From a5a94f52ffcbf3119d272a9369021a213ea6dad2 Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Wed, 9 Feb 2022 15:00:05 +0100 Subject: MacroHLE: Add MultidrawIndirect HLE Macro. --- src/video_core/dma_pusher.cpp | 14 ++++++++------ 1 file changed, 8 insertions(+), 6 deletions(-) (limited to 'src/video_core/dma_pusher.cpp') diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp index 322de2606..eb1371612 100644 --- a/src/video_core/dma_pusher.cpp +++ b/src/video_core/dma_pusher.cpp @@ -61,7 +61,7 @@ bool DmaPusher::Step() { } else { const CommandListHeader command_list_header{ command_list.command_lists[dma_pushbuffer_subindex++]}; - const GPUVAddr dma_get = command_list_header.addr; + dma_state.dma_get = command_list_header.addr; if (dma_pushbuffer_subindex >= command_list.command_lists.size()) { // We've gone through the current list, remove it from the queue @@ -75,11 +75,11 @@ bool DmaPusher::Step() { // Push buffer non-empty, read a word command_headers.resize_destructive(command_list_header.size); - if (Settings::IsGPULevelHigh()) { - memory_manager.ReadBlock(dma_get, command_headers.data(), + if (Settings::IsGPULevelExtreme()) { + memory_manager.ReadBlock(dma_state.dma_get, command_headers.data(), command_list_header.size * sizeof(u32)); } else { - memory_manager.ReadBlockUnsafe(dma_get, command_headers.data(), + memory_manager.ReadBlockUnsafe(dma_state.dma_get, command_headers.data(), command_list_header.size * sizeof(u32)); } ProcessCommands(command_headers); @@ -174,8 +174,10 @@ void DmaPusher::CallMultiMethod(const u32* base_start, u32 num_methods) const { puller.CallMultiMethod(dma_state.method, dma_state.subchannel, base_start, num_methods, dma_state.method_count); } else { - subchannels[dma_state.subchannel]->CallMultiMethod(dma_state.method, base_start, - num_methods, dma_state.method_count); + auto subchannel = subchannels[dma_state.subchannel]; + subchannel->current_dma_segment = dma_state.dma_get; + subchannel->CallMultiMethod(dma_state.method, base_start, num_methods, + dma_state.method_count); } } -- cgit v1.2.3 From 0f89828073a541eaa2cfd985483f839bd2f97b74 Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Wed, 9 Feb 2022 15:39:40 +0100 Subject: MacroHLE: Implement DrawIndexedIndirect & DrawArraysIndirect. --- src/video_core/dma_pusher.cpp | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'src/video_core/dma_pusher.cpp') diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp index eb1371612..13ff64fa3 100644 --- a/src/video_core/dma_pusher.cpp +++ b/src/video_core/dma_pusher.cpp @@ -97,6 +97,7 @@ void DmaPusher::ProcessCommands(std::span commands) { if (dma_state.non_incrementing) { const u32 max_write = static_cast( std::min(index + dma_state.method_count, commands.size()) - index); + dma_state.dma_word_offset = static_cast(index * sizeof(u32)); CallMultiMethod(&command_header.argument, max_write); dma_state.method_count -= max_write; dma_state.is_last_call = true; @@ -175,7 +176,7 @@ void DmaPusher::CallMultiMethod(const u32* base_start, u32 num_methods) const { dma_state.method_count); } else { auto subchannel = subchannels[dma_state.subchannel]; - subchannel->current_dma_segment = dma_state.dma_get; + subchannel->current_dma_segment = dma_state.dma_get + dma_state.dma_word_offset; subchannel->CallMultiMethod(dma_state.method, base_start, num_methods, dma_state.method_count); } -- cgit v1.2.3 From c541559767c3912940ee3d73a122530b3edde9f1 Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Sat, 5 Mar 2022 08:01:13 +0100 Subject: MacroHLE: Refactor MacroHLE system. --- src/video_core/dma_pusher.cpp | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) (limited to 'src/video_core/dma_pusher.cpp') diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp index 13ff64fa3..5ad40abaa 100644 --- a/src/video_core/dma_pusher.cpp +++ b/src/video_core/dma_pusher.cpp @@ -94,10 +94,10 @@ void DmaPusher::ProcessCommands(std::span commands) { if (dma_state.method_count) { // Data word of methods command + dma_state.dma_word_offset = static_cast(index * sizeof(u32)); if (dma_state.non_incrementing) { const u32 max_write = static_cast( std::min(index + dma_state.method_count, commands.size()) - index); - dma_state.dma_word_offset = static_cast(index * sizeof(u32)); CallMultiMethod(&command_header.argument, max_write); dma_state.method_count -= max_write; dma_state.is_last_call = true; @@ -133,6 +133,8 @@ void DmaPusher::ProcessCommands(std::span commands) { case SubmissionMode::Inline: dma_state.method = command_header.method; dma_state.subchannel = command_header.subchannel; + dma_state.dma_word_offset = static_cast( + -static_cast(dma_state.dma_get)); // negate to set address as 0 CallMethod(command_header.arg_count); dma_state.non_incrementing = true; dma_increment_once = false; @@ -165,8 +167,9 @@ void DmaPusher::CallMethod(u32 argument) const { dma_state.method_count, }); } else { - subchannels[dma_state.subchannel]->CallMethod(dma_state.method, argument, - dma_state.is_last_call); + auto subchannel = subchannels[dma_state.subchannel]; + subchannel->current_dma_segment = dma_state.dma_get + dma_state.dma_word_offset; + subchannel->CallMethod(dma_state.method, argument, dma_state.is_last_call); } } -- cgit v1.2.3 From 18637766efd1ff9a0c22967553983cfda69c96ca Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Thu, 17 Nov 2022 16:36:53 +0100 Subject: MacroHLE: Reduce massive calculations on sizing estimation. --- src/video_core/dma_pusher.cpp | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'src/video_core/dma_pusher.cpp') diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp index 5ad40abaa..7a82355da 100644 --- a/src/video_core/dma_pusher.cpp +++ b/src/video_core/dma_pusher.cpp @@ -75,7 +75,8 @@ bool DmaPusher::Step() { // Push buffer non-empty, read a word command_headers.resize_destructive(command_list_header.size); - if (Settings::IsGPULevelExtreme()) { + constexpr u32 MacroRegistersStart = 0xE00; + if (dma_state.method < MacroRegistersStart) { memory_manager.ReadBlock(dma_state.dma_get, command_headers.data(), command_list_header.size * sizeof(u32)); } else { -- cgit v1.2.3 From ce448ce770b6c329caec7ad1ae00e01dddb67b03 Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Fri, 18 Nov 2022 00:21:13 +0100 Subject: Revert Buffer cache changes and setup additional macros. --- src/video_core/dma_pusher.cpp | 17 +++++++++++++---- 1 file changed, 13 insertions(+), 4 deletions(-) (limited to 'src/video_core/dma_pusher.cpp') diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp index 7a82355da..b3e9cb82e 100644 --- a/src/video_core/dma_pusher.cpp +++ b/src/video_core/dma_pusher.cpp @@ -77,11 +77,20 @@ bool DmaPusher::Step() { command_headers.resize_destructive(command_list_header.size); constexpr u32 MacroRegistersStart = 0xE00; if (dma_state.method < MacroRegistersStart) { - memory_manager.ReadBlock(dma_state.dma_get, command_headers.data(), - command_list_header.size * sizeof(u32)); + if (Settings::IsGPULevelHigh()) { + memory_manager.ReadBlock(dma_state.dma_get, command_headers.data(), + command_list_header.size * sizeof(u32)); + } else { + memory_manager.ReadBlockUnsafe(dma_state.dma_get, command_headers.data(), + command_list_header.size * sizeof(u32)); + } } else { - memory_manager.ReadBlockUnsafe(dma_state.dma_get, command_headers.data(), - command_list_header.size * sizeof(u32)); + const size_t copy_size = command_list_header.size * sizeof(u32); + if (subchannels[dma_state.subchannel]) { + subchannels[dma_state.subchannel]->current_dirty = + memory_manager.IsMemoryDirty(dma_state.dma_get, copy_size); + } + memory_manager.ReadBlockUnsafe(dma_state.dma_get, command_headers.data(), copy_size); } ProcessCommands(command_headers); } -- cgit v1.2.3 From cb1497d0d7711a1c0e527aaa3e1dc3f95e5a6644 Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Sun, 27 Nov 2022 00:58:06 +0100 Subject: DMAPusher: Improve collection of non executing methods --- src/video_core/dma_pusher.cpp | 6 ++++++ 1 file changed, 6 insertions(+) (limited to 'src/video_core/dma_pusher.cpp') diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp index b3e9cb82e..551929824 100644 --- a/src/video_core/dma_pusher.cpp +++ b/src/video_core/dma_pusher.cpp @@ -178,6 +178,11 @@ void DmaPusher::CallMethod(u32 argument) const { }); } else { auto subchannel = subchannels[dma_state.subchannel]; + if (!subchannel->execution_mask[dma_state.method]) [[likely]] { + subchannel->method_sink.emplace_back(dma_state.method, argument); + return; + } + subchannel->ConsumeSink(); subchannel->current_dma_segment = dma_state.dma_get + dma_state.dma_word_offset; subchannel->CallMethod(dma_state.method, argument, dma_state.is_last_call); } @@ -189,6 +194,7 @@ void DmaPusher::CallMultiMethod(const u32* base_start, u32 num_methods) const { dma_state.method_count); } else { auto subchannel = subchannels[dma_state.subchannel]; + subchannel->ConsumeSink(); subchannel->current_dma_segment = dma_state.dma_get + dma_state.dma_word_offset; subchannel->CallMultiMethod(dma_state.method, base_start, num_methods, dma_state.method_count); -- cgit v1.2.3