wu-arch/kernels
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Commit Graph

  • 5c6526e414 Merge branch 'kernels' of https://github.com/hansungk/vortex-private into kernels Richard Yan 2024-06-12 02:12:45 -07:00
  • f37f5d5612 dual gemmini kernel + quad core vortex Richard Yan 2024-06-12 02:12:38 -07:00
  • 32e31c51a4 sgemm_tcore: Blocksize 64; Fix kernel launch on larger dim Hansung Kim 2024-06-11 22:27:12 -07:00
  • 03d1df8f53 sgemm_tcore: Separate transpose control on AS read/write Hansung Kim 2024-06-11 21:16:23 -07:00
  • 34eaab4c87 sgemm_tcore: Fix warp-specialized kernel for larger dim Hansung Kim 2024-06-11 20:50:06 -07:00
  • 9febfb9bdc sgemm_tcore: Move global_dmem_load back to kernel.cpp Hansung Kim 2024-06-11 20:12:30 -07:00
  • ca7fd84a83 sgemm_tcore: Split util functions to a header file Hansung Kim 2024-06-11 19:06:22 -07:00
  • dab9d7c6fc sgemm_tcore: Fix kernel launch for smaller TBs than cluster threads Hansung Kim 2024-06-11 14:09:31 -07:00
  • e3c4a4d2f5 sgemm_tcore: Improve agen for !transpose_as smem load Hansung Kim 2024-06-10 22:08:37 -07:00
  • dc7bd6b248 sgemm_tcore: Fix warp_row/col calculation bug Hansung Kim 2024-06-10 19:52:37 -07:00
  • 3b2f5a31de sgemm_tcore: Improve write_result addr gen Hansung Kim 2024-06-10 19:34:00 -07:00
  • a22762db94 sgemm_tcore: Add GEMMINI_DMA to non-warp-specialized mode Hansung Kim 2024-06-10 17:06:46 -07:00
  • 51d9cffb2d Merge remote-tracking branch 'origin/kernels' into kernels Hansung Kim 2024-06-10 16:41:36 -07:00
  • 39449ece37 Add warp-specialized{, and dma enabled} kernel Hansung Kim 2024-06-10 16:39:49 -07:00
  • 357435bc96 Merge branch 'kernels' of https://github.com/hansungk/vortex-private into kernels Richard Yan 2024-06-09 14:35:02 -07:00
  • c327474e3b power specific code for kernel Richard Yan 2024-06-09 14:34:58 -07:00
  • e4eed20de3 sgemm_gemmini_dma: Fix device addr for operands Hansung Kim 2024-06-09 13:32:54 -07:00
  • 90662484b1 sgemm_tcore: inline left out Hansung Kim 2024-06-07 20:34:29 -07:00
  • d61dd85872 sgemm_tcore: Remove unused SIMT core code Hansung Kim 2024-06-07 20:29:08 -07:00
  • aaf4a89b57 Fix asm label already defined error Hansung Kim 2024-06-07 19:55:28 -07:00
  • 9e8988df6b Patch args device address for dma kernel Hansung Kim 2024-06-07 18:32:07 -07:00
  • fc8f0c99f0 Merge branch 'tensor_core' into kernels Hansung Kim 2024-06-07 18:27:02 -07:00
  • 800d9801b5 tensor: Test with multiple accumulators Hansung Kim 2024-06-07 18:19:20 -07:00
  • 080923e869 common.mk: Add more aggressive inline flag Hansung Kim 2024-06-07 18:14:40 -07:00
  • 2cac995db9 tensor: generate 8x8 in correctness script Hansung Kim 2024-06-07 18:13:57 -07:00
  • 7cf59c9480 dma and demo kernels Richard Yan 2024-06-07 18:11:19 -07:00
  • 483f975439 Merge branch 'kernels' into tensor_core Hansung Kim 2024-06-07 16:27:01 -07:00
  • 3a6427a491 sgemm_tcore: Hardcode threadblock id 0 Hansung Kim 2024-06-07 16:08:40 -07:00
  • 856596cbb3 sgemm_tcore: Write reference C before sim Hansung Kim 2024-06-07 10:39:11 -07:00
  • 985c5fc0dc sgemm_tcore: Remove uneffective register asm Hansung Kim 2024-06-06 18:50:31 -07:00
  • 7c4d850074 sgemm_tcore: Experiment with high K; 48% util Hansung Kim 2024-06-06 18:38:43 -07:00
  • 062403066e sgemm_tcore: Bring M/N-loop inside the kernel Hansung Kim 2024-06-06 15:22:01 -07:00
  • d5adacda30 Add args.bin to ELF Hansung Kim 2024-06-06 15:19:05 -07:00
  • 2c50b0cdce sgemm_tcore: Remove BM_d/BN_d Hansung Kim 2024-06-06 14:27:57 -07:00
  • 7f6f096191 sgemm_tcore: Use if constexpr Hansung Kim 2024-06-06 13:43:57 -07:00
  • deb6e5eba2 sgemm_tcore: Move bank-conflicts to SMEM stores from GMEM loads Hansung Kim 2024-06-06 12:43:08 -07:00
  • ab4d525970 sgemm_tcore: More asserts on manual unrolling Hansung Kim 2024-06-06 12:26:07 -07:00
  • a42fa6a113 sgemm_tcore: Swap out mul with bitwise ops for addr ping-pong Hansung Kim 2024-06-05 19:01:59 -07:00
  • 65c653afde sgemm_tcore: Use arithmetic instead of branch for double-buffered addr Hansung Kim 2024-06-05 18:03:08 -07:00
  • c7a6ed03de sgemm_tcore: Use constant offset to reduce SMEM addr calc Hansung Kim 2024-06-05 18:02:34 -07:00
  • 95b5719847 sgemm_tcore: Split K-dim loop between consumer/producer Hansung Kim 2024-06-05 17:14:39 -07:00
  • 150f14af25 sgemm_tcore: Use multiple fp regs for GMEM->SMEM Hansung Kim 2024-06-05 16:53:39 -07:00
  • e44173c65e sgemm_tcore: Deconstruct addr calc for GMEM->SMEM Hansung Kim 2024-06-05 15:11:01 -07:00
  • ff6e5bf6dc sgemm_tcore: Deconstruct smem addr calc to reduce reg alloc Hansung Kim 2024-06-05 14:50:36 -07:00
  • d8944db369 sgemm_tcore: Double-buffer over K-dimension Hansung Kim 2024-06-04 18:23:27 -07:00
  • 18e3653d31 sgemm_tcore: Increase RF data reuse for WMITER/WNITER Hansung Kim 2024-06-03 21:10:42 -07:00
  • 4e723c4655 sgemm_tcore: Support two accumulation reg tiles Hansung Kim 2024-06-01 01:12:08 -07:00
  • c8d6c56dd9 sgemm_tcore: Split global DMEM load into a function Hansung Kim 2024-05-31 17:37:06 -07:00
  • 220ee0aa5e sgemm_tcore: Unroll around WMITER/WNITER Hansung Kim 2024-05-31 17:35:01 -07:00
  • 2b5836022d Also generate kernel.CONFIG.elf Hansung Kim 2024-05-26 13:57:07 -07:00
  • c08a4cba8b Add -ffixed-regs to tests/kernel makefile Hansung Kim 2024-05-26 13:56:34 -07:00
  • 1e48bad4f9 sgemm_tcore: Fix AS transpose Hansung Kim 2024-05-26 13:51:47 -07:00
  • 200fd3e08c sgemm_tcore: Revert to packed smem alloc Hansung Kim 2024-05-25 22:47:59 -07:00
  • bc7bd1a1dd sgemm_tcore: Write reference C matrix to file Hansung Kim 2024-05-25 22:47:15 -07:00
  • 0a884e1ead tensor: spawn on all warps, 8 lanes Hansung Kim 2024-05-25 20:19:57 -07:00
  • b892c22f00 sgemm_tcore: Reflect WMITER/WNITER in threadblock size Hansung Kim 2024-05-16 23:31:52 -07:00
  • 18ecebddc0 sgemm_tcore: Fix round-down error with CORES_PER_CLUSTER Hansung Kim 2024-05-16 21:36:24 -07:00
  • 78b2a318c1 sgemm_tcore: Implement A transpose for coalesced smem access Hansung Kim 2024-05-16 20:22:15 -07:00
  • 8f64fae7a7 sgemm_tcore: Addr gen for local_k; add SIMT-only for reference Hansung Kim 2024-05-16 14:09:55 -07:00
  • df1aa62916 sgemm_tcore: Add warptiling parameters Hansung Kim 2024-05-15 15:23:26 -07:00
  • 5de8e7c33a sgemm_tg: Fix device address to use ELF operands Hansung Kim 2024-05-13 23:09:57 -07:00
  • 9d2b533d5c sgemm_tg: Do operand elf stitching for kernel.elf as well Hansung Kim 2024-05-13 16:48:13 -07:00
  • 09b23ffe87 sgemm_tg: 1-octet 8-lane kernel Hansung Kim 2024-05-13 14:52:33 -07:00
  • d848e88f72 sgemm_tcore: Move C from regF->GMEM directly Hansung Kim 2024-05-13 14:00:50 -07:00
  • 9e60b1834c sgemm_tcore: Rewrite with sgemm_Wg parametrization Hansung Kim 2024-05-13 13:22:06 -07:00
  • 5c298c81df sgemm_tg: Use reg mapping functions Hansung Kim 2024-05-12 22:22:54 -07:00
  • 8a521a1de8 Add 8-lane operand mapping Hansung Kim 2024-05-10 23:23:11 -07:00
  • 33066af56e cisc gemmini Richard Yan 2024-05-08 15:46:20 -07:00
  • 6af0c305ea Fix path to OBJCOPY Hansung Kim 2024-05-08 13:27:11 -07:00
  • 6ba6a1e2e5 Merge branch 'kernels' into tensor_core Hansung Kim 2024-05-08 13:25:31 -07:00
  • 5821bfd10d Repeat vx_wmma issue & hardcode dst address Hansung Kim 2024-05-08 13:22:26 -07:00
  • 7775830814 Hardcode chipyard device addresses Hansung Kim 2024-05-07 16:30:30 -07:00
  • b4c812f9f8 Write expected_C to a binary file Hansung Kim 2024-05-03 17:27:25 -07:00
  • 5bd25985c6 i kinda forgot most of changes joshua 2024-05-04 23:01:47 -07:00
  • 4db59446f3 Merge branch 'kernels' of https://github.com/hansungk/vortex-private into kernels Richard Yan 2024-05-02 15:16:57 -07:00
  • 1b6ebf86a1 update gemmini kernels Richard Yan 2024-05-02 15:16:55 -07:00
  • a606a9ef42 common.mk: properly handle unspecified CONFIG Hansung Kim 2024-04-29 17:14:28 -07:00
  • 01f4a69ae9 dma mvout, double buffering & other opts Richard Yan 2024-04-28 01:18:51 -07:00
  • d21e7b92c7 internal accumulation, forced rematerialization, better unrolling Richard Yan 2024-04-25 15:28:12 -07:00
  • a44edf2b65 Merge branch 'kernels' of https://github.com/hansungk/vortex-private into kernels Richard Yan 2024-04-24 22:10:40 -07:00
  • 6eafa2de54 write operands to elf Richard Yan 2024-04-24 22:09:30 -07:00
  • df881fd69f Generate separate ELF for radiance Hansung Kim 2024-04-24 21:09:01 -07:00
  • 793779aa6c sgemm_wg: 128x128 config Hansung Kim 2024-04-24 21:08:31 -07:00
  • 689043b45e Add regression flops Hansung Kim 2024-04-16 15:22:03 -07:00
  • 6cbfbfb856 sgemm_wg: Output CPU data to binary Hansung Kim 2024-04-14 21:25:14 -07:00
  • 4e9855dc33 highly unrolled a/b load Richard Yan 2024-04-16 22:19:30 -07:00
  • 449d99f0bb dram gemm kernel Richard Yan 2024-04-16 17:15:22 -07:00
  • 99621a0df9 Merge branch 'kernels' of https://github.com/hansungk/vortex-private into kernels Richard Yan 2024-04-15 10:22:19 -07:00
  • 041d49fb58 update gemmini only kernel Richard Yan 2024-04-15 10:22:00 -07:00
  • 0bb7aeb45b add gpu+gemmini gemm kernel Richard Yan 2024-04-15 10:10:20 -07:00
  • d8eddb21ea add gemmini dependency Richard Yan 2024-04-15 10:04:54 -07:00
  • 37a60b1141 sgemm_wg: Output C result to binary Hansung Kim 2024-04-14 12:33:27 -07:00
  • 3383b70732 sgemm_wg: Hardcode device address Hansung Kim 2024-04-14 12:33:02 -07:00
  • 7bf72c9568 cycle counting for fence Richard Yan 2024-04-09 19:53:17 -07:00
  • 93a00101ae sgemm_wg: revert to faster params Hansung Kim 2024-04-04 21:05:06 -07:00
  • 84a31f3384 thread parallel data loading for word strided bank Richard Yan 2024-04-01 11:10:32 -07:00
  • e6db1a83af Merge branch 'kernels' of https://github.com/hansungk/vortex-private into kernels Richard Yan 2024-04-01 11:09:43 -07:00
  • b0c1f77388 vx_start.S: Swizzle stack space Hansung Kim 2024-03-29 12:24:55 -07:00
  • fa2b6e2ad0 sgemm_wg: Explicitly limit unroll to reduce stack spilling Hansung Kim 2024-03-29 02:48:29 -07:00
  • 537b97eb20 common.mk: Don't clean all *.elf Hansung Kim 2024-03-28 20:17:26 -07:00