From eff1491f95c78436e33a7ad4d9b2c02dcdebcc76 Mon Sep 17 00:00:00 2001
From: Nikita Sirgienko <nikita.sirgienko@intel.com>
Date: Mon, 5 Feb 2024 13:04:36 +0100
Subject: [PATCH] Build: Cycles: Backport a DPC++ fix for caching of GPU
 binaries

Since the addition of Meteor Lake binaries, prebuilt GPU binaries
are now stored as fatbinaries. When running on a platform for which
prebuilt binaries are lacking or considered incompatible, the DPC++
SYCL runtime caching logic failed storing the (re)compiled
compatible version. This patch to DPC++ SYCL runtime fixes it.
---
 .../build_environment/patches/dpcpp.diff      | 90 +++++++++++++++++++
 1 file changed, 90 insertions(+)

diff --git a/build_files/build_environment/patches/dpcpp.diff b/build_files/build_environment/patches/dpcpp.diff
index a87d1412caf..5605e550397 100644
--- a/build_files/build_environment/patches/dpcpp.diff
+++ b/build_files/build_environment/patches/dpcpp.diff
@@ -187,3 +187,93 @@ index 17eeaafae194..09e6d2217aaa 100644
    return SYCLDeviceImages;
  }
  
+diff --git a/sycl/source/detail/pi.cpp b/sycl/source/detail/pi.cpp
+index fb228cd85979..94e8438ee964 100644
+--- a/sycl/source/detail/pi.cpp
++++ b/sycl/source/detail/pi.cpp
+@@ -635,45 +635,47 @@ static uint16_t getELFHeaderType(const unsigned char *ImgData, size_t ImgSize) {
+ RT::PiDeviceBinaryType getBinaryImageFormat(const unsigned char *ImgData,
+                                             size_t ImgSize) {
+   // Top-level magic numbers for the recognized binary image formats.
+-  struct {
+-    RT::PiDeviceBinaryType Fmt;
+-    const uint32_t Magic;
+-  } Fmts[] = {{PI_DEVICE_BINARY_TYPE_SPIRV, 0x07230203},
+-              {PI_DEVICE_BINARY_TYPE_LLVMIR_BITCODE, 0xDEC04342},
+-              // 'I', 'N', 'T', 'C' ; Intel native
+-              {PI_DEVICE_BINARY_TYPE_NATIVE, 0x43544E49}};
+-
+-  if (ImgSize >= sizeof(Fmts[0].Magic)) {
+-    detail::remove_const_t<decltype(Fmts[0].Magic)> Hdr = 0;
+-    std::copy(ImgData, ImgData + sizeof(Hdr), reinterpret_cast<char *>(&Hdr));
+-
+-    // Check headers for direct formats.
+-    for (const auto &Fmt : Fmts) {
+-      if (Hdr == Fmt.Magic)
+-        return Fmt.Fmt;
+-    }
+-
+-    // ELF e_type for recognized binary image formats.
+-    struct {
+-      RT::PiDeviceBinaryType Fmt;
+-      const uint16_t Magic;
+-    } ELFFmts[] = {{PI_DEVICE_BINARY_TYPE_NATIVE, 0xFF04},  // OpenCL executable
+-                   {PI_DEVICE_BINARY_TYPE_NATIVE, 0xFF12}}; // ZEBIN executable
+-
+-    // ELF files need to be parsed separately. The header type ends after 18
+-    // bytes.
+-    if (Hdr == 0x464c457F && ImgSize >= 18) {
+-      uint16_t HdrType = getELFHeaderType(ImgData, ImgSize);
+-      for (const auto &ELFFmt : ELFFmts) {
+-        if (HdrType == ELFFmt.Magic)
+-          return ELFFmt.Fmt;
+-      }
+-      // Newer ZEBIN format does not have a special header type, but can instead
+-      // be identified by having a required .ze_info section.
+-      if (checkELFSectionPresent(".ze_info", ImgData, ImgSize))
+-        return PI_DEVICE_BINARY_TYPE_NATIVE;
+-    }
++  auto MatchMagicNumber = [&](auto Number) {
++    return ImgSize >= sizeof(Number) &&
++           std::memcmp(ImgData, &Number, sizeof(Number)) == 0;
++  };
++
++  if (MatchMagicNumber(uint32_t{0x07230203}))
++    return PI_DEVICE_BINARY_TYPE_SPIRV;
++
++  if (MatchMagicNumber(uint32_t{0xDEC04342}))
++    return PI_DEVICE_BINARY_TYPE_LLVMIR_BITCODE;
++
++  if (MatchMagicNumber(uint32_t{0x43544E49}))
++    // 'I', 'N', 'T', 'C' ; Intel native
++    return PI_DEVICE_BINARY_TYPE_LLVMIR_BITCODE;
++
++  // Check for ELF format, size requirements include data we'll read in case of
++  // succesful match.
++  if (ImgSize >= 18 && MatchMagicNumber(uint32_t{0x464c457F})) {
++    uint16_t ELFHdrType = getELFHeaderType(ImgData, ImgSize);
++    if (ELFHdrType == 0xFF04)
++      // OpenCL executable.
++      return PI_DEVICE_BINARY_TYPE_NATIVE;
++
++    if (ELFHdrType == 0xFF12)
++      // ZEBIN executable.
++      return PI_DEVICE_BINARY_TYPE_NATIVE;
++
++    // Newer ZEBIN format does not have a special header type, but can instead
++    // be identified by having a required .ze_info section.
++    if (checkELFSectionPresent(".ze_info", ImgData, ImgSize))
++      return PI_DEVICE_BINARY_TYPE_NATIVE;
+   }
++
++  // "ar" format is used to pack binaries for multiple devices, e.g. via
++  //
++  //   -Xsycl-target-backend=spir64_gen "-device acm-g10,acm-g11"
++  //
++  // option.
++  if (MatchMagicNumber(std::array{'!', '<', 'a', 'r', 'c', 'h', '>', '\n'}))
++    return PI_DEVICE_BINARY_TYPE_NATIVE;
++
+   return PI_DEVICE_BINARY_TYPE_NONE;
+ }
+ 
-- 
2.30.2