Check for and require AVX for CNNScoreVariants (#5291)

README.md

@@ -73,13 +73,10 @@ releases of the toolkit.
    docker client, which can be found on the [docker website](https://www.docker.com/get-docker).
 * Python Dependencies:<a name="python"></a>
     * GATK4 uses the [Conda](https://conda.io/docs/index.html) package manager to establish and manage the
-      Python environment and dependencies required by GATK tools that have a Python dependency. There are two different
-      conda environments that can be used:
-        * The ```gatk``` environment, which has no special hardware requirements. The GATK Docker image comes with the
-          "gatk" environment pre-configured.
-        * The ```gatk-intel``` environment, which requires and uses Intel (AVX2 or AVX-512) hardware acceleration to
-          increase performance.
-    * To establish the conda environment when not using the Docker image, a conda environment must first be "created", and
+      Python environment and dependencies required by GATK tools that have a Python dependency. The ```gatk``` environment, 
+      requires hardware with AVX support for tools that depend on TensorFlow (e.g. CNNScoreVariant). The GATK Docker image 
+      comes with the ```gatk``` environment pre-configured.
+    * To establish the  environment when not using the Docker image, a conda environment must first be "created", and
       then "activated":
         * First, make sure [Miniconda or Conda](https://conda.io/docs/index.html) is installed (Miniconda is sufficient).
         * To "create" the conda environment:
@@ -273,7 +270,7 @@ You can download and run pre-built versions of GATK4 from the following places:
 
   * Examples:
 
-      ```
+      ```      
       ./gatk PrintReadsSpark \
           -I gs://my-gcs-bucket/path/to/input.bam \
           -O gs://my-gcs-bucket/path/to/output.bam \

build.gradle

@@ -522,24 +522,10 @@ task condaStandardEnvironmentDefinition(type: Copy) {
     }
 }
 
-task condaIntelEnvironmentDefinition(type: Copy) {
-    from "scripts"
-    into buildDir
-    include gatkCondaTemplate
-    rename { file -> gatkCondaIntelYML }
-    expand(["condaEnvName":"gatk-intel",
-            "condaEnvDescription" : "Conda environment for GATK Python Tools running with Intel hardware acceleration",
-            "tensorFlowDependency" :
-                    "https://anaconda.org/intel/tensorflow/$tensorflowVersion/download/tensorflow-$tensorflowVersion-cp36-cp36m-linux_x86_64.whl"])
-    doLast {
-        logger.lifecycle("Created Intel Conda environment yml file: $gatkCondaIntelYML")
-    }
-}
 
-// Create two GATK conda environment yml files from the conda enc template
-// (one for standard GATK and one for running GATK with Intel hardware).
+// Create GATK conda environment yml file from the conda enc template
 task condaEnvironmentDefinition() {
-    dependsOn 'pythonPackageArchive', 'condaStandardEnvironmentDefinition', 'condaIntelEnvironmentDefinition'
+    dependsOn 'pythonPackageArchive', 'condaStandardEnvironmentDefinition'
 }
 
 // Create the Python package archive file

scripts/gatkcondaenv.yml.template

@@ -47,4 +47,4 @@ dependencies:
   - theano==0.9.0
   - tqdm==4.19.4
   - werkzeug==0.12.2
-  - gatkPythonPackageArchive.zip
+  - gatkPythonPackageArchive.zip

src/main/java/org/broadinstitute/hellbender/tools/walkers/vqsr/CNNScoreVariants.java

@@ -20,8 +20,11 @@
 import org.broadinstitute.hellbender.utils.runtime.AsynchronousStreamWriter;
 import org.broadinstitute.hellbender.utils.variant.GATKVCFConstants;
 import org.broadinstitute.hellbender.utils.variant.GATKVCFHeaderLines;
+import org.broadinstitute.hellbender.exceptions.UserException;
 import picard.cmdline.programgroups.VariantFilteringProgramGroup;
 
+import com.intel.gkl.IntelGKLUtils;
+
 import java.io.*;
 import java.util.*;
 
@@ -108,6 +111,10 @@ public class CNNScoreVariants extends TwoPassVariantWalker {
             "1D models will look at the reference sequence and variant annotations." +
             "2D models look at aligned reads, reference sequence, and variant annotations." +
             "2D models require a BAM file as input as well as the tensor-type argument to be set.";
+    static final String DISABLE_AVX_CHECK_NAME = "disable-avx-check";
+    static final String AVXREQUIRED_ERROR = "This tool requires AVX instruction set support by default due to its dependency on recent versions of the TensorFlow library.\n" +
+            " If you have an older (pre-1.6) version of TensorFlow installed that does not require AVX you may attempt to re-run the tool with the %s argument to bypass this check.\n" +
+            " Note that such configurations are not officially supported.";
 
     private static final int CONTIG_INDEX = 0;
     private static final int POS_INDEX = 1;
@@ -160,6 +167,11 @@ public class CNNScoreVariants extends TwoPassVariantWalker {
     @Argument(fullName = "output-tensor-dir", shortName = "output-tensor-dir", doc = "Optional directory where tensors can be saved for debugging or visualization.", optional = true)
     private String outputTensorsDir = "";
 
+    @Advanced
+    @Argument(fullName = DISABLE_AVX_CHECK_NAME, shortName = DISABLE_AVX_CHECK_NAME, doc = "If set, no check will be made for AVX support.  " +
+            "Use only if you have installed a pre-1.6 TensorFlow build. ", optional = true)
+    private boolean disableAVXCheck = false;
+
     @Hidden
     @Argument(fullName = "enable-journal", shortName = "enable-journal", doc = "Enable streaming process journal.", optional = true)
     private boolean enableJournal = false;
@@ -232,8 +244,14 @@ public List<ReadFilter> getDefaultReadFilters() {
 
     @Override
     public void onTraversalStart() {
-        if (getHeaderForVariants().getGenotypeSamples().size() > 1) {
-            logger.warn("CNNScoreVariants is a single sample tool, but the input VCF has more than 1 sample.");
+        // Users can disable the AVX check to allow an older version of TF that doesn't require AVX to be used.
+        if(this.disableAVXCheck == false) {
+            IntelGKLUtils utils = new IntelGKLUtils();
+            utils.load(null);
+            if (utils.isAvxSupported() == false) {
+                // Give user the bad news, suggest remedies.
+                throw new UserException.HardwareFeatureException(String.format(CNNScoreVariants.AVXREQUIRED_ERROR, DISABLE_AVX_CHECK_NAME));
+            }
         }
 
         // Start the Python process and initialize a stream writer for streaming data to the Python code