diff --git a/.github/workflows/build-and-test-aarch64.yml b/.github/workflows/build-and-test-aarch64.yml
deleted file mode 100644
index eb9552f26..000000000
--- a/.github/workflows/build-and-test-aarch64.yml
+++ /dev/null
@@ -1,61 +0,0 @@
-name: build and test on aarch64 with QEMU
-on: [push, pull_request]
-jobs:
-  test-on-qemu-linux-aarch64:
-    runs-on: ubuntu-latest
-    name: test on debian-bullseye-aarch64
-    steps:
-      - name: install prerequisites
-        uses: awalsh128/cache-apt-pkgs-action@latest
-        with:
-          packages: g++-aarch64-linux-gnu libc6-dev-arm64-cross
-          version: 1.0 # version of cache to load
-      - name: git checkout
-        uses: actions/checkout@v3
-        with:
-          fetch-depth: 1
-      - name: set up qemu
-        uses: docker/setup-qemu-action@v2
-        with:
-          platforms: arm64
-      - name: cache rust dependencies
-        uses: actions/cache@v3
-        with:
-          path: |
-            ~/.cargo/bin/
-            ~/.cargo/registry/index/
-            ~/.cargo/registry/cache/
-            ~/.cargo/git/db/
-            target/
-          key: aarch64-unknown-linux-gnu-cargo-and-target-${{ hashFiles('**/Cargo.lock') }}
-      # - name: cache dav1d object files
-      #   uses: actions/cache@v3
-      #   with:
-      #     path: build/
-      #     key: aarch64-unknown-linux-gnu-c-object-files-${{ hashFiles('**/meson.build', '**/*.c', '**/*.h') }}
-      - name: cargo build for aarch64-unknown-linux-gnu
-        run: |
-          rustup target add aarch64-unknown-linux-gnu
-          cargo build --release --target aarch64-unknown-linux-gnu
-      - name: docker pull
-        run: docker pull ghcr.io/immunant/rav1d/debian-bullseye-aarch64:latest
-      - name: build and run tests in docker
-        run: |
-          docker run \
-            --rm \
-            -v ~/.cargo/git:/home/prossimo/.cargo/git \
-            -v ~/.cargo/registry:/home/prossimo/.cargo/registry \
-            -v $(pwd):/${{ github.workspace }} \
-            -w ${{ github.workspace }} \
-            --platform linux/arm64 \
-            ghcr.io/immunant/rav1d/debian-bullseye-aarch64:latest \
-            .github/workflows/test.sh -t 2 \
-                -r ../target/aarch64-unknown-linux-gnu/release/dav1d \
-                -s ../target/aarch64-unknown-linux-gnu/release/seek_stress
-      - name: upload build artifacts
-        if: ${{ !cancelled() }}
-        uses: actions/upload-artifact@v3
-        with:
-          name: meson-test-logs
-          path: |
-             ${{ github.workspace }}/build/meson-logs/testlog.txt
\ No newline at end of file
diff --git a/.github/workflows/build-and-test-arm.yml b/.github/workflows/build-and-test-arm.yml
new file mode 100644
index 000000000..08a37e62f
--- /dev/null
+++ b/.github/workflows/build-and-test-arm.yml
@@ -0,0 +1,95 @@
+name: build and test on arm with QEMU
+on: [push, pull_request]
+jobs:
+  test-on-qemu:
+    strategy:
+      matrix:
+        target: [
+          {
+            rust: "aarch64-unknown-linux-gnu",
+            gcc: "g++-aarch64-linux-gnu",
+            libc_cross: "libc6-dev-arm64-cross",
+            docker_image: "debian-bullseye-aarch64",
+            docker_platform: "linux/arm64",
+          },
+          {
+            rust: "armv7-unknown-linux-gnueabihf",
+            gcc: "g++-arm-linux-gnueabihf",
+            libc_cross: "libc6-dev-armhf-cross",
+            docker_image: "debian-bullseye-arm7",
+            docker_platform: "linux/arm/v7",
+          },
+        ]
+        build: [
+          {profile: "checked-release", features: "asm", timeout_multiplier: 2},
+          # The asm fallback "should" be the same on all platforms,
+          # so if we tested it for x86, we shouldn't need to for arm either
+          # (which is slower due to qemu).
+          # {profile: "checked-release", features: ""   , timeout_multiplier: 2},
+        ]
+    runs-on: ubuntu-latest
+    name: meson tests on ${{ matrix.target.rust }} (qemu) with --features ${{ matrix.build.features }}
+    steps:
+      - name: install prerequisites
+        uses: awalsh128/cache-apt-pkgs-action@latest
+        with:
+          packages: ${{ matrix.target.gcc }} ${{ matrix.target.libc_cross }}
+          version: 1.0 # version of cache to load
+      - name: git checkout
+        uses: actions/checkout@v3
+        with:
+          fetch-depth: 1
+      - name: set up qemu
+        uses: docker/setup-qemu-action@v2
+        with:
+          platforms: arm64
+      - name: cache rust dependencies
+        uses: actions/cache@v3
+        with:
+          path: |
+            ~/.cargo/bin/
+            ~/.cargo/registry/index/
+            ~/.cargo/registry/cache/
+            ~/.cargo/git/db/
+            target/
+          key: ${{ matrix.target.rust }}-cargo-and-target-${{ hashFiles('**/Cargo.lock') }}
+      # - name: cache dav1d object files
+      #   uses: actions/cache@v3
+      #   with:
+      #     path: build/
+      #     key: ${{ matrix.target.rust }}-c-object-files-${{ hashFiles('**/meson.build', '**/*.c', '**/*.h') }}
+      - name: cargo build
+        run: |
+          rustup target add ${{ matrix.target.rust }}
+          cargo build \
+            --target ${{ matrix.target.rust }} \
+            --profile ${{ matrix.build.profile }} \
+            --no-default-features --features=bitdepth_8,bitdepth_16,${{ matrix.build.features }}
+      - name: docker pull
+        run: docker pull ghcr.io/immunant/rav1d/${{ matrix.target.docker_image }}:latest
+      - name: build and run tests in docker
+        run: |
+          docker run \
+            --rm \
+            -v ~/.cargo/git:/home/prossimo/.cargo/git \
+            -v ~/.cargo/registry:/home/prossimo/.cargo/registry \
+            -v $(pwd):/${{ github.workspace }} \
+            -w ${{ github.workspace }} \
+            --platform ${{ matrix.target.docker_platform }} \
+            ghcr.io/immunant/rav1d/${{ matrix.target.docker_image }}:latest \
+            .github/workflows/test.sh \
+                -r ../target/${{ matrix.target.rust }}/${{ matrix.build.profile }}/dav1d \
+                -s ../target/${{ matrix.target.rust }}/${{ matrix.build.profile }}/seek_stress \
+                -t ${{ matrix.build.timeout_multiplier }}
+      - name: copy log files
+        if: ${{ !cancelled() }}
+        run: |
+          cp ${{ github.workspace }}/build/meson-logs/testlog.txt \
+             ${{ github.workspace }}/build/meson-logs/testlog-${{ matrix.target }}-${{ matrix.build.features }}.txt
+      - name: upload build artifacts
+        if: ${{ !cancelled() }}
+        uses: actions/upload-artifact@v3
+        with:
+          name: meson-test-logs
+          path: |
+             ${{ github.workspace }}/build/meson-logs/testlog-*.txt
diff --git a/.github/workflows/build-and-test-arm7.yml b/.github/workflows/build-and-test-arm7.yml
deleted file mode 100644
index ddde93e21..000000000
--- a/.github/workflows/build-and-test-arm7.yml
+++ /dev/null
@@ -1,61 +0,0 @@
-name: build and test on arm7 with QEMU
-on: [push, pull_request]
-jobs:
-  test-on-qemu-linux-arm7:
-    runs-on: ubuntu-latest
-    name: test on debian-bullseye-arm7
-    steps:
-      - name: install prerequisites
-        uses: awalsh128/cache-apt-pkgs-action@latest
-        with:
-          packages: g++-arm-linux-gnueabihf libc6-dev-armhf-cross
-          version: 1.0 # version of cache to load
-      - name: git checkout
-        uses: actions/checkout@v3
-        with:
-          fetch-depth: 1
-      - name: set up qemu
-        uses: docker/setup-qemu-action@v2
-        with:
-          platforms: arm
-      - name: cache rust dependencies
-        uses: actions/cache@v3
-        with:
-          path: |
-            ~/.cargo/bin/
-            ~/.cargo/registry/index/
-            ~/.cargo/registry/cache/
-            ~/.cargo/git/db/
-            target/
-          key: arm7-unknown-linux-gnu-cargo-and-target-${{ hashFiles('**/Cargo.lock') }}
-      # - name: cache dav1d object files
-      #   uses: actions/cache@v3
-      #   with:
-      #     path: build/
-      #     key: aarch64-unknown-linux-gnu-c-object-files-${{ hashFiles('**/meson.build', '**/*.c', '**/*.h') }}
-      - name: cargo build for armv7-unknown-linux-gnueabihf
-        run: |
-          rustup target add armv7-unknown-linux-gnueabihf
-          cargo build --release --target armv7-unknown-linux-gnueabihf
-      - name: docker pull
-        run: docker pull ghcr.io/immunant/rav1d/debian-bullseye-arm7:latest
-      - name: build and run tests in docker
-        run: |
-          docker run \
-            --rm \
-            -v ~/.cargo/git:/home/prossimo/.cargo/git \
-            -v ~/.cargo/registry:/home/prossimo/.cargo/registry \
-            -v $(pwd):/${{ github.workspace }} \
-            -w ${{ github.workspace }} \
-            --platform linux/arm/v7 \
-            ghcr.io/immunant/rav1d/debian-bullseye-arm7:latest \
-            .github/workflows/test.sh -t 2 \
-                -r ../target/armv7-unknown-linux-gnueabihf/release/dav1d \
-                -s ../target/armv7-unknown-linux-gnueabihf/release/seek_stress
-      - name: upload build artifacts
-        if: ${{ !cancelled() }}
-        uses: actions/upload-artifact@v3
-        with:
-          name: meson-test-logs
-          path: |
-             ${{ github.workspace }}/build/meson-logs/testlog.txt
\ No newline at end of file
diff --git a/.github/workflows/build-and-test-x86-extra.yml b/.github/workflows/build-and-test-x86-extra.yml
index 816e02010..6dea0422c 100644
--- a/.github/workflows/build-and-test-x86-extra.yml
+++ b/.github/workflows/build-and-test-x86-extra.yml
@@ -12,15 +12,14 @@ jobs:
     strategy:
       matrix:
         target: [
-          "x86_64-unknown-linux-gnu"
+          "x86_64-unknown-linux-gnu",
         ]
-        build: [ # we do release with overflow checks since debug takes forever
-          # release build with overflow checks without optimized assembly routines
-          {name: "release", flags: "--release --no-default-features --features=bitdepth_8,bitdepth_16"},
-          # release build with overflow checks and optimized assembly routines
-          {name: "release", flags: "--release"}
+        build: [
+          {profile: "checked-release", features: "asm"},
+          {profile: "checked-release", features: ""},
         ]
     runs-on: ubuntu-latest
+    name: argon tests on ${{ matrix.target }} with --features ${{ matrix.build.features }}
     steps:
       - name: install prerequisites
         uses: awalsh128/cache-apt-pkgs-action@latest
@@ -58,12 +57,13 @@ jobs:
           key: argon_coveragetool_av1_base_and_extended_profiles_v2.1.1.zip
           restore-keys: | # it is perfectly fine to restore from other branches
             argon_coveragetool_av1_base_and_extended_profiles_v2.1.1.zip
-      - name: cargo build for ${{ matrix.target }} ${{ matrix.build.name }}
+      - name: cargo build
         run: |
           rustup target add ${{ matrix.target }}
-          cargo build --target ${{ matrix.target }} ${{ matrix.build.flags }}
-        env:
-          RUSTFLAGS: "-C overflow-checks=on"
+          cargo build \
+            --target ${{ matrix.target }} \
+            --profile ${{ matrix.build.profile }} \
+            --no-default-features --features=bitdepth_8,bitdepth_16,${{ matrix.build.features }}
       - name: download, check, and unpack argon test vectors
         if: ${{ steps.cache-argon.outputs.cache-hit != 'true' }}
         run: | # delete the downloaded .zip to avoid out-of-diskspace errors
@@ -76,7 +76,6 @@ jobs:
         env:
           ARGON_URL: https://storage.googleapis.com/downloads.aomedia.org/assets/zip/argon_coveragetool_av1_base_and_extended_profiles_v2.1.1.zip
           ARGON_URL_MD5: https://storage.googleapis.com/downloads.aomedia.org/assets/zip/argon_coveragetool_av1_base_and_extended_profiles_v2.1.1.zip.md5sum
-      - name: run argon tests for ${{ matrix.target }} ${{ matrix.build.name }}
+      - name: run argon tests
         run: |
-          tests/dav1d_argon.bash -d target/${{ matrix.target }}/${{ matrix.build.name }}/dav1d -j `nproc`
-
+          tests/dav1d_argon.bash -d target/${{ matrix.target }}/${{ matrix.build.profile }}/dav1d -j $(nproc)
diff --git a/.github/workflows/build-and-test-x86.yml b/.github/workflows/build-and-test-x86.yml
index fbca33900..227ff3178 100644
--- a/.github/workflows/build-and-test-x86.yml
+++ b/.github/workflows/build-and-test-x86.yml
@@ -6,15 +6,14 @@ jobs:
       matrix:
         target: [
           "x86_64-unknown-linux-gnu",
-          "i686-unknown-linux-gnu"
+          "i686-unknown-linux-gnu",
         ]
         build: [
-          # release build without optimized assembly routines
-          {name: "release", flags: "--release --no-default-features --features=bitdepth_8,bitdepth_16", timeout_multiplier: 1},
-          # debug build to catch overflows with optimized assembly routines
-          {name: "debug", flags: "", timeout_multiplier: 2}
+          {profile: "checked-release", features: "asm", timeout_multiplier: 2},
+          {profile: "checked-release", features: ""   , timeout_multiplier: 2},
         ]
     runs-on: ubuntu-latest
+    name: meson tests on ${{ matrix.target }} with --features ${{ matrix.build.features }}
     steps:
       - name: install prerequisites
         uses: awalsh128/cache-apt-pkgs-action@latest
@@ -43,22 +42,25 @@ jobs:
             ~/.cargo/git/db/
             target/
           key: ${{ runner.os }}-cargo-${{ hashFiles('**/Cargo.lock') }}
-      - name: cargo build for ${{ matrix.target }} ${{ matrix.build.name }}
+      - name: cargo build
         run: |
           cargo clean
           rustup target add ${{ matrix.target }}
-          cargo build --target ${{ matrix.target }} ${{ matrix.build.flags }}
-      - name: meson test for ${{ matrix.target }} ${{ matrix.build.name }}
+          cargo build \
+            --target ${{ matrix.target }} \
+            --profile ${{ matrix.build.profile }} \
+            --no-default-features --features=bitdepth_8,bitdepth_16,${{ matrix.build.features }}
+      - name: meson test
         run: |
           .github/workflows/test.sh \
-              -r ../target/${{ matrix.target }}/${{ matrix.build.name }}/dav1d \
-              -s ../target/${{ matrix.target }}/${{ matrix.build.name }}/seek_stress \
+              -r ../target/${{ matrix.target }}/${{ matrix.build.profile }}/dav1d \
+              -s ../target/${{ matrix.target }}/${{ matrix.build.profile }}/seek_stress \
               -t ${{ matrix.build.timeout_multiplier }}
       - name: copy log files
         if: ${{ !cancelled() }}
         run: |
           cp ${{ github.workspace }}/build/meson-logs/testlog.txt \
-             ${{ github.workspace }}/build/meson-logs/testlog-${{ matrix.target }}-${{ matrix.build.name }}.txt
+             ${{ github.workspace }}/build/meson-logs/testlog-${{ matrix.target }}-${{ matrix.build.features }}.txt
       - name: upload build artifacts
         if: ${{ !cancelled() }}
         uses: actions/upload-artifact@v3
@@ -95,18 +97,18 @@ jobs:
             ~/.cargo/git/db/
             target/
           key: ${{ runner.os }}-cargo-${{ hashFiles('**/Cargo.lock') }}
-      - name: cargo build for x86_64-apple-darwin
+      - name: cargo build for x86_64-apple-darwin with --features asm
         run: |
-          cargo build --release
-      - name: meson test for x86_64-apple-darwin
+          cargo build --profile checked-release
+      - name: meson test for x86_64-apple-darwin with --features asm
         run: |
-          .github/workflows/test.sh -r ../target/release/dav1d \
-              -s ../target/release/seek_stress
+          .github/workflows/test.sh -r ../target/checked-release/dav1d \
+              -s ../target/checked-release/seek_stress
       - name: copy log files
         if: ${{ !cancelled() }}
         run: |
           cp ${{ github.workspace }}/build/meson-logs/testlog.txt \
-              ${{ github.workspace }}/build/meson-logs/testlog-x86_64-apple-darwin.txt
+              ${{ github.workspace }}/build/meson-logs/testlog-x86_64-apple-darwin-asm.txt
       - name: upload build artifacts
         if: ${{ !cancelled() }}
         uses: actions/upload-artifact@v3
diff --git a/Cargo.toml b/Cargo.toml
index 03e25f442..4889422ba 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -43,12 +43,40 @@ asm = []
 bitdepth_8 = []
 bitdepth_16 = []
 
+# For debugging.
+#
+# This enables no optimizations, so optimizations can't interfere with debugging,
+# but it will probably be very slow.
 [profile.dev]
-# As we port, we increasingly use operations that require more optimizations to be zero-cost,
-# so while `--release` perf stays the same, debug perf suffers a lot,
-# unless we enable the basic optimizations (`opt-level = 1`),
-# which don't increase compile-time that much.
+
+# Fastest for local development.
+#
+# `debug`/`dev` builds are very slow*, so this enables some optimizations
+# to be fastest for the local development cycle of compiling and running tests.
+# Only some optimizations are enabled, so compile time doesn't increase very much.
+# Full debug info is still enabled, so this can also be used for debugging,
+# but there are still some cases where optimizations can still interfere with debugging.
+#
+# * Especially as we increasingly use more zero-cost abstractions,
+# which are only zero-cost with optimizations.
+[profile.fast-dev]
+inherits = "dev"
 opt-level = 1
 
+# Fastest.
+# For release builds.
 [profile.release]
 codegen-units = 1
+
+# Fast but checked.
+# For CI and sometimes local development with qemu.
+#
+# That is, it compiles with all runtime checks and limited debug info (as opposed to none).
+[profile.checked-release]
+inherits = "release"
+opt-level = 3
+debug = "limited"
+debug-assertions = true
+overflow-checks = true
+lto = "thin"
+codegen-units = 16      # release default