Update doc

hw/rtl/core/VX_commit.sv

@@ -177,7 +177,7 @@ module VX_commit import VX_gpu_pkg::*; #(
     // probably want to change this at some point
     // (i.e. pass a "don't count this towards pending instructions" signal down the pipeline)
     wire [`ISSUE_WIDTH-1:0] final_hmma;
-    // if this is a "ghost" commit generated from the tensor core, don't count
+    // if this is a "ghost" commit generated at the tensor core, don't count
     // toward committed
     wire [`ISSUE_WIDTH-1:0] tensor_ghost;
 `ifdef EXT_T_ENABLE

hw/rtl/core/VX_tensor_hopper_core.sv

@@ -158,9 +158,6 @@ module VX_tensor_hopper_core_block import VX_gpu_pkg::*; #(
             // mark as "ghost" commit.  This will prevent this commit from
             // decrementing from pending_instr buffer
             commit_if.data.tensor = 1'b1;
-            // eop is deliberately set so that we don't underflow the pending_instr
-            // buffer in VX_schedule.  An instruction is considered committed only
-            // when the eop bit is set to one (see VX_commit).
             // only the last ghost commit has eop set, which will trigger
             // scoreboard to clear out the busy bit.
             commit_if.data.eop    = writeback_last;

hw/rtl/fpu/VX_tensor_dpu.sv

@@ -290,7 +290,7 @@ module VX_tensor_threadgroup #(
         // d_col_sel: 1, 3, 1, 3
         //
         // substep 0:
-        // [ 0 x 2 x ]
+        // [ 0 x 2 x ] (0~3 is 'i', the dpu id)
         // [ 1 x 3 x ]
         // substep 1:
         // [ x 0 x 2 ]