update isa spec (#289)

Signed-off-by: Brian Xu <brianx@xilinx.com>

Author: xuhz

specification/aie2ps/isa-spec.html

@@ -630,9 +630,14 @@ <h2 id="start_cond_job_preempt-0x1f">START_COND_JOB_PREEMPT (0x1f)</h2>
 </table>
 <p>Indicates the start of a new job and creates a new entry in the job table,
 if preemption flag has been set. The flag is cleared when preemption is
-checked, and is set if preemption happens.
+checked, and is set if preemption happens, so the job runs only if the flag indicates that a
+preemption occurred when the condition is checked.
 The job size is auto-calculated and inserted by the assembler and not supplied
-explicitly by the user.</p>
+explicitly by the user.
+In multi-uC case, CERT will do barrier sync after this job, so each uC should have the same number
+of this type of conditional jobs.
+The conditional job can share page with other jobs, and all jobs following this job in the page
+don't start before this job is done.</p>
 <h2 id="uc_dma_write_des-0x01">UC_DMA_WRITE_DES (0x01)</h2>
 <p>Enqueues a DM2MM uC-DMA transfer and returns a wait handle for it.</p>
 <table>
@@ -1369,8 +1374,10 @@ <h2 id="preempt-0x19">PREEMPT (0x19)</h2>
 unit of page to the 1st page of control code, and this address is where the <code>RESTORE</code> control code resides. The opcode
 can determine whether preemption is required, and if required, it can also distinguish whether it is to <code>SAVE</code> or to
 <code>RESTORE</code>, and run the <code>SAVE</code> or <code>RESTORE</code> control code accordingly.
-In multi-uc case, for each preemption point id, the control code of each uc should have this opcode with same <code>id</code>.
-This opcode should take one whole job which in turn should take one whole page.</p>
+In multi-uC case, at each preemption point, CERT will do barrier sync regardless of whether the preemption happens or not, so
+for each preemption point id, the control code of each uc should have this opcode with same <code>id</code>.
+This opcode should take one whole job but the job can share page with other jobs. All following jobs in same page don't start
+before this preemption job is done</p>
 <h2 id="load_pdi-0x1a">LOAD_PDI (0x1a)</h2>
 <p>load pdi</p>
 <table>
@@ -1398,10 +1405,13 @@ <h2 id="load_pdi-0x1a">LOAD_PDI (0x1a)</h2>
 </tbody>
 </table>
 <p>pdi itself is also a piece of control code. It can be loaded by other control code at anywhere anytime.
-<code>pdi_id</code> is an elf wide unique id and specifies an unique pdi. consecutive loading of same pdi results in following
+<code>pdi_id</code> is an elf wide unique id and specifies a unique pdi. If multiple control code elfs run in one hwctx,
+the pdi_id should be hwctx wide unique. When this opcode runs, consecutive loading of same pdi results in following
 loading skipped by the uC. <code>pdi_host_addr_offset</code> specifies a relative address in unit of page to the 1st page of
 the control code and is where the pdi control code resides
-This opcode should take one whole job which in turn should take one whole page.</p>
+This opcode should take one whole job but the job can share page with other jobs. All jobs following the load pdi job
+don't start before the load pdi job is done.
+In multi-uc case, cert will do barrier sync after the load_pdi job, so each uc needs to have same number of load pdi jobs</p>
 <h2 id="load_cores-0x04">LOAD_CORES (0x04)</h2>
 <p>load cores</p>
 <table>

specification/aie2ps/isa-spec.md

@@ -238,9 +238,14 @@ Indicates a new job which can be conditionally started according to preemption.
 
 Indicates the start of a new job and creates a new entry in the job table,
 if preemption flag has been set. The flag is cleared when preemption is
-checked, and is set if preemption happens.
+checked, and is set if preemption happens, so the job runs only if the flag indicates that a
+preemption occurred when the condition is checked.
 The job size is auto-calculated and inserted by the assembler and not supplied
 explicitly by the user.
+In multi-uC case, CERT will do barrier sync after this job, so each uC should have the same number
+of this type of conditional jobs.
+The conditional job can share page with other jobs, and all jobs following this job in the page
+don't start before this job is done.
 
 
 ## UC_DMA_WRITE_DES (0x01)
@@ -691,8 +696,10 @@ and this address is where the `SAVE` control code resides. `restore_control_code
 unit of page to the 1st page of control code, and this address is where the `RESTORE` control code resides. The opcode
 can determine whether preemption is required, and if required, it can also distinguish whether it is to `SAVE` or to
 `RESTORE`, and run the `SAVE` or `RESTORE` control code accordingly.
-In multi-uc case, for each preemption point id, the control code of each uc should have this opcode with same `id`.
-This opcode should take one whole job which in turn should take one whole page.
+In multi-uC case, at each preemption point, CERT will do barrier sync regardless of whether the preemption happens or not, so
+for each preemption point id, the control code of each uc should have this opcode with same `id`.
+This opcode should take one whole job but the job can share page with other jobs. All following jobs in same page don't start
+before this preemption job is done
 
 
 ## LOAD_PDI (0x1a)
@@ -704,10 +711,13 @@ load pdi
 | opcode (8b) | pad (8b) | pad (16b) | const (32b) | page_id (16b) | pad (16b) | 12B |
 
 pdi itself is also a piece of control code. It can be loaded by other control code at anywhere anytime.
-`pdi_id` is an elf wide unique id and specifies an unique pdi. consecutive loading of same pdi results in following
+`pdi_id` is an elf wide unique id and specifies a unique pdi. If multiple control code elfs run in one hwctx,
+the pdi_id should be hwctx wide unique. When this opcode runs, consecutive loading of same pdi results in following
 loading skipped by the uC. `pdi_host_addr_offset` specifies a relative address in unit of page to the 1st page of
 the control code and is where the pdi control code resides
-This opcode should take one whole job which in turn should take one whole page.
+This opcode should take one whole job but the job can share page with other jobs. All jobs following the load pdi job
+don't start before the load pdi job is done.
+In multi-uc case, cert will do barrier sync after the load_pdi job, so each uc needs to have same number of load pdi jobs
 
 
 ## LOAD_CORES (0x04)

specification/aie2ps/isa-spec.yaml

@@ -113,9 +113,14 @@ operations:
     description: |
       Indicates the start of a new job and creates a new entry in the job table,
       if preemption flag has been set. The flag is cleared when preemption is
-      checked, and is set if preemption happens.
+      checked, and is set if preemption happens, so the job runs only if the flag indicates that a
+      preemption occurred when the condition is checked.
       The job size is auto-calculated and inserted by the assembler and not supplied
       explicitly by the user.
+      In multi-uC case, CERT will do barrier sync after this job, so each uC should have the same number
+      of this type of conditional jobs.
+      The conditional job can share page with other jobs, and all jobs following this job in the page
+      don't start before this job is done.
   - mnemonic: UC_DMA_WRITE_DES
     opcode: 0x01
     regular: True
@@ -624,8 +629,10 @@ operations:
       unit of page to the 1st page of control code, and this address is where the `RESTORE` control code resides. The opcode
       can determine whether preemption is required, and if required, it can also distinguish whether it is to `SAVE` or to
       `RESTORE`, and run the `SAVE` or `RESTORE` control code accordingly.
-      In multi-uc case, for each preemption point id, the control code of each uc should have this opcode with same `id`.
-      This opcode should take one whole job which in turn should take one whole page.
+      In multi-uC case, at each preemption point, CERT will do barrier sync regardless of whether the preemption happens or not, so
+      for each preemption point id, the control code of each uc should have this opcode with same `id`.
+      This opcode should take one whole job but the job can share page with other jobs. All following jobs in same page don't start
+      before this preemption job is done
   - mnemonic: LOAD_PDI
     opcode: 0x1a
     regular: True
@@ -643,10 +650,13 @@ operations:
     brief: load pdi
     description: |
       pdi itself is also a piece of control code. It can be loaded by other control code at anywhere anytime.
-      `pdi_id` is an elf wide unique id and specifies an unique pdi. consecutive loading of same pdi results in following
+      `pdi_id` is an elf wide unique id and specifies a unique pdi. If multiple control code elfs run in one hwctx,
+      the pdi_id should be hwctx wide unique. When this opcode runs, consecutive loading of same pdi results in following
       loading skipped by the uC. `pdi_host_addr_offset` specifies a relative address in unit of page to the 1st page of
       the control code and is where the pdi control code resides
-      This opcode should take one whole job which in turn should take one whole page.
+      This opcode should take one whole job but the job can share page with other jobs. All jobs following the load pdi job
+      don't start before the load pdi job is done.
+      In multi-uc case, cert will do barrier sync after the load_pdi job, so each uc needs to have same number of load pdi jobs
   - mnemonic: LOAD_CORES
     opcode: 0x04
     regular: True