tools/gen-device-svd: include the last register in cluster size

[zombieleet]

For a cluster with dim 1 or no dim, gen-device-svd computes the size from the start of the last register instead of its end. The padding after the cluster is then too large, and every following field lands at the wrong offset. This is an existing bug from #3335.

-lastAddress := lastReg.Address
+lastAddress := lastReg.Address + uint64(lastReg.ElementSize)

Affected today: nrf9160 (SPU) and nrf5340_application (SPU, DCNF). For example SPU.DPPI is at 0x484 instead of 0x480. No other generated file changes. The 315 "struct size" warnings gen-device prints for the nrf files are gone, and so are the 6 for Atmel, whose output does not change.

Why here and not in tinygo-org/esp32-svd: the esp-pacs SVD files in tinygo-org/esp32-svd#1 use many such clusters, so without this fix the TIMG watchdog registers end up at the wrong address on several ESP chips. Unlike #5797, this is a bug in existing code and the SVD files are correct. The only SVD workaround is to flatten the clusters, which renames TIMG0.T[0].CONFIG to TIMG0.T0CONFIG on single-timer chips only and breaks shared code like runtime_esp32xx.go.

Testing: ran make gen-device before and after, and only the two nRF files changed. Compared the real Go field offsets of every generated struct with the SVD offsets. Both nRF files and all 15 esp32-svd chips now match. Added TestProcessClusterSizeDim1, which fails on dev (ElementSize 4) and passes with this change.

[deadprogram]

Thanks for finding this @zombieleet. I regenerated the device files before and after this change and took a good look. The following is edited from an automated review:

1. Only nrf9160 and nrf5340_application change, and the new offsets match the SVD. As a side effect, the 315 "struct size" warnings that gen-device prints for the nrf files are gone. Perhaps mention that in the description?

2. Please add a test for this. tools/gen-device-svd: Increment lsb/msb last in bitfield ranges. #5821 adds gen-device-svd_test.go, so a small processCluster case with a dim 1 cluster of two registers that checks ElementSize could go there once it lands.

What do you think?

[zombieleet]

Thanks for checking @deadprogram . Good point on the warnings: with this change the nrf struct size warnings go from 315 to 0, and the 6 for Atmel go away too, with no change to the Atmel output. I added that to the description.

I added TestProcessClusterSizeDim1 in gen-device-svd_test.go. It uses a dim 1 cluster with two 32-bit registers and checks that ElementSize is 8. It fails on dev (4) and passes with this change. #5821 also adds gen-device-svd_test.go, so whichever lands second will need a small rebase to merge the two test files. I'm happy to do that on my side if #5821 goes in first.