fix: build for cuda 13 + mac (#1844)

* fix: build for cuda 13 + mac

* ci: build reactant jll

* ci: pass in enzyme_jax_commit to reduce builds

* Apply suggestion from @avik-pal

* Apply suggestion from @avik-pal

Author: avik-pal

.github/workflows/build-reactantjll.yml

@@ -0,0 +1,46 @@
+name: "Build Reactant_jll"
+
+on:
+  pull_request:
+    branches:
+      - main
+    paths:
+      - ".github/workflows/build-reactantjll.yml"
+      - "deps/ReactantExtra/API.cpp"
+      - "deps/ReactantExtra/BUILD"
+      - "deps/ReactantExtra/WORKSPACE"
+      - "deps/ReactantExtra/workspace.bzl"
+
+concurrency:
+  # Skip intermediate builds: always.
+  # Cancel intermediate builds: only if it is a pull request build.
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: ${{ startsWith(github.ref, 'refs/pull/') }}
+
+jobs:
+  enzyme-jax-commit:
+    name: Extract ENZYMEXLA_COMMIT from WORKSPACE
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Extract ENZYMEXLA_COMMIT from WORKSPACE
+        id: extract_enzyme_jax_commit
+        run: |
+          ENZYMEXLA_COMMIT=$(grep -oP 'ENZYMEXLA_COMMIT = "\K[^"]+' deps/ReactantExtra/WORKSPACE)
+          echo "enzyme_jax_commit=$ENZYMEXLA_COMMIT" >> $GITHUB_OUTPUT
+    outputs:
+      enzyme_jax_commit: ${{ steps.extract_enzyme_jax_commit.outputs.enzyme_jax_commit }}
+
+  build-jll:
+    name: Build Reactant_jll
+    if: github.event.pull_request.draft == false
+    uses: EnzymeAD/ReactantBuilder/.github/workflows/build-reactant-reusable.yml@main
+    needs: enzyme-jax-commit
+    with:
+      reactantbuilder_ref: "main"
+      reactant_commit: ${{ github.event.pull_request.head.sha }}
+      enzyme_jax_commit: ${{ needs.enzyme-jax-commit.outputs.enzyme_jax_commit }}

deps/ReactantExtra/API.cpp

@@ -914,8 +914,9 @@ CudaGetStreamExecutorDeviceDescription(int32_t device_id) {
 
   // Memory bandwidth (bytes/sec) ≈ 2 * memClock(Hz) * busWidth(bytes)
   // props.memoryClockRate is in kHz; bus width is in bits.
-  const double mem_clock_hz =
-      static_cast<double>(props.memoryClockRate) * 1000.0;
+  const double mem_clock_hz = static_cast<double>(GetCudaIntegerAttribute(
+                                  cudaDevAttrMemoryClockRate, device_id)) *
+                              1000.0;
   const double bus_bytes = static_cast<double>(props.memoryBusWidth) / 8.0;
   const double bandwidth_Bps = 2.0 * mem_clock_hz * bus_bytes; // DDR assumption
   device_description->set_memory_bandwidth(
@@ -925,8 +926,10 @@ CudaGetStreamExecutorDeviceDescription(int32_t device_id) {
       GetCudaIntegerAttribute(cudaDevAttrL2CacheSize, device_id));
 
   // SM clock (GHz). props.clockRate is kHz.
-  device_description->set_clock_rate_ghz(static_cast<double>(props.clockRate) /
-                                         1.0e6);
+  device_description->set_clock_rate_ghz(
+      static_cast<double>(
+          GetCudaIntegerAttribute(cudaDevAttrClockRate, device_id)) /
+      1.0e6);
   device_description->set_device_memory_size(props.totalGlobalMem);
 
   // Registers
@@ -3480,7 +3483,7 @@ REACTANT_ABI void EstimateRunTimeForInstruction(
 
 #else
 
-REACTANT_ABI void *CreateGPUPerformanceModelWrapper(
+REACTANT_ABI void *CreateGPUPerformanceModel(
     MlirContext ctx, stream_executor::DeviceDescription *device_description) {
   return nullptr;
 }