Use peek-poke for compute pass encoding

2026-04-22 03:02:01 -04:00 · 2020-01-08 14:26:51 -05:00
parent 20cd803d67
commit c01a7c6abe
9 changed files with 257 additions and 28 deletions
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -365,11 +365,37 @@ dependencies = [
 "winapi 0.3.8 (registry+https://github.com/rust-lang/crates.io-index)",
 ]

+[[package]]
+name = "peek-poke"
+version = "0.2.0"
+source = "git+https://github.com/djg/peek-poke?rev=728591c140ead9a6b646a27eaad29e5136fb96b8#728591c140ead9a6b646a27eaad29e5136fb96b8"
+dependencies = [
+ "peek-poke-derive 0.2.0 (git+https://github.com/djg/peek-poke?rev=728591c140ead9a6b646a27eaad29e5136fb96b8)",
+]
+
+[[package]]
+name = "peek-poke-derive"
+version = "0.2.0"
+source = "git+https://github.com/djg/peek-poke?rev=728591c140ead9a6b646a27eaad29e5136fb96b8#728591c140ead9a6b646a27eaad29e5136fb96b8"
+dependencies = [
+ "proc-macro2 0.4.30 (registry+https://github.com/rust-lang/crates.io-index)",
+ "quote 0.6.13 (registry+https://github.com/rust-lang/crates.io-index)",
+ "syn 0.15.44 (registry+https://github.com/rust-lang/crates.io-index)",
+]
+
 [[package]]
 name = "pkg-config"
 version = "0.3.17"
 source = "registry+https://github.com/rust-lang/crates.io-index"

+[[package]]
+name = "proc-macro2"
+version = "0.4.30"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+dependencies = [
+ "unicode-xid 0.1.0 (registry+https://github.com/rust-lang/crates.io-index)",
+]
+
 [[package]]
 name = "proc-macro2"
 version = "1.0.6"
@@ -378,6 +404,14 @@ dependencies = [
 "unicode-xid 0.2.0 (registry+https://github.com/rust-lang/crates.io-index)",
 ]

+[[package]]
+name = "quote"
+version = "0.6.13"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+dependencies = [
+ "proc-macro2 0.4.30 (registry+https://github.com/rust-lang/crates.io-index)",
+]
+
 [[package]]
 name = "quote"
 version = "1.0.2"
@@ -527,6 +561,16 @@ dependencies = [
 "lock_api 0.3.2 (registry+https://github.com/rust-lang/crates.io-index)",
 ]

+[[package]]
+name = "syn"
+version = "0.15.44"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+dependencies = [
+ "proc-macro2 0.4.30 (registry+https://github.com/rust-lang/crates.io-index)",
+ "quote 0.6.13 (registry+https://github.com/rust-lang/crates.io-index)",
+ "unicode-xid 0.1.0 (registry+https://github.com/rust-lang/crates.io-index)",
+]
+
 [[package]]
 name = "syn"
 version = "1.0.11"
@@ -537,6 +581,11 @@ dependencies = [
 "unicode-xid 0.2.0 (registry+https://github.com/rust-lang/crates.io-index)",
 ]

+[[package]]
+name = "unicode-xid"
+version = "0.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+
 [[package]]
 name = "unicode-xid"
 version = "0.2.0"
@@ -612,6 +661,7 @@ dependencies = [
 "gfx-hal 0.4.1 (registry+https://github.com/rust-lang/crates.io-index)",
 "log 0.4.8 (registry+https://github.com/rust-lang/crates.io-index)",
 "parking_lot 0.9.0 (registry+https://github.com/rust-lang/crates.io-index)",
+ "peek-poke 0.2.0 (git+https://github.com/djg/peek-poke?rev=728591c140ead9a6b646a27eaad29e5136fb96b8)",
 "rendy-descriptor 0.5.1 (registry+https://github.com/rust-lang/crates.io-index)",
 "rendy-memory 0.5.2 (registry+https://github.com/rust-lang/crates.io-index)",
 "serde 1.0.103 (registry+https://github.com/rust-lang/crates.io-index)",
@@ -717,8 +767,12 @@ dependencies = [
 "checksum objc_exception 0.1.2 (registry+https://github.com/rust-lang/crates.io-index)" = "ad970fb455818ad6cba4c122ad012fae53ae8b4795f86378bce65e4f6bab2ca4"
 "checksum parking_lot 0.9.0 (registry+https://github.com/rust-lang/crates.io-index)" = "f842b1982eb6c2fe34036a4fbfb06dd185a3f5c8edfaacdf7d1ea10b07de6252"
 "checksum parking_lot_core 0.6.2 (registry+https://github.com/rust-lang/crates.io-index)" = "b876b1b9e7ac6e1a74a6da34d25c42e17e8862aa409cbbbdcfc8d86c6f3bc62b"
+"checksum peek-poke 0.2.0 (git+https://github.com/djg/peek-poke?rev=728591c140ead9a6b646a27eaad29e5136fb96b8)" = "<none>"
+"checksum peek-poke-derive 0.2.0 (git+https://github.com/djg/peek-poke?rev=728591c140ead9a6b646a27eaad29e5136fb96b8)" = "<none>"
 "checksum pkg-config 0.3.17 (registry+https://github.com/rust-lang/crates.io-index)" = "05da548ad6865900e60eaba7f589cc0783590a92e940c26953ff81ddbab2d677"
+"checksum proc-macro2 0.4.30 (registry+https://github.com/rust-lang/crates.io-index)" = "cf3d2011ab5c909338f7887f4fc896d35932e29146c12c8d01da6b22a80ba759"
 "checksum proc-macro2 1.0.6 (registry+https://github.com/rust-lang/crates.io-index)" = "9c9e470a8dc4aeae2dee2f335e8f533e2d4b347e1434e5671afc49b054592f27"
+"checksum quote 0.6.13 (registry+https://github.com/rust-lang/crates.io-index)" = "6ce23b6b870e8f94f81fb0a363d65d86675884b34a09043c81e5562f11c1f8e1"
 "checksum quote 1.0.2 (registry+https://github.com/rust-lang/crates.io-index)" = "053a8c8bcc71fcce321828dc897a98ab9760bef03a4fc36693c231e5b3216cfe"
 "checksum range-alloc 0.1.0 (registry+https://github.com/rust-lang/crates.io-index)" = "dd5927936723a9e8b715d37d7e4b390455087c4bdf25b9f702309460577b14f9"
 "checksum raw-window-handle 0.3.3 (registry+https://github.com/rust-lang/crates.io-index)" = "0a441a7a6c80ad6473bd4b74ec1c9a4c951794285bf941c2126f607c72e48211"
@@ -738,7 +792,9 @@ dependencies = [
 "checksum smallvec 1.0.0 (registry+https://github.com/rust-lang/crates.io-index)" = "4ecf3b85f68e8abaa7555aa5abdb1153079387e60b718283d732f03897fcfc86"
 "checksum spirv_cross 0.16.0 (registry+https://github.com/rust-lang/crates.io-index)" = "fbbe441b3ac8ec0ae6a4f05234239bd372a241ce15793eef694e8b24afc267bb"
 "checksum storage-map 0.2.0 (registry+https://github.com/rust-lang/crates.io-index)" = "fd0a4829a5c591dc24a944a736d6b1e4053e51339a79fd5d4702c4c999a9c45e"
+"checksum syn 0.15.44 (registry+https://github.com/rust-lang/crates.io-index)" = "9ca4b3b69a77cbe1ffc9e198781b7acb0c7365a883670e8f1c1bc66fba79a5c5"
 "checksum syn 1.0.11 (registry+https://github.com/rust-lang/crates.io-index)" = "dff0acdb207ae2fe6d5976617f887eb1e35a2ba52c13c7234c790960cdad9238"
+"checksum unicode-xid 0.1.0 (registry+https://github.com/rust-lang/crates.io-index)" = "fc72304796d0818e357ead4e000d19c9c174ab23dc11093ac919054d20a6a7fc"
 "checksum unicode-xid 0.2.0 (registry+https://github.com/rust-lang/crates.io-index)" = "826e7639553986605ec5979c7dd957c7895e93eabed50ab2ffa7f6128a75097c"
 "checksum vec_map 0.8.1 (registry+https://github.com/rust-lang/crates.io-index)" = "05c78687fb1a80548ae3250346c3db86a80a7cdd77bda190189f2d0a0987c81a"
 "checksum wasm-bindgen 0.2.55 (registry+https://github.com/rust-lang/crates.io-index)" = "29ae32af33bacd663a9a28241abecf01f2be64e6a185c6139b04f18b6385c5f2"
--- a/ffi/wgpu.h
+++ b/ffi/wgpu.h
@@ -243,6 +243,10 @@ typedef enum {
  WGPUVertexFormat_Int4 = 48,
 } WGPUVertexFormat;

+typedef struct WGPURawPass WGPURawPass;
+
+typedef struct WGPURenderCommand WGPURenderCommand;
+
 typedef uint64_t WGPUId_Device_Dummy;

 typedef WGPUId_Device_Dummy WGPUDeviceId;
@@ -698,6 +702,8 @@ WGPUComputePassId wgpu_command_encoder_begin_compute_pass(WGPUCommandEncoderId e
 WGPURenderPassId wgpu_command_encoder_begin_render_pass(WGPUCommandEncoderId encoder_id,
                                                        const WGPURenderPassDescriptor *desc);

+void wgpu_command_encoder_compute_pass(WGPUCommandEncoderId self_id, const WGPURawPass *pass);
+
 void wgpu_command_encoder_copy_buffer_to_buffer(WGPUCommandEncoderId command_encoder_id,
                                                WGPUBufferId source,
                                                WGPUBufferAddress source_offset,
@@ -725,6 +731,15 @@ void wgpu_command_encoder_destroy(WGPUCommandEncoderId command_encoder_id);
 WGPUCommandBufferId wgpu_command_encoder_finish(WGPUCommandEncoderId encoder_id,
                                                const WGPUCommandBufferDescriptor *desc);

+void wgpu_command_encoder_render_pass(WGPUCommandEncoderId self_id,
+                                      const WGPURenderPassColorAttachmentDescriptor *color_attachments,
+                                      uintptr_t color_attachment_length,
+                                      const WGPURenderPassDepthStencilAttachmentDescriptor *depth_stencil_attachment,
+                                      const WGPURenderCommand *commands,
+                                      uintptr_t command_length,
+                                      const WGPUBufferAddress *offsets,
+                                      uintptr_t offset_length);
+
 void wgpu_compute_pass_dispatch(WGPUComputePassId pass_id, uint32_t x, uint32_t y, uint32_t z);

 void wgpu_compute_pass_dispatch_indirect(WGPUComputePassId pass_id,
--- a/wgpu-core/Cargo.toml
+++ b/wgpu-core/Cargo.toml
@@ -29,6 +29,7 @@ log = "0.4"
 hal = { package = "gfx-hal", version = "0.4" }
 gfx-backend-empty = { version = "0.4" }
 parking_lot = "0.9"
+peek-poke = { git = "https://github.com/djg/peek-poke", rev = "728591c140ead9a6b646a27eaad29e5136fb96b8" }
 rendy-memory = "0.5"
 rendy-descriptor = "0.5"
 serde = { version = "1.0", features = ["serde_derive"], optional = true }
--- a/wgpu-core/src/command/compute.rs
+++ b/wgpu-core/src/command/compute.rs
@@ -6,7 +6,7 @@ use crate::{
    command::{
        bind::{Binder, LayoutChange},
        CommandBuffer,
-        OffsetIndex,
+        RawPass,
    },
    device::{all_buffer_stages, BIND_BUFFER_ALIGNMENT},
    hub::{GfxBackend, Global, IdentityFilter, Token},
@@ -18,17 +18,17 @@ use crate::{
 };

 use hal::command::CommandBuffer as _;
+use peek_poke::{Peek, PeekCopy, Poke};

-use std::{iter, ops::Range};
+use std::{convert::TryInto, iter, mem, ptr, slice};


-#[non_exhaustive]
-#[derive(Debug)]
+#[derive(Clone, Copy, Debug, PeekCopy, Poke)]
 pub enum ComputeCommand {
    SetBindGroup {
        index: u32,
+        num_dynamic_offsets: u8,
        bind_group_id: id::BindGroupId,
-        offset_indices: Range<OffsetIndex>,
    },
    SetPipeline(id::ComputePipelineId),
    Dispatch([u32; 3]),
@@ -38,12 +38,6 @@ pub enum ComputeCommand {
    },
 }

-#[derive(Copy, Clone, Debug)]
-pub struct StandaloneComputePass<'a> {
-    pub commands: &'a [ComputeCommand],
-    pub offsets: &'a [BufferAddress],
-}
-
 #[repr(C)]
 #[derive(Clone, Debug, Default)]
 pub struct ComputePassDescriptor {
@@ -96,7 +90,7 @@ impl<F> Global<F> {
    pub fn command_encoder_run_compute_pass<B: GfxBackend>(
        &self,
        encoder_id: id::CommandEncoderId,
-        pass: StandaloneComputePass,
+        raw_data: &[u8],
    ) {
        let hub = B::hub(self);
        let mut token = Token::root();
@@ -112,10 +106,23 @@ impl<F> Global<F> {
        let (buffer_guard, mut token) = hub.buffers.read(&mut token);
        let (texture_guard, _) = hub.textures.read(&mut token);

-        for command in pass.commands {
-            match *command {
-                ComputeCommand::SetBindGroup { index, bind_group_id, ref offset_indices } => {
-                    let offsets = &pass.offsets[offset_indices.start as usize .. offset_indices.end as usize];
+        let mut peeker = raw_data.as_ptr();
+        let raw_data_end = unsafe {
+            raw_data.as_ptr().add(raw_data.len())
+        };
+        let mut command = ComputeCommand::Dispatch([0; 3]); // dummy
+        while unsafe { peeker.add(mem::size_of::<ComputeCommand>()) } <= raw_data_end {
+            peeker = unsafe { command.peek_from(peeker) };
+            match command {
+                ComputeCommand::SetBindGroup { index, num_dynamic_offsets, bind_group_id } => {
+                    debug_assert_eq!(peeker.align_offset(mem::align_of::<BufferAddress>()), 0);
+                    let extra_size = (num_dynamic_offsets as usize) * mem::size_of::<BufferAddress>();
+                    let end = unsafe { peeker.add(extra_size) };
+                    assert!(end <= raw_data_end);
+                    let offsets = unsafe {
+                        slice::from_raw_parts(peeker as *const BufferAddress, num_dynamic_offsets as usize)
+                    };
+                    peeker = end;
                    if cfg!(debug_assertions) {
                        for off in offsets {
                            assert_eq!(
@@ -234,6 +241,8 @@ impl<F> Global<F> {
                }
            }
        }
+
+        assert_eq!(peeker, raw_data_end);
    }

    pub fn compute_pass_set_bind_group<B: GfxBackend>(
@@ -413,3 +422,73 @@ impl<F> Global<F> {
        }
    }
 }
+
+impl RawPass {
+    #[inline]
+    unsafe fn encode(&mut self, command: &ComputeCommand) {
+        self.ensure_extra_size(mem::size_of::<ComputeCommand>());
+        self.data = command.poke_into(self.data);
+    }
+
+    #[inline]
+    unsafe fn encode_with<T>(&mut self, command: &ComputeCommand, extra: &[T]) {
+        let extra_size = extra.len() * mem::size_of::<T>();
+        self.ensure_extra_size(mem::size_of::<ComputeCommand>() + extra_size);
+        self.data = command.poke_into(self.data);
+        debug_assert_eq!(self.data.align_offset(mem::align_of::<T>()), 0);
+        ptr::copy_nonoverlapping(extra.as_ptr(), self.data as *mut T, extra.len());
+        self.data = self.data.add(extra_size);
+    }
+
+    #[no_mangle]
+    pub unsafe extern "C" fn wgpu_raw_compute_pass_set_bind_group(
+        &mut self,
+        index: u32,
+        bind_group_id: id::BindGroupId,
+        offsets: *const BufferAddress,
+        offset_length: usize,
+    ) {
+        self.encode_with(
+            &ComputeCommand::SetBindGroup {
+                index,
+                num_dynamic_offsets: offset_length.try_into().unwrap(),
+                bind_group_id,
+            },
+            slice::from_raw_parts(offsets, offset_length),
+        );
+
+        for offset in slice::from_raw_parts(offsets, offset_length) {
+            self.data = offset.poke_into(self.data);
+        }
+    }
+
+    #[no_mangle]
+    pub unsafe extern "C" fn wgpu_standalone_compute_pass_set_pipeline(
+        &mut self,
+        pipeline_id: id::ComputePipelineId,
+    ) {
+        self.encode(&ComputeCommand::SetPipeline(pipeline_id));
+    }
+
+    #[no_mangle]
+    pub unsafe extern "C" fn wgpu_standalone_compute_pass_dispatch(
+        &mut self,
+        groups_x: u32,
+        groups_y: u32,
+        groups_z: u32,
+    ) {
+        self.encode(&ComputeCommand::Dispatch([groups_x, groups_y, groups_z]));
+    }
+
+    #[no_mangle]
+    pub unsafe extern "C" fn wgpu_standalone_compute_pass_dispatch_indirect(
+        &mut self,
+        buffer_id: id::BufferId,
+        offset: BufferAddress,
+    ) {
+        self.encode(&ComputeCommand::DispatchIndirect {
+            buffer_id,
+            offset,
+        });
+    }
+}
--- a/wgpu-core/src/command/mod.rs
+++ b/wgpu-core/src/command/mod.rs
@@ -54,7 +54,45 @@ use std::{
 };


-pub type OffsetIndex = u16;
+pub struct RawPass {
+    data: *mut u8,
+    base: *mut u8,
+    capacity: usize,
+}
+
+impl RawPass {
+    pub fn new() -> Self {
+        let mut vec = Vec::with_capacity(16);
+        RawPass {
+            data: vec.as_mut_ptr(),
+            base: vec.as_mut_ptr(),
+            capacity: vec.capacity(),
+        }
+    }
+
+    pub unsafe fn delete(self) {
+        let size = self.data as usize - self.base as usize;
+        let _ = Vec::from_raw_parts(self.base, size, self.capacity);
+    }
+
+    pub unsafe fn to_slice(&self) -> &[u8] {
+        let size = self.data as usize - self.base as usize;
+        assert!(size <= self.capacity);
+        slice::from_raw_parts(self.base, size)
+    }
+
+    unsafe fn ensure_extra_size(&mut self, extra_size: usize) {
+        let size = self.data as usize - self.base as usize;
+        if let Some(extra_capacity) = (size + extra_size).checked_sub(self.capacity) {
+            let mut vec = Vec::from_raw_parts(self.base, size, self.capacity);
+            vec.reserve(extra_capacity);
+            //let (data, size, capacity) = vec.into_raw_parts(); //TODO: when stable
+            self.data = vec.as_mut_ptr().add(vec.len());
+            self.base = vec.as_mut_ptr();
+            self.capacity = vec.capacity();
+        }
+    }
+}

 pub struct RenderBundle<B: hal::Backend> {
    _raw: B::CommandBuffer,
--- a/wgpu-core/src/command/render.rs
+++ b/wgpu-core/src/command/render.rs
@@ -6,7 +6,6 @@ use crate::{
    command::{
        bind::{Binder, LayoutChange},
        CommandBuffer,
-        OffsetIndex,
    },
    conv,
    device::{
@@ -38,6 +37,8 @@ use std::{
 };


+type OffsetIndex = u16;
+
 #[repr(C)]
 #[derive(Copy, Clone, Debug, Hash, Eq, PartialEq)]
 pub enum LoadOp {
--- a/wgpu-core/src/id.rs
+++ b/wgpu-core/src/id.rs
@@ -5,7 +5,7 @@
 use crate::{Backend, Epoch, Index};
 #[cfg(feature = "serde")]
 use serde::{Deserialize, Serialize};
-use std::{fmt, marker::PhantomData};
+use std::{fmt, marker::PhantomData, mem};

 const BACKEND_BITS: usize = 3;
 const EPOCH_MASK: u32 = (1 << (32 - BACKEND_BITS)) - 1;
@@ -57,6 +57,21 @@ impl<T> PartialEq for Id<T> {
    }
 }

+unsafe impl<T> peek_poke::Poke for Id<T> {
+    fn max_size() -> usize {
+         mem::size_of::<u64>()
+    }
+    unsafe fn poke_into(&self, data: *mut u8) -> *mut u8 {
+        self.0.poke_into(data)
+    }
+}
+
+impl<T> peek_poke::Peek for Id<T> {
+    unsafe fn peek_from(&mut self, data: *const u8) -> *const u8 {
+        self.0.peek_from(data)
+    }
+}
+
 pub trait TypedId {
    fn zip(index: Index, epoch: Epoch, backend: Backend) -> Self;
    fn unzip(self) -> (Index, Epoch, Backend);
--- a/wgpu-native/src/command.rs
+++ b/wgpu-native/src/command.rs
@@ -315,3 +315,32 @@ pub extern "C" fn wgpu_compute_pass_set_pipeline(
 ) {
    gfx_select!(pass_id => GLOBAL.compute_pass_set_pipeline(pass_id, pipeline_id))
 }
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_command_encoder_compute_pass(
+    self_id: id::CommandEncoderId,
+    pass: &core::command::RawPass,
+) {
+    let raw_data = pass.to_slice();
+    gfx_select!(self_id => GLOBAL.command_encoder_run_compute_pass(self_id, raw_data));
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_command_encoder_render_pass(
+    self_id: id::CommandEncoderId,
+    color_attachments: *const core::command::RenderPassColorAttachmentDescriptor,
+    color_attachment_length: usize,
+    depth_stencil_attachment: Option<&core::command::RenderPassDepthStencilAttachmentDescriptor>,
+    commands: *const core::command::RenderCommand,
+    command_length: usize,
+    offsets:  *const core::BufferAddress,
+    offset_length: usize,
+) {
+    let pass = core::command::StandaloneRenderPass {
+        color_attachments: slice::from_raw_parts(color_attachments, color_attachment_length),
+        depth_stencil_attachment,
+        commands: slice::from_raw_parts(commands, command_length),
+        offsets: slice::from_raw_parts(offsets, offset_length),
+    };
+    gfx_select!(self_id => GLOBAL.command_encoder_run_render_pass(self_id, pass));
+}
--- a/wgpu-remote/src/server.rs
+++ b/wgpu-remote/src/server.rs
@@ -127,16 +127,11 @@ pub extern "C" fn wgpu_server_encoder_destroy(
 pub unsafe extern "C" fn wgpu_server_encode_compute_pass(
    global: &Global,
    self_id: id::CommandEncoderId,
-    commands: *const core::command::ComputeCommand,
-    command_length: usize,
-    offsets:  *const core::BufferAddress,
-    offset_length: usize,
+    bytes: *const u8,
+    byte_length: usize,
 ) {
-    let pass = core::command::StandaloneComputePass {
-        commands: slice::from_raw_parts(commands, command_length),
-        offsets: slice::from_raw_parts(offsets, offset_length),
-    };
-    gfx_select!(self_id => global.command_encoder_run_compute_pass(self_id, pass));
+    let raw_data = slice::from_raw_parts(bytes, byte_length);
+    gfx_select!(self_id => global.command_encoder_run_compute_pass(self_id, raw_data));
 }

 #[no_mangle]