hwacha icon indicating copy to clipboard operation
hwacha copied to clipboard

rv64uv-p-amoadd_w bug

Open abejgonzalez opened this issue 6 years ago • 4 comments

Chipyard/dev found a bug in running asm tests with Hwacha on Rocket.

https://circleci.com/gh/ucb-bar/chipyard/2056?utm_campaign=vcs-integration-link&utm_medium=referral&utm_source=github-build-link

abejgonzalez avatar Jul 11 '19 17:07 abejgonzalez

Hwacha does atomics in the L2. If you aren't building with the L2, the AMO tests should fail.

jerryz123 avatar Jul 11 '19 17:07 jerryz123

I think we can now fix this with an AtomicAutomata that will do nothing if some outer manager supports them (i.e. if there is an L2).

colinschmidt avatar Jul 11 '19 17:07 colinschmidt

That sounds fine with me.

abejgonzalez avatar Jul 11 '19 18:07 abejgonzalez

I am running into atomics failures with the Hwacha L2 config as well:

mkdir -p /home/theodore/dev/clones/chipyard/sims/verilator/output/example.HwachaL2Config
ln -fs /home/theodore/dev/clones/chipyard/esp-tools-install/riscv64-unknown-elf/share/riscv-tests/isa/rv64uv-vp-amoor_w /home/theodore/dev/clones/chipyard/sims/verilator/output/example.HwachaL2Config/rv64uv-vp-amoor_w
(set -o pipefail && /home/theodore/dev/clones/chipyard/sims/verilator/simulator-example-HwachaL2Config  +max-cycles=10000000 +verbose  /home/theodore/dev/clones/chipyard/sims/verilator/output/example.HwachaL2Config/rv64uv-vp-amoor_w 3>&1 1>&2 2>&3 | spike-dasm > /home/theodore/dev/clones/chipyard/sims/verilator/output/example.HwachaL2Config/rv64uv-vp-amoor_w.out)
[194040] %Error: example.HwachaL2Config.top.v:324950: Assertion failed in TOP.TestHarness.TestHarness.tile.hwacha.vus.vxuInst.dccInst.vluInst
%Error: /home/theodore/dev/clones/chipyard/sims/verilator/generated-src/example.HwachaL2Config/example.HwachaL2Config.top.v:324950: Verilog $stop
Aborting...
/home/theodore/dev/clones/chipyard/common.mk:111: recipe for target '/home/theodore/dev/clones/chipyard/sims/verilator/output/example.HwachaL2Config/rv64uv-vp-amoor_w.out' failed
C0:     154711 [0] pc=[000000000000407c] W[r 0=0000000000004078][0] R[r10=0000000000004078] R[r16=00000000000040ef] inst=[0305282b] custom1.rs1 (args unknown)
H: [0] pc=[547d3a05f0] SW[r196=000000007d3a05f0][0] SR[r 98=2cad5d873cc24fcc] SR[r 71=000000007d3a05f0] inst=[3300320c5a232669] c.addiw a2, 26
C0:     154712 [0] pc=[000000000000407c] W[r 0=0000000000004078][0] R[r10=0000000000004078] R[r16=00000000000040ef] inst=[0305282b] custom1.rs1 (args unknown)
H: [0] pc=[547d3a05f0] SW[r  0=000000007d3a05f0][0] SR[r  0=2cad5d873cc24fcc] SR[r  0=000000007d3a05f0] inst=[3300320c5a232669] c.addiw a2, 26
C0:     154713 [0] pc=[000000000000407c] W[r 0=0000000000004078][0] R[r10=0000000000004078] R[r16=00000000000040ef] inst=[0305282b] custom1.rs1 (args unknown)
H: [0] pc=[547d3a05f0] SW[r  2=000000007d3a05f0][0] SR[r  2=2cad5d873cc24fcc] SR[r  0=000000007d3a05f0] inst=[3300320c5a232669] c.addiw a2, 26
H: write_prf 0 0   0 0 1
H: write_prf 0 0   0 1 1
Assertion failed: VLU: non-zero quiescent bias_tail
    at dcc-mem.scala:505 assert(bias_tail === SInt(0), "VLU: non-zero quiescent bias_tail")

Ravenwater avatar Sep 16 '19 11:09 Ravenwater