Lower byte swap intrinsic.

tests_lit/llvm2ice_tests/nacl-other-intrinsics.ll

 ; This tests the NaCl intrinsics not related to atomic operations.
 
 ; RUN: %llvm2ice -O2 --verbose none %s | FileCheck %s
 ; RUN: %llvm2ice -Om1 --verbose none %s | FileCheck %s
 
 ; Do another run w/ O2 and a different check-prefix (otherwise O2 and Om1
 ; share the same "CHECK" prefix). This separate run helps check that
 ; some code is optimized out.
 ; RUN: %llvm2ice -O2 --verbose none %s | FileCheck %s --check-prefix=CHECKO2REM
 
 ; RUN: %llvm2ice --verbose none %s | FileCheck --check-prefix=ERRORS %s
 ; RUN: %llvm2iceinsts %s | %szdiff %s | FileCheck --check-prefix=DUMP %s
 ; RUN: %llvm2iceinsts --pnacl %s | %szdiff %s \
 ; RUN:                           | FileCheck --check-prefix=DUMP %s
 
 declare i8* @llvm.nacl.read.tp()
 declare void @llvm.memcpy.p0i8.p0i8.i32(i8*, i8*, i32, i32, i1)
 declare void @llvm.memmove.p0i8.p0i8.i32(i8*, i8*, i32, i32, i1)
 declare void @llvm.memset.p0i8.i32(i8*, i8, i32, i32, i1)
 declare void @llvm.nacl.longjmp(i8*, i32)
 declare i32 @llvm.nacl.setjmp(i8*)
 declare float @llvm.sqrt.f32(float)
 declare double @llvm.sqrt.f64(double)
 declare void @llvm.trap()
+declare i16 @llvm.bswap.i16(i16)
+declare i32 @llvm.bswap.i32(i32)
+declare i64 @llvm.bswap.i64(i64)
 declare i32 @llvm.ctlz.i32(i32, i1)
 declare i64 @llvm.ctlz.i64(i64, i1)
 declare i32 @llvm.cttz.i32(i32, i1)
 declare i64 @llvm.cttz.i64(i64, i1)
 declare i32 @llvm.ctpop.i32(i32)
 declare i64 @llvm.ctpop.i64(i64)
 
 define i32 @test_nacl_read_tp() {
 entry:
   %ptr = call i8* @llvm.nacl.read.tp()
@@ skipping 200 matching lines @@
   br i1 %r1, label %Zero, label %NonZero
 Zero:
   call void @llvm.trap()
   unreachable
 NonZero:
   ret i32 1
 }
 ; CHECK-LABEL: test_trap
 ; CHECK: ud2
 
+define i32 @test_bswap_16(i32 %x) {
+entry:
+  %x_trunc = trunc i32 %x to i16
+  %r = call i16 @llvm.bswap.i16(i16 %x_trunc)
+  %r_zext = zext i16 %r to i32
+  ret i32 %r_zext
+}
+; CHECK-LABEL: test_bswap_16
+; CHECK: rol {{.*}}, 8
+
+define i32 @test_bswap_32(i32 %x) {
+entry:
+  %r = call i32 @llvm.bswap.i32(i32 %x)
+  ret i32 %r
+}
+; CHECK-LABEL: test_bswap_32
+; CHECK: bswap e{{.*}}
+
+define i64 @test_bswap_64(i64 %x) {
+entry:
+  %r = call i64 @llvm.bswap.i64(i64 %x)
+  ret i64 %r
+}
+; CHECK-LABEL: test_bswap_64
+; CHECK: bswap e{{.*}}
+; CHECK: bswap e{{.*}}
+
 define i32 @test_ctlz_32(i32 %x) {
 entry:
   %r = call i32 @llvm.ctlz.i32(i32 %x, i1 0)
   ret i32 %r
 }
 ; CHECK-LABEL: test_ctlz_32
 ; TODO(jvoung): If we detect that LZCNT is supported, then use that
 ; and avoid the need to do the cmovne and xor stuff to guarantee that
 ; the result is well-defined w/ input == 0.
 ; CHECK: bsr [[REG_TMP:e.*]], {{.*}}
@@ skipping 105 matching lines @@
   ret i32 %r
 }
 ; If there is a trunc, then the mov {{.*}}, 0 is dead and gets optimized out.
 ; CHECKO2REM-LABEL: test_popcount_64_ret_i32
 ; CHECKO2REM: call __popcountdi2
 ; CHECKO2REM-NOT: mov {{.*}}, 0
 
 
 ; ERRORS-NOT: ICE translation error
 ; DUMP-NOT: SZ