[stdlib] Add gfx1030 RX 6900 support

bazel/common.MODULE.bazel

@@ -196,6 +196,7 @@ mojo.gpu_toolchains(
         "MI355": "mi355",
         "Navi": "radeon",
         "AMD Radeon Graphics": "radeon",
+        "AMD Radeon RX 6900 XT": "rx6900xt",
         "Apple M1": "m1",
         "Apple M2": "m2",
         "Apple M3": "m3",
@@ -218,6 +219,7 @@ mojo.gpu_toolchains(
         "rtx3090": "nvidia:86",
         "rtx4090": "nvidia:89",
         "rtx5090": "nvidia:120a",
+        "rx6900xt": "amdgpu:gfx1030",
         "m1": "metal:10",
         "m2": "metal:20",
         "m3": "metal:30",

mojo/docs/changelog.md

@@ -52,6 +52,8 @@ what we publish.
   alias EDEADLK = platform_map["EDEADLK", linux = 35, macos = 11]()
   ```
 
+- Added support for AMD RX 6900 XT consumer-grade GPU.
+
 ### Tooling changes
 
 ### ❌ Removed

mojo/stdlib/stdlib/gpu/host/info.mojo

@@ -932,6 +932,25 @@ fn _get_7600_target() -> _TargetType:
     ]
 
 
+fn _get_6900_target() -> _TargetType:
+    """
+    Creates an MLIR target configuration for AMD Radeon 6900 GPU.
+
+    Returns:
+        MLIR target configuration for 6900.
+    """
+
+    return __mlir_attr[
+        `#kgen.target<triple = "amdgcn-amd-amdhsa", `,
+        `arch = "gfx1030", `,
+        `features = "", `,
+        `data_layout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-p7:160:256:256:32-p8:128:128:128:48-p9:192:256:256:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5-G1-ni:7:8:9",`,
+        `index_bit_width = 64,`,
+        `simd_bit_width = 128`,
+        `> : !kgen.target`,
+    ]
+
+
 fn _get_780m_target() -> _TargetType:
     """
     Creates an MLIR target configuration for AMD Radeon 780m GPU.
@@ -1081,6 +1100,33 @@ alias Radeon7600 = GPUInfo(
     max_thread_block_size=1024,
 )
 
+alias Radeon6900 = Info(
+    name="Radeon 6900",
+    vendor=Vendor.AMD_GPU,
+    api="hip",
+    arch_name="gfx1102",
+    compile_options="",
+    compute=10.3,
+    version="RDNA2",
+    sm_count=60,
+    warp_size=32,
+    threads_per_sm=1024,
+    threads_per_warp=32,
+    warps_per_multiprocessor=32,  # 1024 threads per sm / 32 threads per warp = 32 warps per sm
+    threads_per_multiprocessor=1024,
+    thread_blocks_per_multiprocessor=2,
+    shared_memory_per_multiprocessor=32768,
+    register_file_size=32768,
+    register_allocation_unit_size=256,
+    allocation_granularity="warp",
+    max_registers_per_thread=255,
+    max_registers_per_block=32768,
+    max_blocks_per_multiprocessor=2,
+    shared_memory_allocation_unit_size=128,
+    warp_allocation_granularity=4,
+    max_thread_block_size=1024,
+)
+
 
 alias Radeon780m = GPUInfo(
     name="Radeon 780M",
@@ -1225,6 +1271,8 @@ struct GPUInfo(Stringable, Writable):
             return _get_mi300x_target()
         if self.name == "Radeon 780M":
             return _get_780m_target()
+        if self.name == "Radeon 6900":
+            return _get_6900_target()
         if self.name == "Radeon 7900":
             return _get_7900_target()
         if self.name == "Radeon 7800/7700":
@@ -1800,6 +1848,7 @@ fn _get_info_from_target[target_arch0: StaticString]() -> GPUInfo:
             # AMD
             StaticString("mi300x"),
             StaticString("gfx942"),
+            StaticString("gfx1030"),
             StaticString("gfx1100"),
             StaticString("gfx1101"),
             StaticString("gfx1102"),
@@ -1833,6 +1882,8 @@ fn _get_info_from_target[target_arch0: StaticString]() -> GPUInfo:
         return RTX5090
     elif target_arch == "gfx942" or target_arch == "mi300x":
         return MI300X
+    elif target_arch == "gfx1030":
+        return Radeon6900
     elif target_arch == "gfx1100":
         return Radeon7900
     elif target_arch == "gfx1101":