diff --git a/tinygrad/runtime/ops_amd.py b/tinygrad/runtime/ops_amd.py index 5a311cc08b..b47ee3259f 100644 --- a/tinygrad/runtime/ops_amd.py +++ b/tinygrad/runtime/ops_amd.py @@ -151,7 +151,7 @@ class AMDComputeQueue(HWQueue): # CUs you want to by disabling other CUs via bits in regCOMPUTE_STATIC_THREAD_MGMT_SE and trace even kernels that only have one wavefront. self.wreg(self.gc.regSQ_THREAD_TRACE_MASK, wtype_include=self.soc.SQ_TT_WTYPE_INCLUDE_CS_BIT, simd_sel=0, wgp_sel=0, sa_sel=0) reg_include = self.soc.SQ_TT_TOKEN_MASK_SQDEC_BIT | self.soc.SQ_TT_TOKEN_MASK_SHDEC_BIT | self.soc.SQ_TT_TOKEN_MASK_GFXUDEC_BIT | \ - self.soc.SQ_TT_TOKEN_MASK_COMP_BIT | self.soc.SQ_TT_TOKEN_MASK_CONTEXT_BIT | self.soc.SQ_TT_TOKEN_MASK_CONTEXT_BIT + self.soc.SQ_TT_TOKEN_MASK_COMP_BIT | self.soc.SQ_TT_TOKEN_MASK_CONTEXT_BIT token_exclude = 1 << self.soc.SQ_TT_TOKEN_EXCLUDE_PERF_SHIFT if not (se_mask >> se) & 0b1: token_exclude |= 1 << self.soc.SQ_TT_TOKEN_EXCLUDE_VMEMEXEC_SHIFT | 1 << self.soc.SQ_TT_TOKEN_EXCLUDE_ALUEXEC_SHIFT | \