Subzero: Add fabs intrinsic support.

tests_lit/llvm2ice_tests/nacl-other-intrinsics.ll

 ; This tests the NaCl intrinsics not related to atomic operations.
 
 ; RUN: %p2i -i %s --filetype=obj --disassemble --args -O2 -sandbox \
 ; RUN:   | FileCheck %s
 ; RUN: %p2i -i %s --filetype=obj --disassemble --args -Om1 -sandbox \
 ; RUN:   | 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: %p2i -i %s --filetype=obj --disassemble --args -O2 -sandbox \
 ; RUN:   | FileCheck --check-prefix=CHECKO2REM %s
 
 ; Do O2 runs without -sandbox to make sure llvm.nacl.read.tp gets
 ; lowered to __nacl_read_tp instead of gs:0x0.
 ; We also know that because it's O2, it'll have the O2REM optimizations.
 ; RUN: %p2i -i %s --filetype=obj --disassemble --args -O2 \
 ; RUN:   | FileCheck --check-prefix=CHECKO2UNSANDBOXEDREM %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 float @llvm.fabs.f32(float)
+declare double @llvm.fabs.f64(double)
+declare <4 x float> @llvm.fabs.v4f32(<4 x float>)
 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)
@@ skipping 223 matching lines @@
 define float @test_sqrt_ignored(float %x, double %y) {
 entry:
   %ignored1 = call float @llvm.sqrt.f32(float %x)
   %ignored2 = call double @llvm.sqrt.f64(double %y)
   ret float 0.0
 }
 ; CHECKO2REM-LABEL: test_sqrt_ignored
 ; CHECKO2REM-NOT: sqrtss
 ; CHECKO2REM-NOT: sqrtsd
 
+define float @test_fabs_float(float %x) {
+entry:
+  %r = call float @llvm.fabs.f32(float %x)
+  %r2 = call float @llvm.fabs.f32(float %r)
+  %r3 = call float @llvm.fabs.f32(float -0.0)
+  %r4 = fadd float %r2, %r3
+  ret float %r4
+}
+; CHECK-LABEL: test_fabs_float
+; CHECK: pcmpeqd
+; CHECK: psrld
+; CHECK: pand
+; CHECK: pcmpeqd
+; CHECK: psrld
+; CHECK: pand
+; CHECK: pcmpeqd
+; CHECK: psrld
+; CHECK: pand
+
+define double @test_fabs_double(double %x) {
+entry:
+  %r = call double @llvm.fabs.f64(double %x)
+  %r2 = call double @llvm.fabs.f64(double %r)
+  %r3 = call double @llvm.fabs.f64(double -0.0)
+  %r4 = fadd double %r2, %r3
+  ret double %r4
+}
+; CHECK-LABEL: test_fabs_double
+; CHECK: pcmpeqd
+; CHECK: psrlq
+; CHECK: pand
+; CHECK: pcmpeqd
+; CHECK: psrlq
+; CHECK: pand
+; CHECK: pcmpeqd
+; CHECK: psrlq
+; CHECK: pand
+
+define <4 x float> @test_fabs_v4f32(<4 x float> %x) {
+entry:
+  %r = call <4 x float> @llvm.fabs.v4f32(<4 x float> %x)
+  %r2 = call <4 x float> @llvm.fabs.v4f32(<4 x float> %r)
+  %r3 = call <4 x float> @llvm.fabs.v4f32(<4 x float> undef)
+  %r4 = fadd <4 x float> %r2, %r3
+  ret <4 x float> %r4
+}
+; CHECK-LABEL: test_fabs_v4f32
+; CHECK: pcmpeqd
+; CHECK: psrld
+; CHECK: pand
+; CHECK: pcmpeqd
+; CHECK: psrld
+; CHECK: pand
+; CHECK: pcmpeqd
+; CHECK: psrld
+; CHECK: pand
+
 define i32 @test_trap(i32 %br) {
 entry:
   %r1 = icmp eq i32 %br, 0
   br i1 %r1, label %Zero, label %NonZero
 Zero:
   call void @llvm.trap()
   unreachable
 NonZero:
   ret i32 1
 }
@@ skipping 192 matching lines @@
 
   call void @llvm.stackrestore(i8* %sp1)
   ret void
 }
 ; CHECK-LABEL: test_stacksave_multiple
 ; At least 3 copies of esp, but probably more from having to do the allocas.
 ; CHECK: mov {{.*}},esp
 ; CHECK: mov {{.*}},esp
 ; CHECK: mov {{.*}},esp
 ; CHECK: mov esp,{{.*}}