fix: skip cs nopw 0x0(%rax,%rax,1)

.github/workflows/test.yml

@@ -25,3 +25,21 @@ jobs:
       - name: Run tests
         run: go test -v ./...
         working-directory: example
+
+  cases:
+    name: cases
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v2
+      - uses: actions/setup-go@v4
+      - name: Install clang
+        uses: ConorMacBride/install-package@v1
+        with:
+          apt: clang libc6-dev-i386
+      - name: Install GOAT
+        run: go install .
+      - name: Run tests
+        run: |
+          for file in tests/*.c
+            goat $file -O3 -mavx -mfma
+          done

parser_amd64.go

@@ -16,11 +16,12 @@ package main
 import (
 	"bufio"
 	"fmt"
-	"github.com/klauspost/asmfmt"
 	"os"
 	"regexp"
 	"strings"
 	"unicode"
+
+	"github.com/klauspost/asmfmt"
 )
 
 const buildTags = "//go:build !noasm && amd64\n"
@@ -163,7 +164,8 @@ func parseObjectDump(dump string, functions map[string][]Line) error {
 			if assembly == "" {
 				return fmt.Errorf("try to increase --insn-width of objdump")
 			} else if strings.HasPrefix(assembly, "nop") ||
-				assembly == "xchg   %ax,%ax" {
+				assembly == "xchg   %ax,%ax" ||
+				assembly == "cs nopw 0x0(%rax,%rax,1)" {
 				continue
 			}
 			if lineNumber >= len(functions[functionName]) {

tests/cs_npw.c

@@ -0,0 +1,15 @@
+void l2(float *a, float *b, float *res, long *len)
+{
+    int size = *len;
+
+    float sum = 0;
+    // add the remaining vectors
+    for (int i = 0; i < size; i++)
+    {
+        float diff = a[i] - b[i];
+        float sq = diff * diff;
+        sum += sq;
+    }
+
+    *res = sum;
+}