING, + ACTION_CREATE_NEW, + ACTION_EXTEND, + } action = ACTION_CREATE_NEW; + + /* We've already contructed loop intervals for the back + * edges, there's nothing more to do. */ + if (vsir_block_dominates(successor, block)) + continue; + + assert(block->order_pos < successor->order_pos); + + /* Jumping from a block to the following one is always + * possible, so nothing to do. */ + if (block->order_pos + 1 == successor->order_pos) + continue; + + /* Let's look for a loop interval that already breaks at + * `successor' and either contains or can be extended to + * contain `block'. */ + for (k = 0; k < cfg->loop_interval_count; ++k) + { + struct cfg_loop_interval *interval = &cfg->loop_intervals[k]; + + if (interval->end != successor->order_pos) + continue; + + if (interval->begin <= block->order_pos) + { + action = ACTION_DO_NOTHING; + break; + } + + if (interval->synthetic) + { + action = ACTION_EXTEND; + extend = interval; + break; + } + } + + if (action == ACTION_DO_NOTHING) + continue; + + /* Ok, we have to decide where the new or replacing + * interval has to begin. These are the rules: 1. it must + * begin before `block'; 2. intervals must be properly + * nested; 3. the new interval should begin as late as + * possible, to limit control flow depth and extension. */ + begin = block->order_pos; + + /* Our candidate interval is always [begin, + * successor->order_pos), and we move `begin' backward + * until the candidate interval contains all the intervals + * whose endpoint lies in the candidate interval + * itself. */ + for (k = 0; k < cfg->loop_interval_count; ++k) + { + struct cfg_loop_interval *interval = &cfg->loop_intervals[k]; + + if (begin < interval->end && interval->end < successor->order_pos) + begin = min(begin, interval->begin); + } + + /* New we have to care about the intervals whose begin + * point lies in the candidate interval. We cannot move + * the candidate interval endpoint, because it is + * important that the loop break target matches + * `successor'. So we have to move that interval's begin + * point to the begin point of the candidate interval, + * i.e. `begin'. But what if the interval we should extend + * backward is not synthetic? This cannot happen, + * fortunately, because it would mean that there is a jump + * entering a loop via a block which is not the loop + * header, so the CFG would not be reducible. */ + for (k = 0; k < cfg->loop_interval_count; ++k) + { + struct cfg_loop_interval *interval = &cfg->loop_intervals[k]; + + if (interval->begin < successor->order_pos && successor->order_pos < interval->end) + { + if (interval->synthetic) + interval->begin = min(begin, interval->begin); + assert(begin >= interval->begin); + } + } + + if (action == ACTION_EXTEND) + extend->begin = begin; + else if ((ret = vsir_cfg_add_loop_interval(cfg, begin, successor->order_pos, true)) < 0) + return ret; + } + } + + if (TRACE_ON()) + for (i = 0; i < cfg->loop_interval_count; ++i) + TRACE("%s loop interval %u - %u\n", cfg->loop_intervals[i].synthetic ? "Synthetic" : "Natural", + cfg->loop_intervals[i].begin, cfg->loop_intervals[i].end); + + return VKD3D_OK; +} + enum vkd3d_result vkd3d_shader_normalise(struct vkd3d_shader_parser *parser, const struct vkd3d_shader_compile_info *compile_info) { @@ -3784,6 +3925,12 @@ enum vkd3d_result vkd3d_shader_normalise(struct vkd3d_shader_parser *parser, return result; } + if ((result = vsir_cfg_generate_synthetic_loop_intervals(&cfg)) < 0) + { + vsir_cfg_cleanup(&cfg); + return result; + } + if ((result = simple_structurizer_run(parser)) < 0) { vsir_cfg_cleanup(&cfg);

1 0

Giovanni Mascellani : vkd3d-shader/ir: Sort loop intervals.
by Alexandre Julliard 12 Mar '24

12 Mar '24

Module: vkd3d Branch: master Commit: f2b2040ff6e61c30160948f1ec5a160a7128981b URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/f2b2040ff6e61c30160948f1ec5a1… Author: Giovanni Mascellani <gmascellani(a)codeweavers.com> Date: Tue Feb 6 14:02:59 2024 +0100 vkd3d-shader/ir: Sort loop intervals. --- libs/vkd3d-shader/ir.c | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) diff --git a/libs/vkd3d-shader/ir.c b/libs/vkd3d-shader/ir.c index 88d54e56..759c8995 100644 --- a/libs/vkd3d-shader/ir.c +++ b/libs/vkd3d-shader/ir.c @@ -3766,6 +3766,20 @@ fail: return ret; } +/* Sort loop intervals first by ascending begin time and then by + * descending end time, so that inner intervals appear after outer + * ones and disjoint intervals appear in their proper order. */ +static int compare_loop_intervals(const void *ptr1, const void *ptr2) +{ + const struct cfg_loop_interval *interval1 = ptr1; + const struct cfg_loop_interval *interval2 = ptr2; + + if (interval1->begin != interval2->begin) + return vkd3d_u32_compare(interval1->begin, interval2->begin); + + return -vkd3d_u32_compare(interval1->end, interval2->end); +} + static enum vkd3d_result vsir_cfg_generate_synthetic_loop_intervals(struct vsir_cfg *cfg) { enum vkd3d_result ret; @@ -3879,6 +3893,8 @@ static enum vkd3d_result vsir_cfg_generate_synthetic_loop_intervals(struct vsir_ } } + qsort(cfg->loop_intervals, cfg->loop_interval_count, sizeof(*cfg->loop_intervals), compare_loop_intervals); + if (TRACE_ON()) for (i = 0; i < cfg->loop_interval_count; ++i) TRACE("%s loop interval %u - %u\n", cfg->loop_intervals[i].synthetic ? "Synthetic" : "Natural",

1 0

Nikolay Sivov : vkd3d-shader: Remove a token for 'precise' modifier.
by Alexandre Julliard 12 Mar '24

12 Mar '24

Module: vkd3d Branch: master Commit: cf7c05f4315dd21a96f60653497c1228a38f61af URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/cf7c05f4315dd21a96f60653497c1… Author: Nikolay Sivov <nsivov(a)codeweavers.com> Date: Sun Feb 25 02:42:54 2024 +0100 vkd3d-shader: Remove a token for 'precise' modifier. Signed-off-by: Nikolay Sivov <nsivov(a)codeweavers.com> --- Makefile.am | 1 + libs/vkd3d-shader/hlsl.l | 1 - libs/vkd3d-shader/hlsl.y | 14 ++++++----- libs/vkd3d-shader/vkd3d_shader_private.h | 1 + tests/hlsl/precise-modifier.shader_test | 41 ++++++++++++++++++++++++++++++++ 5 files changed, 51 insertions(+), 7 deletions(-) diff --git a/Makefile.am b/Makefile.am index d07218d8..52e154fe 100644 --- a/Makefile.am +++ b/Makefile.am @@ -155,6 +155,7 @@ vkd3d_shader_tests = \ tests/hlsl/object-parameters.shader_test \ tests/hlsl/object-references.shader_test \ tests/hlsl/pow.shader_test \ + tests/hlsl/precise-modifier.shader_test \ tests/hlsl/rasteriser-ordered-views.shader_test \ tests/hlsl/reflect.shader_test \ tests/hlsl/register-reservations-numeric.shader_test \ diff --git a/libs/vkd3d-shader/hlsl.l b/libs/vkd3d-shader/hlsl.l index 8dcceb94..600e2cf2 100644 --- a/libs/vkd3d-shader/hlsl.l +++ b/libs/vkd3d-shader/hlsl.l @@ -109,7 +109,6 @@ packoffset {return KW_PACKOFFSET; } pass {return KW_PASS; } PixelShader {return KW_PIXELSHADER; } pixelshader {return KW_PIXELSHADER; } -precise {return KW_PRECISE; } RasterizerOrderedBuffer {return KW_RASTERIZERORDEREDBUFFER; } RasterizerOrderedStructuredBuffer {return KW_RASTERIZERORDEREDSTRUCTUREDBUFFER; } RasterizerOrderedTexture1D {return KW_RASTERIZERORDEREDTEXTURE1D; } diff --git a/libs/vkd3d-shader/hlsl.y b/libs/vkd3d-shader/hlsl.y index 47d126ac..fda6f597 100644 --- a/libs/vkd3d-shader/hlsl.y +++ b/libs/vkd3d-shader/hlsl.y @@ -5363,7 +5363,6 @@ static void validate_uav_type(struct hlsl_ctx *ctx, enum hlsl_sampler_dim dim, %token KW_PACKOFFSET %token KW_PASS %token KW_PIXELSHADER -%token KW_PRECISE %token KW_RASTERIZERORDEREDBUFFER %token KW_RASTERIZERORDEREDSTRUCTUREDBUFFER %token KW_RASTERIZERORDEREDTEXTURE1D @@ -6823,10 +6822,6 @@ var_modifiers: { $$ = add_modifiers(ctx, $2, HLSL_STORAGE_NOPERSPECTIVE, &@1); } - | KW_PRECISE var_modifiers - { - $$ = add_modifiers(ctx, $2, HLSL_MODIFIER_PRECISE, &@1); - } | KW_SHARED var_modifiers { $$ = add_modifiers(ctx, $2, HLSL_STORAGE_SHARED, &@1); @@ -6875,7 +6870,14 @@ var_modifiers: { $$ = add_modifiers(ctx, $2, HLSL_MODIFIER_INLINE, &@1); } - + | var_identifier var_modifiers + { + if (!strcmp($1, "precise")) + $$ = add_modifiers(ctx, $2, HLSL_MODIFIER_PRECISE, &@1); + else + hlsl_error(ctx, &@1, VKD3D_SHADER_ERROR_HLSL_UNKNOWN_MODIFIER, + "Unknown modifier %s.", debugstr_a($1)); + } complex_initializer: initializer_expr diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index ea52f151..07b9f94c 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -149,6 +149,7 @@ enum vkd3d_shader_error VKD3D_SHADER_ERROR_HLSL_NON_FINITE_RESULT = 5027, VKD3D_SHADER_ERROR_HLSL_DUPLICATE_SWITCH_CASE = 5028, VKD3D_SHADER_ERROR_HLSL_MISSING_TECHNIQUE = 5029, + VKD3D_SHADER_ERROR_HLSL_UNKNOWN_MODIFIER = 5030, VKD3D_SHADER_WARNING_HLSL_IMPLICIT_TRUNCATION = 5300, VKD3D_SHADER_WARNING_HLSL_DIVISION_BY_ZERO = 5301, diff --git a/tests/hlsl/precise-modifier.shader_test b/tests/hlsl/precise-modifier.shader_test new file mode 100644 index 00000000..7ad8f39e --- /dev/null +++ b/tests/hlsl/precise-modifier.shader_test @@ -0,0 +1,41 @@ +[pixel shader] +// 'precise' is not a keyword +float4 precise; + +float4 main() : sv_target +{ + return precise; +} + +[test] +uniform 0 float4 0.1 0.2 0.3 0.4 +draw quad +probe all rgba (0.1, 0.2, 0.3, 0.4) + +[pixel shader fail(sm>=6)] +precise float4 var; +typedef float4 precise; + +precise main() : sv_target +{ + return var; +} + +[test] +uniform 0 float4 0.1 0.2 0.3 0.4 +draw quad +probe all rgba (0.1, 0.2, 0.3, 0.4) + +[pixel shader] +float4 precise; +precise float4 var; + +float4 main() : sv_target +{ + return precise; +} + +[test] +uniform 0 float4 0.1 0.2 0.3 0.4 +draw quad +probe all rgba (0.1, 0.2, 0.3, 0.4)

1 0

Nikolay Sivov : vkd3d-shader: Parse a 'single' modifier.
by Alexandre Julliard 12 Mar '24

12 Mar '24

Module: vkd3d Branch: master Commit: e0ef45c0a37b70f7e0ecc49605410512a6180bcd URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/e0ef45c0a37b70f7e0ecc49605410… Author: Nikolay Sivov <nsivov(a)codeweavers.com> Date: Sun Feb 25 02:51:58 2024 +0100 vkd3d-shader: Parse a 'single' modifier. Signed-off-by: Nikolay Sivov <nsivov(a)codeweavers.com> --- libs/vkd3d-shader/hlsl.h | 1 + libs/vkd3d-shader/hlsl.y | 2 ++ tests/hlsl/cbuffer.shader_test | 60 ++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 63 insertions(+) diff --git a/libs/vkd3d-shader/hlsl.h b/libs/vkd3d-shader/hlsl.h index 1fb8b834..9781dd33 100644 --- a/libs/vkd3d-shader/hlsl.h +++ b/libs/vkd3d-shader/hlsl.h @@ -374,6 +374,7 @@ struct hlsl_attribute #define HLSL_STORAGE_CENTROID 0x00004000 #define HLSL_STORAGE_NOPERSPECTIVE 0x00008000 #define HLSL_STORAGE_LINEAR 0x00010000 +#define HLSL_MODIFIER_SINGLE 0x00020000 #define HLSL_TYPE_MODIFIERS_MASK (HLSL_MODIFIER_PRECISE | HLSL_MODIFIER_VOLATILE | \ HLSL_MODIFIER_CONST | HLSL_MODIFIER_ROW_MAJOR | \ diff --git a/libs/vkd3d-shader/hlsl.y b/libs/vkd3d-shader/hlsl.y index fda6f597..ec8b3d22 100644 --- a/libs/vkd3d-shader/hlsl.y +++ b/libs/vkd3d-shader/hlsl.y @@ -6874,6 +6874,8 @@ var_modifiers: { if (!strcmp($1, "precise")) $$ = add_modifiers(ctx, $2, HLSL_MODIFIER_PRECISE, &@1); + else if (!strcmp($1, "single")) + $$ = add_modifiers(ctx, $2, HLSL_MODIFIER_SINGLE, &@1); else hlsl_error(ctx, &@1, VKD3D_SHADER_ERROR_HLSL_UNKNOWN_MODIFIER, "Unknown modifier %s.", debugstr_a($1)); diff --git a/tests/hlsl/cbuffer.shader_test b/tests/hlsl/cbuffer.shader_test index bb0937ed..517fe636 100644 --- a/tests/hlsl/cbuffer.shader_test +++ b/tests/hlsl/cbuffer.shader_test @@ -29,11 +29,71 @@ float4 main() : sv_target return foo; } +[pixel shader fail(sm>=6)] +// The 'single' modifier is not a keyword. It's meaningful only for fx_5_0. +single cbuffer cb +{ + float4 single; +}; + +float4 main() : sv_target +{ + return single; +} + [test] uniform 0 float4 1.0 2.0 3.0 4.0 draw quad probe all rgba (1.0, 2.0, 3.0, 4.0) +[pixel shader fail(sm>=6)] +// 'single' modifier is usable when there is a variable with the same name +cbuffer cb +{ + float4 single; +}; + +single cbuffer cb2 +{ + float4 var; +}; + +float4 main() : sv_target +{ + return single; +} + +[test] +uniform 0 float4 1.0 2.0 3.0 4.0 +draw quad +probe all rgba (1.0, 2.0, 3.0, 4.0) + +[pixel shader] +typedef float4 single; +single var; + +float4 main() : sv_target +{ + return var; +} + +[test] +uniform 0 float4 1.0 2.0 3.0 4.0 +draw quad +probe all rgba (1.0, 2.0, 3.0, 4.0) + +[pixel shader fail] +typedef float4 single; +single cbuffer cb +{ + float4 var; +}; + +float4 main() : sv_target +{ + return var; +} + % SM1 buffer offset allocation follows different rules than SM4. % Those would have to be tested separately. [require]

1 0