Add code generation tests

Introduces a suite of code generation tests to the `benches` folder.
Each tests consists of a pair of files: a Rust source containing a
function named `codegen_test`, and a `.x86-64.mca` file containing
the llvm-mca analysis of `codegen_test`. In practice, these tests
operate similarly to UI tests; CI checks a fresh llvm-mca analysis
against that in the file, and fails if they mismatch. Changes can
be expressly blessed via `BLESS=1`.

gherrit-pr-id: G5964d13c9ffb7a47ed4662892ef92c90ddc51e12

Author: jswrenn

.github/workflows/ci.yml

@@ -379,7 +379,8 @@ jobs:
           $FEATURES \
           --verbose \
           -- \
-          --skip ui
+          --skip ui \
+          --skip codegen
 
       # Only run tests when targetting Linux x86 (32- or 64-bit) - we're
       # executing on Linux x86_64, so we can't run tests for any non-x86 target.
@@ -609,6 +610,28 @@ jobs:
         matrix.target != 'wasm32-unknown-unknown' &&
         env.ZC_SKIP_CARGO_SEMVER_CHECKS != '1'
 
+  codegen:
+    runs-on: ubuntu-latest
+    name: Run codegen tests
+    needs: generate_cache
+    steps:
+      - uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
+        with:
+          persist-credentials: false
+      - name: Populate cache
+        uses: ./.github/actions/cache
+      - name: Run tests
+        run: |
+          set -eo pipefail
+          sudo apt install -qq llvm
+          ./cargo.sh +nightly install --quiet cargo-show-asm
+          ./cargo.sh +nightly test \
+            --package zerocopy \
+            --target x86_64-unknown-linux-gnu \
+            --all-features \
+            --verbose \
+            --test codegen
+
   coverage:
     runs-on: ubuntu-latest
     name: Generate code coverage
@@ -631,7 +654,9 @@ jobs:
             --lcov \
             --output-path lcov.info \
             --verbose \
-            -- --skip ui
+            -- \
+            --skip ui \
+            --skip codegen
       - name: Upload coverage to Codecov
         uses: codecov/codecov-action@671740ac38dd9b0130fbe1cec585b89eea48d3de # v5.5.2
         with:
@@ -885,6 +910,7 @@ jobs:
             cargo install --locked kani-verifier                    &> /dev/null &
             cargo install cargo-nextest                             &> /dev/null &
             cargo kani setup                                        &> /dev/null &
+            sudo apt-get install llvm                               &> /dev/null &
 
             wait
 
@@ -988,7 +1014,7 @@ jobs:
       # https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/collaborating-on-repositories-with-code-quality-features/troubleshooting-required-status-checks#handling-skipped-but-required-checks
       if: failure()
       runs-on: ubuntu-latest
-      needs: [build_test, coverage, kani, check_be_aarch64, check_avr_atmega, check_fmt, check_actions, check_readme, check_versions, check_msrv_is_minimal, check_stale_stderr, generate_cache, check-all-toolchains-tested, check-job-dependencies, check-todo, run-git-hooks, zizmor]
+      needs: [build_test, codegen, coverage, kani, check_be_aarch64, check_avr_atmega, check_fmt, check_actions, check_readme, check_versions, check_msrv_is_minimal, check_stale_stderr, generate_cache, check-all-toolchains-tested, check-job-dependencies, check-todo, run-git-hooks, zizmor]
       steps:
         - name: Mark the job as failed
           run: exit 1

benches/formats/coco.rs

@@ -0,0 +1,25 @@
+// Copyright 2026 The Fuchsia Authors
+//
+// Licensed under a BSD-style license <LICENSE-BSD>, Apache License, Version 2.0
+// <LICENSE-APACHE or https://www.apache.org/licenses/LICENSE-2.0>, or the MIT
+// license <LICENSE-MIT or https://opensource.org/licenses/MIT>, at your option.
+// This file may not be copied, modified, or distributed except according to
+// those terms.
+
+use zerocopy_derive::*;
+
+// The only valid value of this type are the bytes `0xC0C0`.
+#[derive(TryFromBytes, KnownLayout, Immutable)]
+#[repr(u16, align(2))]
+enum C0C0 {
+    _XC0C0 = 0xC0C0,
+}
+
+#[derive(TryFromBytes, KnownLayout, Immutable)]
+#[repr(C)]
+pub struct Packet {
+    magic_number: C0C0,
+    mug_size: u8,
+    temperature: u8,
+    marshmallows: [[u8; 2]],
+}

benches/formats/loco.rs

@@ -0,0 +1,18 @@
+// Copyright 2026 The Fuchsia Authors
+//
+// Licensed under a BSD-style license <LICENSE-BSD>, Apache License, Version 2.0
+// <LICENSE-APACHE or https://www.apache.org/licenses/LICENSE-2.0>, or the MIT
+// license <LICENSE-MIT or https://opensource.org/licenses/MIT>, at your option.
+// This file may not be copied, modified, or distributed except according to
+// those terms.
+
+use zerocopy_derive::*;
+
+#[derive(FromBytes, KnownLayout, Immutable)]
+#[repr(C, align(2))]
+pub struct Packet {
+    magic_number: [u8; 2],
+    mug_size: u8,
+    temperature: u8,
+    marshmallows: [[u8; 2]],
+}

benches/ref_from_bytes.rs

@@ -0,0 +1,17 @@
+// Copyright 2026 The Fuchsia Authors
+//
+// Licensed under a BSD-style license <LICENSE-BSD>, Apache License, Version 2.0
+// <LICENSE-APACHE or https://www.apache.org/licenses/LICENSE-2.0>, or the MIT
+// license <LICENSE-MIT or https://opensource.org/licenses/MIT>, at your option.
+// This file may not be copied, modified, or distributed except according to
+// those terms.
+
+use zerocopy::FromBytes;
+
+#[path = "formats/loco.rs"]
+mod loco;
+
+#[unsafe(no_mangle)]
+fn codegen_test(source: &[u8]) -> Option<&loco::Packet> {
+    FromBytes::ref_from_bytes(source).ok()
+}

benches/ref_from_bytes.x86-64.mca

@@ -0,0 +1,75 @@
+Iterations:        100
+Instructions:      1800
+Total Cycles:      704
+Total uOps:        2000
+
+Dispatch Width:    4
+uOps Per Cycle:    2.84
+IPC:               2.56
+Block RThroughput: 5.0
+
+
+Instruction Info:
+[1]: #uOps
+[2]: Latency
+[3]: RThroughput
+[4]: MayLoad
+[5]: MayStore
+[6]: HasSideEffects (U)
+
+[1]    [2]    [3]    [4]    [5]    [6]    Instructions:
+ 1      1     0.33                        mov	rdx, rsi
+ 1      1     0.33                        cmp	rsi, 4
+ 1      1     0.50                        setb	al
+ 1      1     0.33                        or	al, dil
+ 1      1     0.33                        test	al, 1
+ 1      1     1.00                        je	.LBB5_2
+ 1      0     0.25                        xor	eax, eax
+ 1      1     1.00                  U     ret
+ 1      1     0.50                        lea	rcx, [rdx - 4]
+ 1      1     0.33                        mov	rsi, rcx
+ 1      1     0.33                        and	rsi, -2
+ 1      1     0.33                        add	rsi, 4
+ 1      1     0.50                        shr	rcx
+ 1      0     0.25                        xor	eax, eax
+ 1      1     0.33                        cmp	rdx, rsi
+ 2      2     0.67                        cmove	rdx, rcx
+ 2      2     0.67                        cmove	rax, rdi
+ 1      1     1.00                  U     ret
+
+
+Resources:
+[0]   - SBDivider
+[1]   - SBFPDivider
+[2]   - SBPort0
+[3]   - SBPort1
+[4]   - SBPort4
+[5]   - SBPort5
+[6.0] - SBPort23
+[6.1] - SBPort23
+
+
+Resource pressure per iteration:
+[0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  
+ -      -     5.97   5.98    -     6.05    -      -     
+
+Resource pressure by instruction:
+[0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
+ -      -     0.97   0.01    -     0.02    -      -     mov	rdx, rsi
+ -      -     0.01   0.02    -     0.97    -      -     cmp	rsi, 4
+ -      -     0.03    -      -     0.97    -      -     setb	al
+ -      -     0.01   0.02    -     0.97    -      -     or	al, dil
+ -      -      -     0.98    -     0.02    -      -     test	al, 1
+ -      -      -      -      -     1.00    -      -     je	.LBB5_2
+ -      -      -      -      -      -      -      -     xor	eax, eax
+ -      -      -      -      -     1.00    -      -     ret
+ -      -     0.98   0.02    -      -      -      -     lea	rcx, [rdx - 4]
+ -      -     0.01   0.99    -      -      -      -     mov	rsi, rcx
+ -      -      -     0.98    -     0.02    -      -     and	rsi, -2
+ -      -     0.98   0.01    -     0.01    -      -     add	rsi, 4
+ -      -     0.99    -      -     0.01    -      -     shr	rcx
+ -      -      -      -      -      -      -      -     xor	eax, eax
+ -      -     0.02   0.97    -     0.01    -      -     cmp	rdx, rsi
+ -      -     0.99   0.99    -     0.02    -      -     cmove	rdx, rcx
+ -      -     0.98   0.99    -     0.03    -      -     cmove	rax, rdi
+ -      -      -      -      -     1.00    -      -     ret

benches/ref_from_bytes_with_elems.rs

@@ -0,0 +1,17 @@
+// Copyright 2026 The Fuchsia Authors
+//
+// Licensed under a BSD-style license <LICENSE-BSD>, Apache License, Version 2.0
+// <LICENSE-APACHE or https://www.apache.org/licenses/LICENSE-2.0>, or the MIT
+// license <LICENSE-MIT or https://opensource.org/licenses/MIT>, at your option.
+// This file may not be copied, modified, or distributed except according to
+// those terms.
+
+use zerocopy::FromBytes;
+
+#[path = "formats/loco.rs"]
+mod loco;
+
+#[unsafe(no_mangle)]
+fn codegen_test(source: &[u8], count: usize) -> Option<&loco::Packet> {
+    FromBytes::ref_from_bytes_with_elems(source, count).ok()
+}

benches/ref_from_bytes_with_elems.x86-64.mca

@@ -0,0 +1,65 @@
+Iterations:        100
+Instructions:      1300
+Total Cycles:      439
+Total uOps:        1400
+
+Dispatch Width:    4
+uOps Per Cycle:    3.19
+IPC:               2.96
+Block RThroughput: 3.5
+
+
+Instruction Info:
+[1]: #uOps
+[2]: Latency
+[3]: RThroughput
+[4]: MayLoad
+[5]: MayStore
+[6]: HasSideEffects (U)
+
+[1]    [2]    [3]    [4]    [5]    [6]    Instructions:
+ 1      1     0.33                        movabs	rax, 9223372036854775805
+ 1      1     0.33                        cmp	rdx, rax
+ 2      2     1.00                        seta	cl
+ 1      1     0.33                        mov	rax, rdi
+ 1      1     0.33                        or	dil, cl
+ 1      1     0.33                        test	dil, 1
+ 1      1     1.00                        jne	.LBB5_2
+ 1      1     0.50                        lea	rcx, [2*rdx + 4]
+ 1      1     0.33                        cmp	rsi, rcx
+ 1      1     1.00                        je	.LBB5_3
+ 1      0     0.25                        xor	eax, eax
+ 1      1     0.33                        mov	rdx, rsi
+ 1      1     1.00                  U     ret
+
+
+Resources:
+[0]   - SBDivider
+[1]   - SBFPDivider
+[2]   - SBPort0
+[3]   - SBPort1
+[4]   - SBPort4
+[5]   - SBPort5
+[6.0] - SBPort23
+[6.1] - SBPort23
+
+
+Resource pressure per iteration:
+[0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  
+ -      -     4.32   4.33    -     4.35    -      -     
+
+Resource pressure by instruction:
+[0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
+ -      -      -     0.99    -     0.01    -      -     movabs	rax, 9223372036854775805
+ -      -     0.33   0.67    -      -      -      -     cmp	rdx, rax
+ -      -     1.98    -      -     0.02    -      -     seta	cl
+ -      -     0.01   0.99    -      -      -      -     mov	rax, rdi
+ -      -     1.00    -      -      -      -      -     or	dil, cl
+ -      -     0.99   0.01    -      -      -      -     test	dil, 1
+ -      -      -      -      -     1.00    -      -     jne	.LBB5_2
+ -      -      -     1.00    -      -      -      -     lea	rcx, [2*rdx + 4]
+ -      -     0.01    -      -     0.99    -      -     cmp	rsi, rcx
+ -      -      -      -      -     1.00    -      -     je	.LBB5_3
+ -      -      -      -      -      -      -      -     xor	eax, eax
+ -      -      -     0.67    -     0.33    -      -     mov	rdx, rsi
+ -      -      -      -      -     1.00    -      -     ret

benches/ref_from_prefix.rs

@@ -0,0 +1,20 @@
+// Copyright 2026 The Fuchsia Authors
+//
+// Licensed under a BSD-style license <LICENSE-BSD>, Apache License, Version 2.0
+// <LICENSE-APACHE or https://www.apache.org/licenses/LICENSE-2.0>, or the MIT
+// license <LICENSE-MIT or https://opensource.org/licenses/MIT>, at your option.
+// This file may not be copied, modified, or distributed except according to
+// those terms.
+
+use zerocopy::FromBytes;
+
+#[path = "formats/loco.rs"]
+mod loco;
+
+#[unsafe(no_mangle)]
+fn codegen_test(source: &[u8]) -> Option<&loco::Packet> {
+    match FromBytes::ref_from_prefix(source) {
+        Ok((packet, _rest)) => Some(packet),
+        _ => None,
+    }
+}

benches/ref_from_prefix.x86-64.mca

@@ -0,0 +1,67 @@
+Iterations:        100
+Instructions:      1400
+Total Cycles:      405
+Total uOps:        1400
+
+Dispatch Width:    4
+uOps Per Cycle:    3.46
+IPC:               3.46
+Block RThroughput: 4.0
+
+
+Instruction Info:
+[1]: #uOps
+[2]: Latency
+[3]: RThroughput
+[4]: MayLoad
+[5]: MayStore
+[6]: HasSideEffects (U)
+
+[1]    [2]    [3]    [4]    [5]    [6]    Instructions:
+ 1      0     0.25                        xor	edx, edx
+ 1      1     0.33                        mov	eax, 0
+ 1      1     0.33                        test	dil, 1
+ 1      1     1.00                        jne	.LBB5_4
+ 1      1     0.33                        cmp	rsi, 4
+ 1      1     1.00                        jae	.LBB5_3
+ 1      1     0.33                        mov	edx, 1
+ 1      0     0.25                        xor	eax, eax
+ 1      1     1.00                  U     ret
+ 1      1     0.33                        add	rsi, -4
+ 1      1     0.50                        shr	rsi
+ 1      1     0.33                        mov	rdx, rsi
+ 1      1     0.33                        mov	rax, rdi
+ 1      1     1.00                  U     ret
+
+
+Resources:
+[0]   - SBDivider
+[1]   - SBFPDivider
+[2]   - SBPort0
+[3]   - SBPort1
+[4]   - SBPort4
+[5]   - SBPort5
+[6.0] - SBPort23
+[6.1] - SBPort23
+
+
+Resource pressure per iteration:
+[0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  
+ -      -     3.99   3.99    -     4.02    -      -     
+
+Resource pressure by instruction:
+[0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
+ -      -      -      -      -      -      -      -     xor	edx, edx
+ -      -     0.01   0.98    -     0.01    -      -     mov	eax, 0
+ -      -     0.98   0.02    -      -      -      -     test	dil, 1
+ -      -      -      -      -     1.00    -      -     jne	.LBB5_4
+ -      -     0.02   0.98    -      -      -      -     cmp	rsi, 4
+ -      -      -      -      -     1.00    -      -     jae	.LBB5_3
+ -      -     0.98   0.01    -     0.01    -      -     mov	edx, 1
+ -      -      -      -      -      -      -      -     xor	eax, eax
+ -      -      -      -      -     1.00    -      -     ret
+ -      -     0.01   0.99    -      -      -      -     add	rsi, -4
+ -      -     1.00    -      -      -      -      -     shr	rsi
+ -      -      -     1.00    -      -      -      -     mov	rdx, rsi
+ -      -     0.99   0.01    -      -      -      -     mov	rax, rdi
+ -      -      -      -      -     1.00    -      -     ret

benches/ref_from_prefix_with_elems.rs

@@ -0,0 +1,20 @@
+// Copyright 2026 The Fuchsia Authors
+//
+// Licensed under a BSD-style license <LICENSE-BSD>, Apache License, Version 2.0
+// <LICENSE-APACHE or https://www.apache.org/licenses/LICENSE-2.0>, or the MIT
+// license <LICENSE-MIT or https://opensource.org/licenses/MIT>, at your option.
+// This file may not be copied, modified, or distributed except according to
+// those terms.
+
+use zerocopy::FromBytes;
+
+#[path = "formats/loco.rs"]
+mod loco;
+
+#[unsafe(no_mangle)]
+fn codegen_test(source: &[u8], count: usize) -> Option<&loco::Packet> {
+    match FromBytes::ref_from_prefix_with_elems(source, count) {
+        Ok((packet, _rest)) => Some(packet),
+        _ => None,
+    }
+}