[SDAG] Honor signed arguments in floating point libcalls (#109134)

In ExpandFPLibCall, an assumption is made that all floating point
libcalls that take integer arguments use unsigned integers. In the case
of ldexp and frexp, this assumption is incorrect, leading to
miscompilation and subsequent target-dependent incorrect operation.

Indicate that ldexp and frexp utilize signed arguments in
ExpandFPLibCall.

Fixes #108904

Signed-off-by: Timothy Pearson <tpearson@solidsilicon.com>
(cherry picked from commit 90c14748638f1e10e31173b145fdbb5c4529c922)

diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
index 7f5b46af01c62fe49361987bc2e1f79d3bd46536..4b25f553ffae91ebf8ccce02b38ca3eeb90560bc 100644 (file)
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
@@ -2190,7 +2190,8 @@ void SelectionDAGLegalize::ExpandFPLibCall(SDNode* Node,
     Results.push_back(Tmp.first);
     Results.push_back(Tmp.second);
   } else {
-    SDValue Tmp = ExpandLibCall(LC, Node, false).first;
+    bool IsSignedArgument = Node->getOpcode() == ISD::FLDEXP;
+    SDValue Tmp = ExpandLibCall(LC, Node, IsSignedArgument).first;
     Results.push_back(Tmp);
   }
 }

diff --git a/llvm/test/CodeGen/PowerPC/ldexp-libcall.ll b/llvm/test/CodeGen/PowerPC/ldexp-libcall.ll
index 6144a9d920365152f1ba1f4a3bf68238d1043f22..e531516c37e87ef9fb975bfeeea0e5be3cbb5e00 100644 (file)
--- a/llvm/test/CodeGen/PowerPC/ldexp-libcall.ll
+++ b/llvm/test/CodeGen/PowerPC/ldexp-libcall.ll
@@ -10,7 +10,7 @@ define float @call_ldexpf(float %a, i32 %b) {
 ; CHECK-NEXT:    std r0, 48(r1)
 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-NEXT:    .cfi_offset lr, 16
-; CHECK-NEXT:    clrldi r4, r4, 32
+; CHECK-NEXT:    extsw r4, r4
 ; CHECK-NEXT:    bl ldexpf
 ; CHECK-NEXT:    nop
 ; CHECK-NEXT:    addi r1, r1, 32
@@ -29,7 +29,7 @@ define double @call_ldexp(double %a, i32 %b) {
 ; CHECK-NEXT:    std r0, 48(r1)
 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-NEXT:    .cfi_offset lr, 16
-; CHECK-NEXT:    clrldi r4, r4, 32
+; CHECK-NEXT:    extsw r4, r4
 ; CHECK-NEXT:    bl ldexp
 ; CHECK-NEXT:    nop
 ; CHECK-NEXT:    addi r1, r1, 32

diff --git a/llvm/test/CodeGen/PowerPC/ldexp.ll b/llvm/test/CodeGen/PowerPC/ldexp.ll
index 151df6096b30bdd3bcff645f709794d071317896..ffc826cc86de59dad1bea1a86011b9a2504f2693 100644 (file)
--- a/llvm/test/CodeGen/PowerPC/ldexp.ll
+++ b/llvm/test/CodeGen/PowerPC/ldexp.ll
@@ -57,22 +57,24 @@ define <2 x float> @ldexp_v2f32(<2 x float> %val, <2 x i32> %exp) {
 ; CHECK-NEXT:    .cfi_offset v29, -48
 ; CHECK-NEXT:    .cfi_offset v30, -32
 ; CHECK-NEXT:    .cfi_offset v31, -16
-; CHECK-NEXT:    xxsldwi vs0, v2, v2, 3
 ; CHECK-NEXT:    li r3, 0
+; CHECK-NEXT:    xxsldwi vs0, v2, v2, 3
 ; CHECK-NEXT:    stxv v29, 32(r1) # 16-byte Folded Spill
 ; CHECK-NEXT:    xscvspdpn f1, vs0
-; CHECK-NEXT:    vextuwrx r4, r3, v3
+; CHECK-NEXT:    vextuwrx r3, r3, v3
 ; CHECK-NEXT:    stxv v30, 48(r1) # 16-byte Folded Spill
 ; CHECK-NEXT:    stxv v31, 64(r1) # 16-byte Folded Spill
+; CHECK-NEXT:    extsw r4, r3
 ; CHECK-NEXT:    vmr v31, v3
 ; CHECK-NEXT:    vmr v30, v2
 ; CHECK-NEXT:    bl ldexpf
 ; CHECK-NEXT:    nop
-; CHECK-NEXT:    xxswapd vs0, v30
 ; CHECK-NEXT:    li r3, 4
+; CHECK-NEXT:    xxswapd vs0, v30
 ; CHECK-NEXT:    xscvdpspn v29, f1
 ; CHECK-NEXT:    xscvspdpn f1, vs0
-; CHECK-NEXT:    vextuwrx r4, r3, v31
+; CHECK-NEXT:    vextuwrx r3, r3, v31
+; CHECK-NEXT:    extsw r4, r3
 ; CHECK-NEXT:    bl ldexpf
 ; CHECK-NEXT:    nop
 ; CHECK-NEXT:    xscvdpspn vs0, f1
@@ -100,35 +102,39 @@ define <4 x float> @ldexp_v4f32(<4 x float> %val, <4 x i32> %exp) {
 ; CHECK-NEXT:    .cfi_offset v29, -48
 ; CHECK-NEXT:    .cfi_offset v30, -32
 ; CHECK-NEXT:    .cfi_offset v31, -16
-; CHECK-NEXT:    li r3, 12
-; CHECK-NEXT:    xscvspdpn f1, v2
+; CHECK-NEXT:    li r3, 4
+; CHECK-NEXT:    xxswapd vs0, v2
 ; CHECK-NEXT:    stxv v28, 32(r1) # 16-byte Folded Spill
+; CHECK-NEXT:    xscvspdpn f1, vs0
+; CHECK-NEXT:    vextuwrx r3, r3, v3
 ; CHECK-NEXT:    stxv v29, 48(r1) # 16-byte Folded Spill
 ; CHECK-NEXT:    stxv v30, 64(r1) # 16-byte Folded Spill
 ; CHECK-NEXT:    stxv v31, 80(r1) # 16-byte Folded Spill
 ; CHECK-NEXT:    vmr v31, v3
+; CHECK-NEXT:    extsw r4, r3
 ; CHECK-NEXT:    vmr v30, v2
-; CHECK-NEXT:    vextuwrx r4, r3, v3
 ; CHECK-NEXT:    bl ldexpf
 ; CHECK-NEXT:    nop
-; CHECK-NEXT:    xxswapd vs0, v30
-; CHECK-NEXT:    li r3, 4
+; CHECK-NEXT:    li r3, 12
 ; CHECK-NEXT:    xscpsgndp v29, f1, f1
-; CHECK-NEXT:    xscvspdpn f1, vs0
-; CHECK-NEXT:    vextuwrx r4, r3, v31
+; CHECK-NEXT:    xscvspdpn f1, v30
+; CHECK-NEXT:    vextuwrx r3, r3, v31
+; CHECK-NEXT:    extsw r4, r3
 ; CHECK-NEXT:    bl ldexpf
 ; CHECK-NEXT:    nop
-; CHECK-NEXT:    xxmrghd vs0, v29, vs1
+; CHECK-NEXT:    xxmrghd vs0, vs1, v29
 ; CHECK-NEXT:    li r3, 0
-; CHECK-NEXT:    vextuwrx r4, r3, v31
+; CHECK-NEXT:    vextuwrx r3, r3, v31
 ; CHECK-NEXT:    xvcvdpsp v28, vs0
 ; CHECK-NEXT:    xxsldwi vs0, v30, v30, 3
+; CHECK-NEXT:    extsw r4, r3
 ; CHECK-NEXT:    xscvspdpn f1, vs0
 ; CHECK-NEXT:    bl ldexpf
 ; CHECK-NEXT:    nop
 ; CHECK-NEXT:    xxsldwi vs0, v30, v30, 1
+; CHECK-NEXT:    mfvsrwz r3, v31
 ; CHECK-NEXT:    xscpsgndp v29, f1, f1
-; CHECK-NEXT:    mfvsrwz r4, v31
+; CHECK-NEXT:    extsw r4, r3
 ; CHECK-NEXT:    xscvspdpn f1, vs0
 ; CHECK-NEXT:    bl ldexpf
 ; CHECK-NEXT:    nop
@@ -156,7 +162,7 @@ define half @ldexp_f16(half %arg0, i32 %arg1) {
 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-NEXT:    .cfi_offset lr, 16
 ; CHECK-NEXT:    xscvdphp f0, f1
-; CHECK-NEXT:    clrldi r4, r4, 32
+; CHECK-NEXT:    extsw r4, r4
 ; CHECK-NEXT:    mffprwz r3, f0
 ; CHECK-NEXT:    clrlwi r3, r3, 16
 ; CHECK-NEXT:    mtfprwz f0, r3

diff --git a/llvm/test/CodeGen/PowerPC/negative-integer-fp-libcall.ll b/llvm/test/CodeGen/PowerPC/negative-integer-fp-libcall.ll
new file mode 100644 (file)
index 0000000..010ee6e
--- /dev/null
+++ b/llvm/test/CodeGen/PowerPC/negative-integer-fp-libcall.ll
@@ -0,0 +1,26 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -O1 -mtriple=powerpc64-unknown-linux-gnu -mcpu=pwr8 < %s | FileCheck %s
+
+; Test that a negative parameter smaller than 64 bits (e.g., int)
+; is correctly implemented with sign-extension when passed to
+; a floating point libcall.
+
+define double @ldexp_test(ptr %a, ptr %b) nounwind {
+; CHECK-LABEL: ldexp_test:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    mflr 0
+; CHECK-NEXT:    stdu 1, -112(1)
+; CHECK-NEXT:    std 0, 128(1)
+; CHECK-NEXT:    lfd 1, 0(3)
+; CHECK-NEXT:    lwa 4, 0(4)
+; CHECK-NEXT:    bl ldexp
+; CHECK-NEXT:    nop
+; CHECK-NEXT:    addi 1, 1, 112
+; CHECK-NEXT:    ld 0, 16(1)
+; CHECK-NEXT:    mtlr 0
+; CHECK-NEXT:    blr
+  %base = load double, ptr %a
+  %exp = load i32, ptr %b
+  %call = call double @llvm.ldexp.f64.i32(double %base, i32 signext %exp)
+  ret double %call
+}

diff --git a/llvm/test/CodeGen/X86/fold-int-pow2-with-fmul-or-fdiv.ll b/llvm/test/CodeGen/X86/fold-int-pow2-with-fmul-or-fdiv.ll
index 5a051a9c499e4758c1c9553a4cf6f4389cef3db8..332fbf7188af81c8b34c64b26022eda034680a48 100644 (file)
--- a/llvm/test/CodeGen/X86/fold-int-pow2-with-fmul-or-fdiv.ll
+++ b/llvm/test/CodeGen/X86/fold-int-pow2-with-fmul-or-fdiv.ll
@@ -406,13 +406,15 @@ define <8 x half> @fmul_pow2_ldexp_8xhalf(<8 x i16> %i) {
 ; CHECK-SSE-NEXT:    subq $72, %rsp
 ; CHECK-SSE-NEXT:    .cfi_def_cfa_offset 80
 ; CHECK-SSE-NEXT:    movdqa %xmm0, (%rsp) # 16-byte Spill
-; CHECK-SSE-NEXT:    pextrw $7, %xmm0, %edi
+; CHECK-SSE-NEXT:    pextrw $7, %xmm0, %eax
+; CHECK-SSE-NEXT:    movswl %ax, %edi
 ; CHECK-SSE-NEXT:    movss {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-SSE-NEXT:    callq ldexpf@PLT
 ; CHECK-SSE-NEXT:    callq __truncsfhf2@PLT
 ; CHECK-SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; CHECK-SSE-NEXT:    pextrw $6, %xmm0, %edi
+; CHECK-SSE-NEXT:    pextrw $6, %xmm0, %eax
+; CHECK-SSE-NEXT:    movswl %ax, %edi
 ; CHECK-SSE-NEXT:    movd {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-SSE-NEXT:    callq ldexpf@PLT
 ; CHECK-SSE-NEXT:    callq __truncsfhf2@PLT
@@ -420,13 +422,15 @@ define <8 x half> @fmul_pow2_ldexp_8xhalf(<8 x i16> %i) {
 ; CHECK-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
 ; CHECK-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; CHECK-SSE-NEXT:    pextrw $5, %xmm0, %edi
+; CHECK-SSE-NEXT:    pextrw $5, %xmm0, %eax
+; CHECK-SSE-NEXT:    movswl %ax, %edi
 ; CHECK-SSE-NEXT:    movss {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-SSE-NEXT:    callq ldexpf@PLT
 ; CHECK-SSE-NEXT:    callq __truncsfhf2@PLT
 ; CHECK-SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; CHECK-SSE-NEXT:    pextrw $4, %xmm0, %edi
+; CHECK-SSE-NEXT:    pextrw $4, %xmm0, %eax
+; CHECK-SSE-NEXT:    movswl %ax, %edi
 ; CHECK-SSE-NEXT:    movd {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-SSE-NEXT:    callq ldexpf@PLT
 ; CHECK-SSE-NEXT:    callq __truncsfhf2@PLT
@@ -436,13 +440,15 @@ define <8 x half> @fmul_pow2_ldexp_8xhalf(<8 x i16> %i) {
 ; CHECK-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
 ; CHECK-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; CHECK-SSE-NEXT:    pextrw $3, %xmm0, %edi
+; CHECK-SSE-NEXT:    pextrw $3, %xmm0, %eax
+; CHECK-SSE-NEXT:    movswl %ax, %edi
 ; CHECK-SSE-NEXT:    movss {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-SSE-NEXT:    callq ldexpf@PLT
 ; CHECK-SSE-NEXT:    callq __truncsfhf2@PLT
 ; CHECK-SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; CHECK-SSE-NEXT:    pextrw $2, %xmm0, %edi
+; CHECK-SSE-NEXT:    pextrw $2, %xmm0, %eax
+; CHECK-SSE-NEXT:    movswl %ax, %edi
 ; CHECK-SSE-NEXT:    movd {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-SSE-NEXT:    callq ldexpf@PLT
 ; CHECK-SSE-NEXT:    callq __truncsfhf2@PLT
@@ -450,14 +456,15 @@ define <8 x half> @fmul_pow2_ldexp_8xhalf(<8 x i16> %i) {
 ; CHECK-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
 ; CHECK-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; CHECK-SSE-NEXT:    pextrw $1, %xmm0, %edi
+; CHECK-SSE-NEXT:    pextrw $1, %xmm0, %eax
+; CHECK-SSE-NEXT:    movswl %ax, %edi
 ; CHECK-SSE-NEXT:    movss {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-SSE-NEXT:    callq ldexpf@PLT
 ; CHECK-SSE-NEXT:    callq __truncsfhf2@PLT
 ; CHECK-SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-SSE-NEXT:    movd %xmm0, %eax
-; CHECK-SSE-NEXT:    movzwl %ax, %edi
+; CHECK-SSE-NEXT:    movswl %ax, %edi
 ; CHECK-SSE-NEXT:    movd {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-SSE-NEXT:    callq ldexpf@PLT
 ; CHECK-SSE-NEXT:    callq __truncsfhf2@PLT
@@ -476,13 +483,15 @@ define <8 x half> @fmul_pow2_ldexp_8xhalf(<8 x i16> %i) {
 ; CHECK-AVX2-NEXT:    subq $72, %rsp
 ; CHECK-AVX2-NEXT:    .cfi_def_cfa_offset 80
 ; CHECK-AVX2-NEXT:    vmovdqa %xmm0, (%rsp) # 16-byte Spill
-; CHECK-AVX2-NEXT:    vpextrw $7, %xmm0, %edi
+; CHECK-AVX2-NEXT:    vpextrw $7, %xmm0, %eax
+; CHECK-AVX2-NEXT:    movswl %ax, %edi
 ; CHECK-AVX2-NEXT:    vmovss {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-AVX2-NEXT:    callq ldexpf@PLT
 ; CHECK-AVX2-NEXT:    callq __truncsfhf2@PLT
 ; CHECK-AVX2-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-AVX2-NEXT:    vmovdqa (%rsp), %xmm0 # 16-byte Reload
-; CHECK-AVX2-NEXT:    vpextrw $6, %xmm0, %edi
+; CHECK-AVX2-NEXT:    vpextrw $6, %xmm0, %eax
+; CHECK-AVX2-NEXT:    movswl %ax, %edi
 ; CHECK-AVX2-NEXT:    vmovd {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-AVX2-NEXT:    callq ldexpf@PLT
 ; CHECK-AVX2-NEXT:    callq __truncsfhf2@PLT
@@ -490,13 +499,15 @@ define <8 x half> @fmul_pow2_ldexp_8xhalf(<8 x i16> %i) {
 ; CHECK-AVX2-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
 ; CHECK-AVX2-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-AVX2-NEXT:    vmovdqa (%rsp), %xmm0 # 16-byte Reload
-; CHECK-AVX2-NEXT:    vpextrw $5, %xmm0, %edi
+; CHECK-AVX2-NEXT:    vpextrw $5, %xmm0, %eax
+; CHECK-AVX2-NEXT:    movswl %ax, %edi
 ; CHECK-AVX2-NEXT:    vmovss {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-AVX2-NEXT:    callq ldexpf@PLT
 ; CHECK-AVX2-NEXT:    callq __truncsfhf2@PLT
 ; CHECK-AVX2-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-AVX2-NEXT:    vmovdqa (%rsp), %xmm0 # 16-byte Reload
-; CHECK-AVX2-NEXT:    vpextrw $4, %xmm0, %edi
+; CHECK-AVX2-NEXT:    vpextrw $4, %xmm0, %eax
+; CHECK-AVX2-NEXT:    movswl %ax, %edi
 ; CHECK-AVX2-NEXT:    vmovd {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-AVX2-NEXT:    callq ldexpf@PLT
 ; CHECK-AVX2-NEXT:    callq __truncsfhf2@PLT
@@ -506,13 +517,15 @@ define <8 x half> @fmul_pow2_ldexp_8xhalf(<8 x i16> %i) {
 ; CHECK-AVX2-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
 ; CHECK-AVX2-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-AVX2-NEXT:    vmovdqa (%rsp), %xmm0 # 16-byte Reload
-; CHECK-AVX2-NEXT:    vpextrw $3, %xmm0, %edi
+; CHECK-AVX2-NEXT:    vpextrw $3, %xmm0, %eax
+; CHECK-AVX2-NEXT:    movswl %ax, %edi
 ; CHECK-AVX2-NEXT:    vmovss {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-AVX2-NEXT:    callq ldexpf@PLT
 ; CHECK-AVX2-NEXT:    callq __truncsfhf2@PLT
 ; CHECK-AVX2-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-AVX2-NEXT:    vmovdqa (%rsp), %xmm0 # 16-byte Reload
-; CHECK-AVX2-NEXT:    vpextrw $2, %xmm0, %edi
+; CHECK-AVX2-NEXT:    vpextrw $2, %xmm0, %eax
+; CHECK-AVX2-NEXT:    movswl %ax, %edi
 ; CHECK-AVX2-NEXT:    vmovd {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-AVX2-NEXT:    callq ldexpf@PLT
 ; CHECK-AVX2-NEXT:    callq __truncsfhf2@PLT
@@ -520,14 +533,15 @@ define <8 x half> @fmul_pow2_ldexp_8xhalf(<8 x i16> %i) {
 ; CHECK-AVX2-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
 ; CHECK-AVX2-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-AVX2-NEXT:    vmovdqa (%rsp), %xmm0 # 16-byte Reload
-; CHECK-AVX2-NEXT:    vpextrw $1, %xmm0, %edi
+; CHECK-AVX2-NEXT:    vpextrw $1, %xmm0, %eax
+; CHECK-AVX2-NEXT:    movswl %ax, %edi
 ; CHECK-AVX2-NEXT:    vmovss {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-AVX2-NEXT:    callq ldexpf@PLT
 ; CHECK-AVX2-NEXT:    callq __truncsfhf2@PLT
 ; CHECK-AVX2-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-AVX2-NEXT:    vmovdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-AVX2-NEXT:    vmovd %xmm0, %eax
-; CHECK-AVX2-NEXT:    movzwl %ax, %edi
+; CHECK-AVX2-NEXT:    movswl %ax, %edi
 ; CHECK-AVX2-NEXT:    vmovd {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-AVX2-NEXT:    callq ldexpf@PLT
 ; CHECK-AVX2-NEXT:    callq __truncsfhf2@PLT
@@ -546,7 +560,8 @@ define <8 x half> @fmul_pow2_ldexp_8xhalf(<8 x i16> %i) {
 ; CHECK-AVX512F-NEXT:    subq $72, %rsp
 ; CHECK-AVX512F-NEXT:    .cfi_def_cfa_offset 80
 ; CHECK-AVX512F-NEXT:    vmovdqa %xmm0, (%rsp) # 16-byte Spill
-; CHECK-AVX512F-NEXT:    vpextrw $7, %xmm0, %edi
+; CHECK-AVX512F-NEXT:    vpextrw $7, %xmm0, %eax
+; CHECK-AVX512F-NEXT:    movswl %ax, %edi
 ; CHECK-AVX512F-NEXT:    vmovss {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-AVX512F-NEXT:    callq ldexpf@PLT
 ; CHECK-AVX512F-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
@@ -554,7 +569,8 @@ define <8 x half> @fmul_pow2_ldexp_8xhalf(<8 x i16> %i) {
 ; CHECK-AVX512F-NEXT:    vpinsrw $0, %eax, %xmm0, %xmm0
 ; CHECK-AVX512F-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-AVX512F-NEXT:    vmovdqa (%rsp), %xmm0 # 16-byte Reload
-; CHECK-AVX512F-NEXT:    vpextrw $6, %xmm0, %edi
+; CHECK-AVX512F-NEXT:    vpextrw $6, %xmm0, %eax
+; CHECK-AVX512F-NEXT:    movswl %ax, %edi
 ; CHECK-AVX512F-NEXT:    vmovss {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-AVX512F-NEXT:    callq ldexpf@PLT
 ; CHECK-AVX512F-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
@@ -564,7 +580,8 @@ define <8 x half> @fmul_pow2_ldexp_8xhalf(<8 x i16> %i) {
 ; CHECK-AVX512F-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
 ; CHECK-AVX512F-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-AVX512F-NEXT:    vmovdqa (%rsp), %xmm0 # 16-byte Reload
-; CHECK-AVX512F-NEXT:    vpextrw $5, %xmm0, %edi
+; CHECK-AVX512F-NEXT:    vpextrw $5, %xmm0, %eax
+; CHECK-AVX512F-NEXT:    movswl %ax, %edi
 ; CHECK-AVX512F-NEXT:    vmovss {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-AVX512F-NEXT:    callq ldexpf@PLT
 ; CHECK-AVX512F-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
@@ -572,7 +589,8 @@ define <8 x half> @fmul_pow2_ldexp_8xhalf(<8 x i16> %i) {
 ; CHECK-AVX512F-NEXT:    vpinsrw $0, %eax, %xmm0, %xmm0
 ; CHECK-AVX512F-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-AVX512F-NEXT:    vmovdqa (%rsp), %xmm0 # 16-byte Reload
-; CHECK-AVX512F-NEXT:    vpextrw $4, %xmm0, %edi
+; CHECK-AVX512F-NEXT:    vpextrw $4, %xmm0, %eax
+; CHECK-AVX512F-NEXT:    movswl %ax, %edi
 ; CHECK-AVX512F-NEXT:    vmovss {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-AVX512F-NEXT:    callq ldexpf@PLT
 ; CHECK-AVX512F-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
@@ -584,7 +602,8 @@ define <8 x half> @fmul_pow2_ldexp_8xhalf(<8 x i16> %i) {
 ; CHECK-AVX512F-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
 ; CHECK-AVX512F-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-AVX512F-NEXT:    vmovdqa (%rsp), %xmm0 # 16-byte Reload
-; CHECK-AVX512F-NEXT:    vpextrw $3, %xmm0, %edi
+; CHECK-AVX512F-NEXT:    vpextrw $3, %xmm0, %eax
+; CHECK-AVX512F-NEXT:    movswl %ax, %edi
 ; CHECK-AVX512F-NEXT:    vmovss {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-AVX512F-NEXT:    callq ldexpf@PLT
 ; CHECK-AVX512F-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
@@ -592,7 +611,8 @@ define <8 x half> @fmul_pow2_ldexp_8xhalf(<8 x i16> %i) {
 ; CHECK-AVX512F-NEXT:    vpinsrw $0, %eax, %xmm0, %xmm0
 ; CHECK-AVX512F-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-AVX512F-NEXT:    vmovdqa (%rsp), %xmm0 # 16-byte Reload
-; CHECK-AVX512F-NEXT:    vpextrw $2, %xmm0, %edi
+; CHECK-AVX512F-NEXT:    vpextrw $2, %xmm0, %eax
+; CHECK-AVX512F-NEXT:    movswl %ax, %edi
 ; CHECK-AVX512F-NEXT:    vmovss {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-AVX512F-NEXT:    callq ldexpf@PLT
 ; CHECK-AVX512F-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
@@ -602,7 +622,8 @@ define <8 x half> @fmul_pow2_ldexp_8xhalf(<8 x i16> %i) {
 ; CHECK-AVX512F-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
 ; CHECK-AVX512F-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-AVX512F-NEXT:    vmovdqa (%rsp), %xmm0 # 16-byte Reload
-; CHECK-AVX512F-NEXT:    vpextrw $1, %xmm0, %edi
+; CHECK-AVX512F-NEXT:    vpextrw $1, %xmm0, %eax
+; CHECK-AVX512F-NEXT:    movswl %ax, %edi
 ; CHECK-AVX512F-NEXT:    vmovss {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-AVX512F-NEXT:    callq ldexpf@PLT
 ; CHECK-AVX512F-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
@@ -611,7 +632,7 @@ define <8 x half> @fmul_pow2_ldexp_8xhalf(<8 x i16> %i) {
 ; CHECK-AVX512F-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-AVX512F-NEXT:    vmovdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-AVX512F-NEXT:    vmovd %xmm0, %eax
-; CHECK-AVX512F-NEXT:    movzwl %ax, %edi
+; CHECK-AVX512F-NEXT:    movswl %ax, %edi
 ; CHECK-AVX512F-NEXT:    vmovss {{.*#+}} xmm0 = [8.192E+3,0.0E+0,0.0E+0,0.0E+0]
 ; CHECK-AVX512F-NEXT:    callq ldexpf@PLT
 ; CHECK-AVX512F-NEXT:    vcvtps2ph $4, %xmm0, %xmm0