Merge branch 'main' into feat/ecsm-spec-update

Author: jotabulacios

.github/workflows/benchmark-gpu.yml

@@ -33,7 +33,7 @@ concurrency:
   cancel-in-progress: true
 
 env:
-  # Vast offer search: RTX 5090, >=16 cores, >=64GB RAM, >=64GB disk, verified +
+  # Vast offer search: RTX 5090, >=16 cores, >=96GB RAM, >=64GB disk, verified +
   # rentable, Blackwell-capable driver, <= cap.
   GPU_NAME: RTX_5090
   PRICE_CAP: "1"
@@ -169,7 +169,7 @@ jobs:
           MIN_DRIVER: "580"
         run: |
           # cpu_ram filter is in GB.
-          QUERY="gpu_name=${GPU_NAME} num_gpus=1 cpu_cores_effective>=16 cpu_ram>=64 disk_space>=64 verified=true rentable=true cuda_max_good>=12.8 dph_total<=${PRICE_CAP}"
+          QUERY="gpu_name=${GPU_NAME} num_gpus=1 cpu_cores_effective>=16 cpu_ram>=96 disk_space>=64 verified=true rentable=true cuda_max_good>=12.8 dph_total<=${PRICE_CAP}"
           echo "Query: $QUERY (+ client-side driver_version major >= $MIN_DRIVER)"
           # Keep only offers whose driver major >= MIN_DRIVER, then most expensive first
           # (within the price cap) — premium hosts have faster disks/network (quicker image
@@ -190,7 +190,7 @@ jobs:
             sleep "$OFFER_INTERVAL"
           done
           if [ -z "$OFFER_ID" ]; then
-            echo "::error::No RTX 5090 offer matched after $OFFER_ATTEMPTS attempts (>=16 cores, >=64GB RAM, >=64GB disk, driver>=${MIN_DRIVER}, <= \$${PRICE_CAP}/hr)"
+            echo "::error::No RTX 5090 offer matched after $OFFER_ATTEMPTS attempts (>=16 cores, >=96GB RAM, >=64GB disk, driver>=${MIN_DRIVER}, <= \$${PRICE_CAP}/hr)"
             exit 1
           fi
           echo "id=$OFFER_ID" >> "$GITHUB_OUTPUT"

Cargo.lock

@@ -1,5 +1,6 @@
 .PHONY: deps deps-linux deps-macos compile-programs-asm compile-programs-rust compile-bench \
-compile-programs clean-asm clean-rust clean-bench clean-shared clean test test-asm \
+compile-programs compile-recursion-elfs clean-asm clean-rust clean-bench clean-shared \
+clean-recursion-elfs clean test test-asm \
 test-rust test-executor test-flamegraph flamegraph-prover \
 test-fast test-prover test-prover-all test-disk-spill test-math-cuda test-cuda-integration \
 bench-math-cuda bench-prover bench-prover-cuda build check clippy fmt lint regen-ethrex-fixtures \
@@ -46,6 +47,13 @@ BENCH_PROGRAM_DIRS := $(dir $(wildcard $(BENCH_PROGRAMS_DIR)/*/Cargo.toml))
 BENCH_PROGRAMS := $(notdir $(basename $(BENCH_PROGRAM_DIRS:%/=%)))
 BENCH_ARTIFACTS := $(addprefix $(BENCH_ARTIFACTS_DIR)/, $(addsuffix .elf, $(BENCH_PROGRAMS)))
 
+# Recursion smoke-test guests, in bench_vs/lambda/ (shared with bench_vs/run.sh)
+# rather than executor/programs/. The recursion guest is the in-VM STARK verifier.
+RECURSION_GUESTS_DIR=./bench_vs/lambda
+RECURSION_ARTIFACTS_DIR=./executor/program_artifacts/recursion
+RECURSION_GUESTS := empty fibonacci recursion
+RECURSION_ARTIFACTS := $(addprefix $(RECURSION_ARTIFACTS_DIR)/, $(addsuffix .elf, $(RECURSION_GUESTS)))
+
 # Override with: make ... SYSROOT_DIR=$HOME/.lambda-vm-sysroot
 # to install the sysroot in a user-writable location and avoid sudo.
 SYSROOT_DIR ?= /opt/lambda-vm-sysroot
@@ -133,7 +141,16 @@ compile-programs-rust: prepare-sysroot $(RUST_ARTIFACTS)
 
 compile-bench: prepare-sysroot $(BENCH_ARTIFACTS)
 
-compile-programs: compile-programs-asm compile-programs-rust compile-bench
+# NOTE: the recursion smoke tests are #[ignore]d (not run by `make test` /
+# `test-executor`) because they're too slow for CI today; only `test-prover-all`
+# runs them. We still compile their guest ELFs on every build so they keep
+# compiling until the tests are fast enough to run in CI.
+compile-programs: compile-programs-asm compile-programs-rust compile-bench compile-recursion-elfs
+
+compile-recursion-elfs: prepare-sysroot $(RECURSION_ARTIFACTS)
+
+$(RECURSION_ARTIFACTS_DIR):
+	mkdir -p $@
 
 
 $(RUST_ARTIFACTS_DIR):
@@ -142,34 +159,49 @@ $(RUST_ARTIFACTS_DIR):
 $(BENCH_ARTIFACTS_DIR):
 	mkdir -p $@
 
+# The guest .elf rules depend on FORCE so their recipe always runs: cargo already
+# tracks the full dependency graph, so we let it decide what to rebuild (a fast
+# no-op when nothing changed) rather than re-encode that in Make prereqs.
+.PHONY: FORCE
+FORCE:
+
+# The guest .elf rules all share one canned recipe: the cargo build invocation is
+# identical across the rust, bench, and recursion guests. They differ only in the
+# source directory ($(1)) and the built-binary name suffix ($(2): empty when the
+# binary == crate name, `-bench` for the recursion suite, whose crates are named
+# <name>-bench). cargo owns the dep graph (see FORCE above), so the recipe always
+# runs and lets cargo decide what to actually rebuild.
+define build_guest_elf
+cd $(1)/$* && \
+	CARGO_TARGET_DIR=$(abspath $(SHARED_TARGET_DIR)) \
+	CFLAGS_riscv64im_lambda_vm_elf="$(SYSROOT_CFLAGS)" \
+	rustup run nightly-2026-02-01 cargo build --release \
+		--target $(RV64_TARGET_SPEC) \
+		-Z build-std=core,alloc,std,compiler_builtins,panic_abort \
+		-Z build-std-features=compiler-builtins-mem \
+		-Z json-target-spec
+cp $(SHARED_TARGET_DIR)/riscv64im-lambda-vm-elf/release/$*$(2) $@
+endef
+
 # Compile rust (64-bit)
 # Order-only `| prepare-sysroot` so a direct `make .../foo.elf` provisions the sysroot
 # first (the aggregate compile-programs-rust/compile-bench targets already do, but a
 # bare pattern-rule invocation like `make -B .../ethrex.elf` would otherwise skip it
 # and fail to compile guest C dependencies). Order-only because prepare-sysroot is
 # .PHONY — a normal prereq would force a rebuild every time; its recipe is idempotent.
-$(RUST_ARTIFACTS_DIR)/%.elf: $(RUST_PROGRAMS_DIR)/%/Cargo.toml | prepare-sysroot $(RUST_ARTIFACTS_DIR)
-	cd $(RUST_PROGRAMS_DIR)/$* && \
-		CARGO_TARGET_DIR=$(abspath $(SHARED_TARGET_DIR)) \
-		CFLAGS_riscv64im_lambda_vm_elf="$(SYSROOT_CFLAGS)" \
-		rustup run nightly-2026-02-01 cargo build --release \
-			--target $(RV64_TARGET_SPEC) \
-			-Z build-std=core,alloc,std,compiler_builtins,panic_abort \
-			-Z build-std-features=compiler-builtins-mem \
-			-Z json-target-spec
-	cp $(SHARED_TARGET_DIR)/riscv64im-lambda-vm-elf/release/$* $@
+$(RUST_ARTIFACTS_DIR)/%.elf: FORCE | prepare-sysroot $(RUST_ARTIFACTS_DIR)
+	$(call build_guest_elf,$(RUST_PROGRAMS_DIR),)
 
 # Compile rust benches (64-bit)
-$(BENCH_ARTIFACTS_DIR)/%.elf: $(BENCH_PROGRAMS_DIR)/%/Cargo.toml | prepare-sysroot $(BENCH_ARTIFACTS_DIR)
-	cd $(BENCH_PROGRAMS_DIR)/$* && \
-		CARGO_TARGET_DIR=$(abspath $(SHARED_TARGET_DIR)) \
-		CFLAGS_riscv64im_lambda_vm_elf="$(SYSROOT_CFLAGS)" \
-		rustup run nightly-2026-02-01 cargo build --release \
-			--target $(RV64_TARGET_SPEC) \
-			-Z build-std=core,alloc,std,compiler_builtins,panic_abort \
-			-Z build-std-features=compiler-builtins-mem \
-			-Z json-target-spec
-	cp $(SHARED_TARGET_DIR)/riscv64im-lambda-vm-elf/release/$* $@
+$(BENCH_ARTIFACTS_DIR)/%.elf: FORCE | prepare-sysroot $(BENCH_ARTIFACTS_DIR)
+	$(call build_guest_elf,$(BENCH_PROGRAMS_DIR),)
+
+# Recursion-suite guests (bench_vs/lambda/): the crate's binary is <name>-bench, so
+# copy <name>-bench -> <name>.elf. std-inclusive build-std covers both the no_std
+# inner guests and the std recursion verifier. Prover tests read these prebuilt
+# artifacts like every other program (see prover/src/tests/recursion_smoke_test.rs).
+$(RECURSION_ARTIFACTS_DIR)/%.elf: FORCE | prepare-sysroot $(RECURSION_ARTIFACTS_DIR)
+	$(call build_guest_elf,$(RECURSION_GUESTS_DIR),-bench)
 
 clean-asm:
 	-rm -rf $(ASM_ARTIFACTS_DIR)
@@ -183,7 +215,10 @@ clean-bench:
 clean-shared:
 	-rm -rf $(SHARED_TARGET_DIR)
 
-clean: clean-asm clean-rust clean-bench clean-shared
+clean-recursion-elfs:
+	-rm -rf $(RECURSION_ARTIFACTS_DIR)
+
+clean: clean-asm clean-rust clean-bench clean-shared clean-recursion-elfs
 
 test-executor: compile-programs
 	cargo test -p executor
@@ -225,8 +260,9 @@ test-fast:
 test-prover:
 	cargo test -p lambda-vm-prover
 
-# Prover tests including slow ones
-test-prover-all:
+# Prover tests including slow ones. The recursion smoke tests (#[ignore]d) read
+# prebuilt guest ELFs from executor/program_artifacts/recursion/, so build them first.
+test-prover-all: compile-recursion-elfs
 	cargo test -p lambda-vm-prover -- --include-ignored
 
 # Prover tests with debug-checks (shows bus balance report)

Makefile

@@ -0,0 +1,6 @@
+[target.riscv64im-lambda-vm-elf]
+rustflags = [
+  "-C", "link-arg=-e",
+  "-C", "link-arg=main",
+  "-C", "passes=lower-atomic"
+]