ggml-cuda : fix CDNA2 compute capability constant for gfx90a (MI210)

GGML_CUDA_CC_CDNA2 was set to 0x910 which corresponds to gfx910 (RDNA3),
not gfx90a (CDNA2/MI210). This caused CDNA2 GPUs to be misidentified,
skipping CDNA2-specific code paths such as MFMA acc register renaming.

Fix by setting the constant to 0x90a to match the actual gfx90a ISA.

Author: aviallon

ggml/src/ggml-cuda/common.cuh

@@ -65,7 +65,7 @@
 #define GGML_CUDA_CC_VEGA       (GGML_CUDA_CC_OFFSET_AMD + 0x900)  // Vega56/64, minimum for fp16 dual issue
 #define GGML_CUDA_CC_VEGA20     (GGML_CUDA_CC_OFFSET_AMD + 0x906)  // MI50/Radeon VII, minimum for dp4a
 #define GGML_CUDA_CC_CDNA1      (GGML_CUDA_CC_OFFSET_AMD + 0x908)  // MI100, minimum for MFMA, acc registers
-#define GGML_CUDA_CC_CDNA2      (GGML_CUDA_CC_OFFSET_AMD + 0x910)  // MI210, minimum acc register renameing
+#define GGML_CUDA_CC_CDNA2      (GGML_CUDA_CC_OFFSET_AMD + 0x90a)  // MI210 (gfx90a), minimum acc register renaming
 #define GGML_CUDA_CC_CDNA3      (GGML_CUDA_CC_OFFSET_AMD + 0x942)  // MI300
 
 // RDNA removes MFMA, dp4a, xnack, acc registers, wave size is 32