Regenerate main CUDA header, fix zluda_dump

.devcontainer/Dockerfile

@@ -8,16 +8,27 @@ RUN DEBIAN_FRONTEND=noninteractive apt-get update -y && DEBIAN_FRONTEND=noninter
     python3 \
     ripgrep \
     git \
-    ltrace
+    ltrace \
+    # required by llvm 17
+    lsb-release software-properties-common gnupg
+
+ARG LLVM_VERSION=17
+RUN wget https://apt.llvm.org/llvm.sh && \
+    chmod +x llvm.sh && \
+    ./llvm.sh ${LLVM_VERSION}
 
 # Feel free to change to a newer version if you have a newer verison on your host
-ARG CUDA_VERSION=12-4
+ARG CUDA_PKG_VERSION=12-4
 # Docker <-> host  driver version compatiblity is newer host <-> older docker
 # We don't care about a specific driver version, so pick oldest 5XX
 ARG CUDA_DRIVER=515
-RUN DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends \
+RUN DEBIAN_FRONTEND=noninteractive apt-get update -y && DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends \
+    # CUDA headers need it for interop
+    libgl-dev libegl-dev libvdpau-dev \
     nvidia-utils-${CUDA_DRIVER} \
-    cuda-cudart-${CUDA_VERSION}
+    cuda-cudart-dev-${CUDA_PKG_VERSION} \
+    cuda-cudart-${CUDA_PKG_VERSION} \
+    cuda-profiler-api-${CUDA_PKG_VERSION}
 
 ARG ROCM_VERSION=6.2.2
 RUN mkdir --parents --mode=0755 /etc/apt/keyrings && \
@@ -29,9 +40,10 @@ RUN mkdir --parents --mode=0755 /etc/apt/keyrings && \
     rocminfo \
     rocm-gdb \
     rocm-smi-lib \
+    rocm-llvm-dev \
     hip-runtime-amd &&  \
     echo '/opt/rocm/lib' > /etc/ld.so.conf.d/rocm.conf && \
     ldconfig
 
-ENV PATH=$PATH:/opt/rocm-6.2.2/bin
+ENV PATH=$PATH:/opt/rocm-${ROCM_VERSION}/bin
 

.devcontainer/devcontainer.json

@@ -28,7 +28,7 @@
 	//"hostRequirements": { "gpu": "optional" }
 	"customizations": {
 		"vscode": {
-			"extensions": [ "mhutchie.git-graph" ],
+			"extensions": [ "mhutchie.git-graph" ]
 		}
-}
+	}
 }

cuda_base/README

@@ -1 +1 @@
-bindgen build/wrapper.h -o src/cuda.rs --no-partialeq "CUDA_HOST_NODE_PARAMS_st" --with-derive-eq --whitelist-function="^cu.*" --whitelist-var="^CU.*" --size_t-is-usize --default-enum-style=newtype --no-layout-tests --no-doc-comments --no-derive-debug --new-type-alias "^CUdevice_v\d+$|^CUdeviceptr_v\d+$" --must-use-type "cudaError_enum" -- -I/usr/local/cuda/include
+bindgen build/wrapper.h -o src/cuda.rs --no-partialeq "CUDA_HOST_NODE_PARAMS_st" --with-derive-eq --allowlist-type="^CU.*" --allowlist-function="^cu.*" --allowlist-var="^CU.*" --default-enum-style=newtype --no-layout-tests --no-doc-comments --no-derive-debug --new-type-alias "^CUdevice_v\d+$|^CUdeviceptr_v\d+$" --must-use-type "cudaError_enum" -- -I/usr/local/cuda/include

cuda_base/build/wrapper.h

@@ -1,3 +1,7 @@
 #define __CUDA_API_VERSION_INTERNAL
 #include <cuda.h>
-#include <cudaGL.h>
+#include <cudaProfiler.h>
+#include <cudaGL.h>
+#include <cudaEGL.h>
+#include <vdpau/vdpau.h>
+#include <cudaVDPAU.h>

cuda_base/src/lib.rs

@@ -216,6 +216,8 @@ fn prepend_cuda_path_to_path(base_path: &Path, path: Path) -> Path {
         if ident.starts_with("CU")
             || ident.starts_with("cu")
             || ident.starts_with("GL")
+            || ident.starts_with("EGL")
+            || ident.starts_with("Vdp")
             || ident == "HGPUNV"
         {
             let mut base_path = base_path.clone();
@@ -343,11 +345,10 @@ fn cuda_derive_display_trait_for_item(
             None
         }
         Item::Struct(item_struct) => {
-            let item_struct_name = item_struct.ident.to_string();
             if state.ignore_types.contains(&item_struct.ident) {
                 return None;
             }
-            if item_struct_name.ends_with("_enum") {
+            if state.enums.contains_key(&item_struct.ident) {
                 let enum_ = &item_struct.ident;
                 let enum_iter = iter::repeat(&item_struct.ident);
                 let variants = state.enums.get(&item_struct.ident).unwrap().iter();

ptx_parser/src/lib.rs

@@ -284,20 +284,40 @@ fn immediate_value<'a, 'input>(stream: &mut PtxParser<'a, 'input>) -> PResult<as
     .parse_next(stream)
 }
 
-pub fn parse_module_unchecked<'input>(text: &'input str) -> Option<ast::Module<'input>> {
+pub fn parse_for_errors<'input>(text: &'input str) -> Vec<PtxError> {
-    let input = lex_with_span(text).ok()?;
+    let (tokens, mut errors) = lex_with_span_unchecked(text);
-    let mut errors = Vec::new();
+    let parse_result = {
-    let state = PtxParserState::new(text, &mut errors);
+        let state = PtxParserState::new(text, &mut errors);
-    let parser = PtxParser {
+        let parser = PtxParser {
-        state,
+            state,
-        input: &input[..],
+            input: &tokens[..],
+        };
+        module
+            .parse(parser)
+            .map_err(|err| PtxError::Parser(err.into_inner()))
     };
-    let parsing_result = module.parse(parser).ok();
+    match parse_result {
-    if !errors.is_empty() {
+        Ok(_) => {}
-        None
+        Err(err) => {
-    } else {
+            errors.push(err);
-        parsing_result
+        }
     }
+    errors
+}
+
+fn lex_with_span_unchecked<'input>(
+    text: &'input str,
+) -> (Vec<(Token<'input>, logos::Span)>, Vec<PtxError>) {
+    let lexer = Token::lexer(text);
+    let mut result = Vec::new();
+    let mut errors = Vec::new();
+    for (token, span) in lexer.spanned() {
+        match token {
+            Ok(t) => result.push((t, span)),
+            Err(err) => errors.push(PtxError::Lexer { source: err }),
+        }
+    }
+    (result, errors)
 }
 
 pub fn parse_module_checked<'input>(
@@ -342,17 +362,6 @@ pub fn parse_module_checked<'input>(
     }
 }
 
-fn lex_with_span<'input>(
-    text: &'input str,
-) -> Result<Vec<(Token<'input>, logos::Span)>, TokenError> {
-    let lexer = Token::lexer(text);
-    let mut result = Vec::new();
-    for (token, span) in lexer.spanned() {
-        result.push((token?, span));
-    }
-    Ok(result)
-}
-
 fn module<'a, 'input>(stream: &mut PtxParser<'a, 'input>) -> PResult<ast::Module<'input>> {
     (
         version,

zluda_dump/Cargo.toml

@@ -2,7 +2,7 @@
 name = "zluda_dump"
 version = "0.0.0"
 authors = ["Andrzej Janik <vosen@vosen.pl>"]
-edition = "2018"
+edition = "2021"
 
 [lib]
 name = "zluda_dump"
@@ -10,6 +10,7 @@ crate-type = ["cdylib"]
 
 [dependencies]
 ptx = { path = "../ptx" }
+ptx_parser = { path = "../ptx_parser" }
 lz4-sys = "1.9"
 regex = "1.4"
 dynasm = "1.2"

zluda_dump/src/format.rs

@@ -1,11 +1,10 @@
+use cuda_base::cuda_derive_display_trait;
 use std::{
     ffi::{c_void, CStr},
     fmt::LowerHex,
     mem, ptr, slice,
 };
 
-use cuda_base::cuda_derive_display_trait;
-
 pub(crate) trait CudaDisplay {
     fn write(
         &self,
@@ -494,6 +493,59 @@ impl CudaDisplay
     }
 }
 
+impl CudaDisplay for cuda_types::CUgraphNodeParams_st {
+    fn write(
+        &self,
+        _fn_name: &'static str,
+        _index: usize,
+        _writer: &mut (impl std::io::Write + ?Sized),
+    ) -> std::io::Result<()> {
+        todo!()
+    }
+}
+
+impl CudaDisplay for cuda_types::CUlaunchConfig_st {
+    fn write(
+        &self,
+        _fn_name: &'static str,
+        _index: usize,
+        _writer: &mut (impl std::io::Write + ?Sized),
+    ) -> std::io::Result<()> {
+        todo!()
+    }
+}
+
+impl CudaDisplay for cuda_types::CUeglFrame_st {
+    fn write(
+        &self,
+        _fn_name: &'static str,
+        _index: usize,
+        _writer: &mut (impl std::io::Write + ?Sized),
+    ) -> std::io::Result<()> {
+        todo!()
+    }
+}
+
+impl CudaDisplay for cuda_types::CUdevResource_st {
+    fn write(
+        &self,
+        _fn_name: &'static str,
+        _index: usize,
+        _writer: &mut (impl std::io::Write + ?Sized),
+    ) -> std::io::Result<()> {
+        todo!()
+    }
+}
+impl CudaDisplay for cuda_types::CUlaunchAttribute_st {
+    fn write(
+        &self,
+        _fn_name: &'static str,
+        _index: usize,
+        _writer: &mut (impl std::io::Write + ?Sized),
+    ) -> std::io::Result<()> {
+        todo!()
+    }
+}
 impl<T: CudaDisplay> CudaDisplay for *mut T {
     fn write(
         &self,
@@ -585,27 +637,7 @@ pub fn write_cuGraphKernelNodeGetAttribute(
     CudaDisplay::write(&hNode, "cuGraphKernelNodeGetAttribute", 0, writer)?;
     writer.write_all(b", attr: ")?;
     CudaDisplay::write(&attr, "cuGraphKernelNodeGetAttribute", 1, writer)?;
-    match attr {
+    write_launch_attribute(writer, "cuGraphKernelNodeGetAttribute", 2, attr, value_out)?;
-        cuda_types::CUkernelNodeAttrID::CU_KERNEL_NODE_ATTRIBUTE_ACCESS_POLICY_WINDOW => {
-            writer.write_all(b", value_out: ")?;
-            CudaDisplay::write(
-                unsafe { &(*value_out).accessPolicyWindow },
-                "cuGraphKernelNodeGetAttribute",
-                2,
-                writer,
-            )?;
-        }
-        cuda_types::CUkernelNodeAttrID::CU_KERNEL_NODE_ATTRIBUTE_COOPERATIVE => {
-            writer.write_all(b", value_out: ")?;
-            CudaDisplay::write(
-                unsafe { &(*value_out).cooperative },
-                "cuGraphKernelNodeGetAttribute",
-                2,
-                writer,
-            )?;
-        }
-        _ => return writer.write_all(b", ...) "),
-    }
     writer.write_all(b") ")
 }
 
@@ -630,28 +662,73 @@ pub fn write_cuStreamGetAttribute(
     CudaDisplay::write(&hStream, "cuStreamGetAttribute", 0, writer)?;
     writer.write_all(b", attr: ")?;
     CudaDisplay::write(&attr, "cuStreamGetAttribute", 1, writer)?;
-    match attr {
+    write_launch_attribute(writer, "cuStreamGetAttribute", 2, attr, value_out)?;
-        cuda_types::CUstreamAttrID::CU_STREAM_ATTRIBUTE_ACCESS_POLICY_WINDOW => {
+    writer.write_all(b") ")
+}
+
+fn write_launch_attribute(
+    writer: &mut (impl std::io::Write + ?Sized),
+    fn_name: &'static str,
+    index: usize,
+    attribute: cuda_types::CUlaunchAttributeID,
+    value_out: *mut cuda_types::CUstreamAttrValue,
+) -> std::io::Result<()> {
+    match attribute {
+        cuda_types::CUlaunchAttributeID::CU_LAUNCH_ATTRIBUTE_ACCESS_POLICY_WINDOW => {
             writer.write_all(b", value_out: ")?;
             CudaDisplay::write(
                 unsafe { &(*value_out).accessPolicyWindow },
-                "cuStreamGetAttribute",
+                fn_name,
-                2,
+                index,
                 writer,
-            )?;
+            )
         }
-        cuda_types::CUstreamAttrID::CU_STREAM_ATTRIBUTE_SYNCHRONIZATION_POLICY => {
+        cuda_types::CUlaunchAttributeID::CU_LAUNCH_ATTRIBUTE_COOPERATIVE => {
             writer.write_all(b", value_out: ")?;
-            CudaDisplay::write(
+            CudaDisplay::write(unsafe { &(*value_out).cooperative }, fn_name, index, writer)
-                unsafe { &(*value_out).syncPolicy },
-                "cuStreamGetAttribute",
-                2,
-                writer,
-            )?;
         }
-        _ => return writer.write_all(b", ...) "),
+        cuda_types::CUlaunchAttributeID::CU_LAUNCH_ATTRIBUTE_SYNCHRONIZATION_POLICY => {
+            writer.write_all(b", value_out: ")?;
+            CudaDisplay::write(unsafe { &(*value_out).syncPolicy }, fn_name, index, writer)
+        }
+        cuda_types::CUlaunchAttributeID::CU_LAUNCH_ATTRIBUTE_CLUSTER_DIMENSION => {
+            writer.write_all(b", value_out: ")?;
+            CudaDisplay::write(unsafe { &(*value_out).clusterDim }, fn_name, index, writer)
+        }
+        cuda_types::CUlaunchAttributeID::CU_LAUNCH_ATTRIBUTE_CLUSTER_SCHEDULING_POLICY_PREFERENCE => {
+            writer.write_all(b", value_out: ")?;
+            CudaDisplay::write(unsafe { &(*value_out).clusterSchedulingPolicyPreference }, fn_name, index, writer)
+        }
+        cuda_types::CUlaunchAttributeID::CU_LAUNCH_ATTRIBUTE_PROGRAMMATIC_STREAM_SERIALIZATION => {
+            writer.write_all(b", value_out: ")?;
+            CudaDisplay::write(unsafe { &(*value_out).programmaticStreamSerializationAllowed }, fn_name, index, writer)
+        }
+        cuda_types::CUlaunchAttributeID::CU_LAUNCH_ATTRIBUTE_PROGRAMMATIC_EVENT => {
+            writer.write_all(b", value_out: ")?;
+            CudaDisplay::write(unsafe { &(*value_out).programmaticEvent }, fn_name, index, writer)
+        }
+        cuda_types::CUlaunchAttributeID::CU_LAUNCH_ATTRIBUTE_PRIORITY => {
+            writer.write_all(b", value_out: ")?;
+            CudaDisplay::write(unsafe { &(*value_out).priority }, fn_name, index, writer)
+        }
+        cuda_types::CUlaunchAttributeID::CU_LAUNCH_ATTRIBUTE_MEM_SYNC_DOMAIN_MAP => {
+            writer.write_all(b", value_out: ")?;
+            CudaDisplay::write(unsafe { &(*value_out).memSyncDomainMap }, fn_name, index, writer)
+        }
+        cuda_types::CUlaunchAttributeID::CU_LAUNCH_ATTRIBUTE_MEM_SYNC_DOMAIN => {
+            writer.write_all(b", value_out: ")?;
+            CudaDisplay::write(unsafe { &(*value_out).memSyncDomain }, fn_name, index, writer)
+        }
+        cuda_types::CUlaunchAttributeID::CU_LAUNCH_ATTRIBUTE_LAUNCH_COMPLETION_EVENT => {
+            writer.write_all(b", value_out: ")?;
+            CudaDisplay::write(unsafe { &(*value_out).launchCompletionEvent }, fn_name, index, writer)
+        }
+        cuda_types::CUlaunchAttributeID::CU_LAUNCH_ATTRIBUTE_DEVICE_UPDATABLE_KERNEL_NODE => {
+            writer.write_all(b", value_out: ")?;
+            CudaDisplay::write(unsafe { &(*value_out).deviceUpdatableKernelNode }, fn_name, index, writer)
+        }
+        _ => writer.write_all(b", ... "),
     }
-    writer.write_all(b") ")
 }
 
 #[allow(non_snake_case)]
@@ -737,7 +814,17 @@ cuda_derive_display_trait!(
         CUstreamBatchMemOpParams_union_CUstreamMemOpWaitValueParams_st,
         CUstreamBatchMemOpParams_union_CUstreamMemOpWriteValueParams_st,
         CUuuid_st,
-        HGPUNV
+        HGPUNV,
+        EGLint,
+        EGLSyncKHR,
+        EGLImageKHR,
+        EGLStreamKHR,
+        CUasyncNotificationInfo_st,
+        CUgraphNodeParams_st,
+        CUeglFrame_st,
+        CUdevResource_st,
+        CUlaunchAttribute_st,
+        CUlaunchConfig_st
     ],
     [
         cuCtxCreate_v3,

zluda_dump/src/lib.rs

@@ -1,6 +1,4 @@
-use cuda_types::{
+use cuda_types::*;
-    CUdevice, CUdevice_attribute, CUfunction, CUjit_option, CUmodule, CUresult, CUuuid,
-};
 use paste::paste;
 use side_by_side::CudaDynamicFns;
 use std::io;
@@ -205,10 +203,10 @@ impl Settings {
             }
         };
         let libcuda_path = match env::var("ZLUDA_CUDA_LIB") {
-            Err(env::VarError::NotPresent) => os::LIBCUDA_DEFAULT_PATH.to_owned(),
+            Err(env::VarError::NotPresent) => os::LIBCUDA_DEFAULT_PATH.to_string(),
             Err(e) => {
                 logger.log(log::LogEntry::ErrorBox(Box::new(e) as _));
-                os::LIBCUDA_DEFAULT_PATH.to_owned()
+                os::LIBCUDA_DEFAULT_PATH.to_string()
             }
             Ok(env_string) => env_string,
         };

zluda_dump/src/os_unix.rs

@@ -1,8 +1,8 @@
-use crate::cuda::CUuuid;
+use cuda_types::CUuuid;
 use std::ffi::{c_void, CStr, CString};
 use std::mem;
 
-pub(crate) const LIBCUDA_DEFAULT_PATH: &'static str = b"/usr/lib/x86_64-linux-gnu/libcuda.so.1\0";
+pub(crate) const LIBCUDA_DEFAULT_PATH: &str = "/usr/lib/x86_64-linux-gnu/libcuda.so.1";
 
 pub unsafe fn load_library(libcuda_path: &str) -> *mut c_void {
     let libcuda_path = CString::new(libcuda_path).unwrap();

zluda_dump/src/trace.rs

@@ -1,6 +1,3 @@
-use ptx::{ast::PtxError, Token};
-use ptx::{DisplayParseError, ModuleParserExt};
-
 use crate::{dark_api, log, Settings};
 use cuda_types::CUmodule;
 use std::{
@@ -172,7 +169,7 @@ impl StateTracker {
         submodule_index: Option<usize>,
         module_text: &str,
     ) {
-        let (_ast, errors) = ptx::ModuleParser::parse_unchecked(module_text);
+        let errors = ptx_parser::parse_for_errors(module_text);
         if !errors.is_empty() {
             fn_logger.log(log::LogEntry::ModuleParsingError(
                 DumpWriter::get_file_name(module_index, version, submodule_index, "log"),
@@ -232,7 +229,7 @@ impl DumpWriter {
         module_index: usize,
         version: Option<usize>,
         submodule_index: Option<usize>,
-        errors: &[ptx::ParseError<usize, Token<'input>, PtxError>],
+        errors: &[ptx_parser::PtxError<'input>],
     ) -> io::Result<()> {
         let mut log_file = match &self.dump_dir {
             None => return Ok(()),
@@ -246,8 +243,7 @@ impl DumpWriter {
         ));
         let mut file = File::create(log_file)?;
         for error in errors {
-            let pretty_print_error = unsafe { DisplayParseError::new(error, module_text) };
+            writeln!(file, "{}", error)?;
-            writeln!(file, "{}", pretty_print_error)?;
         }
         Ok(())
     }