[SYCL] Add missing supported AMDGPU architectures to SYCL (#15723)

The three added `gfx7` architectures were previously omitted from the
list of supported ones for SYCL targeting AMDGPU.
There is no rational for them to be excluded rather than a potential
mistake when all the rest were added, so this PR adds them in.
All of these exactly match the LLVM AMDGPU support docs here:
https://llvm.org/docs/AMDGPUUsage.html#processors.
This also makes reusing the current `OffloadArch` enum in clang Driver
easier with no concern of whether we have to filter these out for SYCL.

Author: Georgi Mirazchiyski

clang/lib/Driver/ToolChains/SYCL.cpp

@@ -1328,6 +1328,9 @@ StringRef SYCL::gen::resolveGenDevice(StringRef DeviceName) {
           .Case("amd_gpu_gfx700", "gfx700")
           .Case("amd_gpu_gfx701", "gfx701")
           .Case("amd_gpu_gfx702", "gfx702")
+          .Case("amd_gpu_gfx703", "gfx703")
+          .Case("amd_gpu_gfx704", "gfx704")
+          .Case("amd_gpu_gfx705", "gfx705")
           .Case("amd_gpu_gfx801", "gfx801")
           .Case("amd_gpu_gfx802", "gfx802")
           .Case("amd_gpu_gfx803", "gfx803")
@@ -1415,6 +1418,9 @@ SmallString<64> SYCL::gen::getGenDeviceMacro(StringRef DeviceName) {
                       .Case("gfx700", "AMD_GPU_GFX700")
                       .Case("gfx701", "AMD_GPU_GFX701")
                       .Case("gfx702", "AMD_GPU_GFX702")
+                      .Case("gfx703", "AMD_GPU_GFX703")
+                      .Case("gfx704", "AMD_GPU_GFX704")
+                      .Case("gfx705", "AMD_GPU_GFX705")
                       .Case("gfx801", "AMD_GPU_GFX801")
                       .Case("gfx802", "AMD_GPU_GFX802")
                       .Case("gfx803", "AMD_GPU_GFX803")

clang/test/Driver/sycl-device-traits-macros-amdgcn.cpp

@@ -9,6 +9,15 @@
 // RUN: %clangxx -fsycl -nogpulib -fsycl-targets=amd_gpu_gfx702 \
 // RUN:   -fsycl-libspirv-path=%S/Inputs/SYCL/libspirv.bc -### %s 2>&1 | \
 // RUN:   FileCheck %s --check-prefix=CHECK-SYCL-AMDGCN-AMD-AMDHSA-DEVICE-TRIPLE
+// RUN: %clangxx -fsycl -nogpulib -fsycl-targets=amd_gpu_gfx703 \
+// RUN:   -fsycl-libspirv-path=%S/Inputs/SYCL/libspirv.bc -### %s 2>&1 | \
+// RUN:   FileCheck %s --check-prefix=CHECK-SYCL-AMDGCN-AMD-AMDHSA-DEVICE-TRIPLE
+// RUN: %clangxx -fsycl -nogpulib -fsycl-targets=amd_gpu_gfx704 \
+// RUN:   -fsycl-libspirv-path=%S/Inputs/SYCL/libspirv.bc -### %s 2>&1 | \
+// RUN:   FileCheck %s --check-prefix=CHECK-SYCL-AMDGCN-AMD-AMDHSA-DEVICE-TRIPLE
+// RUN: %clangxx -fsycl -nogpulib -fsycl-targets=amd_gpu_gfx705 \
+// RUN:   -fsycl-libspirv-path=%S/Inputs/SYCL/libspirv.bc -### %s 2>&1 | \
+// RUN:   FileCheck %s --check-prefix=CHECK-SYCL-AMDGCN-AMD-AMDHSA-DEVICE-TRIPLE
 // RUN: %clangxx -fsycl -nogpulib -fsycl-targets=amd_gpu_gfx801 \
 // RUN:   -fsycl-libspirv-path=%S/Inputs/SYCL/libspirv.bc -### %s 2>&1 | \
 // RUN:   FileCheck %s --check-prefix=CHECK-SYCL-AMDGCN-AMD-AMDHSA-DEVICE-TRIPLE
@@ -120,6 +129,15 @@
 // RUN: %clangxx -fsycl -nogpulib -fsycl-targets=amdgcn-amd-amdhsa -Xsycl-target-backend --offload-arch=gfx702 \
 // RUN:   -fsycl-libspirv-path=%S/Inputs/SYCL/libspirv.bc -### %s 2>&1 | \
 // RUN:   FileCheck %s --check-prefix=CHECK-SYCL-AMDGCN-AMD-AMDHSA-OFFLOAD-ARCH
+// RUN: %clangxx -fsycl -nogpulib -fsycl-targets=amdgcn-amd-amdhsa -Xsycl-target-backend --offload-arch=gfx703 \
+// RUN:   -fsycl-libspirv-path=%S/Inputs/SYCL/libspirv.bc -### %s 2>&1 | \
+// RUN:   FileCheck %s --check-prefix=CHECK-SYCL-AMDGCN-AMD-AMDHSA-OFFLOAD-ARCH
+// RUN: %clangxx -fsycl -nogpulib -fsycl-targets=amdgcn-amd-amdhsa -Xsycl-target-backend --offload-arch=gfx704 \
+// RUN:   -fsycl-libspirv-path=%S/Inputs/SYCL/libspirv.bc -### %s 2>&1 | \
+// RUN:   FileCheck %s --check-prefix=CHECK-SYCL-AMDGCN-AMD-AMDHSA-OFFLOAD-ARCH
+// RUN: %clangxx -fsycl -nogpulib -fsycl-targets=amdgcn-amd-amdhsa -Xsycl-target-backend --offload-arch=gfx705 \
+// RUN:   -fsycl-libspirv-path=%S/Inputs/SYCL/libspirv.bc -### %s 2>&1 | \
+// RUN:   FileCheck %s --check-prefix=CHECK-SYCL-AMDGCN-AMD-AMDHSA-OFFLOAD-ARCH
 // RUN: %clangxx -fsycl -nogpulib -fsycl-targets=amdgcn-amd-amdhsa -Xsycl-target-backend --offload-arch=gfx801 \
 // RUN:   -fsycl-libspirv-path=%S/Inputs/SYCL/libspirv.bc -### %s 2>&1 | \
 // RUN:   FileCheck %s --check-prefix=CHECK-SYCL-AMDGCN-AMD-AMDHSA-OFFLOAD-ARCH

clang/test/Driver/sycl-oneapi-gpu-amdgpu.cpp

@@ -9,6 +9,15 @@
 // RUN: %clangxx -fsycl -nogpulib -fsycl-targets=amd_gpu_gfx702 \
 // RUN:   -fsycl-libspirv-path=%S/Inputs/SYCL/libspirv.bc -### %s 2>&1 | \
 // RUN:   FileCheck %s --check-prefixes=DEVICE_AMD,MACRO_AMD -DDEV_STR=gfx702 -DMAC_STR=GFX702
+// RUN: %clangxx -fsycl -nogpulib -fsycl-targets=amd_gpu_gfx703 \
+// RUN:   -fsycl-libspirv-path=%S/Inputs/SYCL/libspirv.bc -### %s 2>&1 | \
+// RUN:   FileCheck %s --check-prefixes=DEVICE_AMD,MACRO_AMD -DDEV_STR=gfx703 -DMAC_STR=GFX703
+// RUN: %clangxx -fsycl -nogpulib -fsycl-targets=amd_gpu_gfx704 \
+// RUN:   -fsycl-libspirv-path=%S/Inputs/SYCL/libspirv.bc -### %s 2>&1 | \
+// RUN:   FileCheck %s --check-prefixes=DEVICE_AMD,MACRO_AMD -DDEV_STR=gfx704 -DMAC_STR=GFX704
+// RUN: %clangxx -fsycl -nogpulib -fsycl-targets=amd_gpu_gfx705 \
+// RUN:   -fsycl-libspirv-path=%S/Inputs/SYCL/libspirv.bc -### %s 2>&1 | \
+// RUN:   FileCheck %s --check-prefixes=DEVICE_AMD,MACRO_AMD -DDEV_STR=gfx705 -DMAC_STR=GFX705
 // RUN: %clangxx -fsycl -nogpulib -fsycl-targets=amd_gpu_gfx801 \
 // RUN:   -fsycl-libspirv-path=%S/Inputs/SYCL/libspirv.bc -### %s 2>&1 | \
 // RUN:   FileCheck %s --check-prefixes=DEVICE_AMD,MACRO_AMD -DDEV_STR=gfx801 -DMAC_STR=GFX801

llvm/include/llvm/SYCLLowerIR/DeviceConfigFile.td

@@ -320,6 +320,9 @@ def : HipTargetInfo<"amd_gpu_gfx906", !listconcat(HipMinAspects, AllUSMAspects),
 def : HipTargetInfo<"amd_gpu_gfx700", HipMinAspects, HipSubgroupSizesGCN2>;
 def : HipTargetInfo<"amd_gpu_gfx701", HipMinAspects, HipSubgroupSizesGCN2>;
 def : HipTargetInfo<"amd_gpu_gfx702", HipMinAspects, HipSubgroupSizesGCN2>;
+def : HipTargetInfo<"amd_gpu_gfx703", HipMinAspects, HipSubgroupSizesGCN2>;
+def : HipTargetInfo<"amd_gpu_gfx704", HipMinAspects, HipSubgroupSizesGCN2>;
+def : HipTargetInfo<"amd_gpu_gfx705", HipMinAspects, HipSubgroupSizesGCN2>;
 def : HipTargetInfo<"amd_gpu_gfx801", HipMinAspects, HipSubgroupSizesGCN3>;
 def : HipTargetInfo<"amd_gpu_gfx802", HipMinAspects, HipSubgroupSizesGCN3>;
 def : HipTargetInfo<"amd_gpu_gfx803", HipMinAspects, HipSubgroupSizesGCN3>;

sycl/doc/UsersManual.md

@@ -95,6 +95,9 @@ and not recommended to use in production environment.
     * amd_gpu_gfx700 - AMD GCN GFX7 (Sea Islands (CI)) architecture
     * amd_gpu_gfx701 - AMD GCN GFX7 (Sea Islands (CI)) architecture
     * amd_gpu_gfx702 - AMD GCN GFX7 (Sea Islands (CI)) architecture
+    * amd_gpu_gfx703 - AMD GCN GFX7 (Sea Islands (CI)) architecture
+    * amd_gpu_gfx704 - AMD GCN GFX7 (Sea Islands (CI)) architecture
+    * amd_gpu_gfx705 - AMD GCN GFX7 (Sea Islands (CI)) architecture
     * amd_gpu_gfx801 - AMD GCN GFX8 (Volcanic Islands (VI)) architecture
     * amd_gpu_gfx802 - AMD GCN GFX8 (Volcanic Islands (VI)) architecture
     * amd_gpu_gfx803 - AMD GCN GFX8 (Volcanic Islands (VI)) architecture
@@ -104,13 +107,13 @@ and not recommended to use in production environment.
     * amd_gpu_gfx902 - AMD GCN GFX9 (Vega) architecture
     * amd_gpu_gfx904 - AMD GCN GFX9 (Vega) architecture
     * amd_gpu_gfx906 - AMD GCN GFX9 (Vega) architecture
-    * amd_gpu_gfx908 - AMD GCN GFX9 (Vega) architecture
+    * amd_gpu_gfx908 - AMD GCN GFX9 (CDNA1) architecture
     * amd_gpu_gfx909 - AMD GCN GFX9 (Vega) architecture
-    * amd_gpu_gfx90a - AMD GCN GFX9 (Vega) architecture
+    * amd_gpu_gfx90a - AMD GCN GFX9 (CDNA2) architecture
     * amd_gpu_gfx90c - AMD GCN GFX9 (Vega) architecture
-    * amd_gpu_gfx940 - AMD GCN GFX9 (Vega) architecture
-    * amd_gpu_gfx941 - AMD GCN GFX9 (Vega) architecture
-    * amd_gpu_gfx942 - AMD GCN GFX9 (Vega) architecture
+    * amd_gpu_gfx940 - AMD GCN GFX9 (CDNA3) architecture
+    * amd_gpu_gfx941 - AMD GCN GFX9 (CDNA3) architecture
+    * amd_gpu_gfx942 - AMD GCN GFX9 (CDNA3) architecture
     * amd_gpu_gfx1010 - AMD GCN GFX10.1 (RDNA 1) architecture
     * amd_gpu_gfx1011 - AMD GCN GFX10.1 (RDNA 1) architecture
     * amd_gpu_gfx1012 - AMD GCN GFX10.1 (RDNA 1) architecture

sycl/doc/design/DeviceIf.md

@@ -88,6 +88,9 @@ recognizes:
 * `amd_gpu_gfx700`
 * `amd_gpu_gfx701`
 * `amd_gpu_gfx702`
+* `amd_gpu_gfx703`
+* `amd_gpu_gfx704`
+* `amd_gpu_gfx705`
 * `amd_gpu_gfx801`
 * `amd_gpu_gfx802`
 * `amd_gpu_gfx803`
@@ -187,6 +190,9 @@ one of the following corresponding C++ macro names:
 * `__SYCL_TARGET_AMD_GPU_GFX700__`
 * `__SYCL_TARGET_AMD_GPU_GFX701__`
 * `__SYCL_TARGET_AMD_GPU_GFX702__`
+* `__SYCL_TARGET_AMD_GPU_GFX703__`
+* `__SYCL_TARGET_AMD_GPU_GFX704__`
+* `__SYCL_TARGET_AMD_GPU_GFX705__`
 * `__SYCL_TARGET_AMD_GPU_GFX801__`
 * `__SYCL_TARGET_AMD_GPU_GFX802__`
 * `__SYCL_TARGET_AMD_GPU_GFX803__`

sycl/doc/extensions/experimental/sycl_ext_oneapi_device_architecture.asciidoc

@@ -551,6 +551,9 @@ a|
 amd_gpu_gfx700
 amd_gpu_gfx701
 amd_gpu_gfx702
+amd_gpu_gfx703
+amd_gpu_gfx704
+amd_gpu_gfx705
 ----
 |-
 |AMD GCN 2.0 architecture.

sycl/include/sycl/ext/oneapi/experimental/architectures.def

@@ -114,6 +114,9 @@ __SYCL_ARCHITECTURE(nvidia_gpu_sm_90a, 0x01000000000090a0)
 __SYCL_ARCHITECTURE(amd_gpu_gfx700, 0x0200000000070000)
 __SYCL_ARCHITECTURE(amd_gpu_gfx701, 0x0200000000070100)
 __SYCL_ARCHITECTURE(amd_gpu_gfx702, 0x0200000000070200)
+__SYCL_ARCHITECTURE(amd_gpu_gfx703, 0x0200000000070300)
+__SYCL_ARCHITECTURE(amd_gpu_gfx704, 0x0200000000070400)
+__SYCL_ARCHITECTURE(amd_gpu_gfx705, 0x0200000000070500)
 __SYCL_ARCHITECTURE(amd_gpu_gfx801, 0x0200000000080100)
 __SYCL_ARCHITECTURE(amd_gpu_gfx802, 0x0200000000080200)
 __SYCL_ARCHITECTURE(amd_gpu_gfx803, 0x0200000000080300)

sycl/include/sycl/ext/oneapi/experimental/device_architecture.hpp

@@ -203,6 +203,15 @@ static constexpr ext::oneapi::experimental::architecture
 #ifndef __SYCL_TARGET_AMD_GPU_GFX702__
 #define __SYCL_TARGET_AMD_GPU_GFX702__ 0
 #endif
+#ifndef __SYCL_TARGET_AMD_GPU_GFX703__
+#define __SYCL_TARGET_AMD_GPU_GFX703__ 0
+#endif
+#ifndef __SYCL_TARGET_AMD_GPU_GFX704__
+#define __SYCL_TARGET_AMD_GPU_GFX704__ 0
+#endif
+#ifndef __SYCL_TARGET_AMD_GPU_GFX705__
+#define __SYCL_TARGET_AMD_GPU_GFX705__ 0
+#endif
 #ifndef __SYCL_TARGET_AMD_GPU_GFX801__
 #define __SYCL_TARGET_AMD_GPU_GFX801__ 0
 #endif
@@ -360,6 +369,9 @@ static constexpr bool is_allowable_aot_mode =
     (__SYCL_TARGET_AMD_GPU_GFX700__ == 1) ||
     (__SYCL_TARGET_AMD_GPU_GFX701__ == 1) ||
     (__SYCL_TARGET_AMD_GPU_GFX702__ == 1) ||
+    (__SYCL_TARGET_AMD_GPU_GFX703__ == 1) ||
+    (__SYCL_TARGET_AMD_GPU_GFX704__ == 1) ||
+    (__SYCL_TARGET_AMD_GPU_GFX705__ == 1) ||
     (__SYCL_TARGET_AMD_GPU_GFX801__ == 1) ||
     (__SYCL_TARGET_AMD_GPU_GFX802__ == 1) ||
     (__SYCL_TARGET_AMD_GPU_GFX803__ == 1) ||
@@ -538,6 +550,15 @@ get_current_architecture_aot() {
 #if __SYCL_TARGET_AMD_GPU_GFX702__
   return ext::oneapi::experimental::architecture::amd_gpu_gfx702;
 #endif
+#if __SYCL_TARGET_AMD_GPU_GFX703__
+  return ext::oneapi::experimental::architecture::amd_gpu_gfx703;
+#endif
+#if __SYCL_TARGET_AMD_GPU_GFX704__
+  return ext::oneapi::experimental::architecture::amd_gpu_gfx704;
+#endif
+#if __SYCL_TARGET_AMD_GPU_GFX705__
+  return ext::oneapi::experimental::architecture::amd_gpu_gfx705;
+#endif
 #if __SYCL_TARGET_AMD_GPU_GFX801__
   return ext::oneapi::experimental::architecture::amd_gpu_gfx801;
 #endif

sycl/source/detail/device_info.hpp

@@ -626,6 +626,9 @@ constexpr std::pair<const char *, oneapi_exp_arch> NvidiaAmdGPUArchitectures[] =
         {"9.0", oneapi_exp_arch::nvidia_gpu_sm_90},
         {"gfx701", oneapi_exp_arch::amd_gpu_gfx701},
         {"gfx702", oneapi_exp_arch::amd_gpu_gfx702},
+        {"gfx703", oneapi_exp_arch::amd_gpu_gfx703},
+        {"gfx704", oneapi_exp_arch::amd_gpu_gfx704},
+        {"gfx705", oneapi_exp_arch::amd_gpu_gfx705},
         {"gfx801", oneapi_exp_arch::amd_gpu_gfx801},
         {"gfx802", oneapi_exp_arch::amd_gpu_gfx802},
         {"gfx803", oneapi_exp_arch::amd_gpu_gfx803},