[llvm] 73cddef - optimize `is_finite` assembly (#169402)
via llvm-commits
llvm-commits at lists.llvm.org
Sun Mar 29 07:28:12 PDT 2026
Author: Folkert de Vries
Date: 2026-03-29T14:28:07Z
New Revision: 73cddef78828d371d8d1c1b05adec9a6f37f462d
URL: https://github.com/llvm/llvm-project/commit/73cddef78828d371d8d1c1b05adec9a6f37f462d
DIFF: https://github.com/llvm/llvm-project/commit/73cddef78828d371d8d1c1b05adec9a6f37f462d.diff
LOG: optimize `is_finite` assembly (#169402)
Fixes https://github.com/llvm/llvm-project/issues/169270
Changes the implementation of `is_finite` to emit fewer instructions,
e.g.
X86_64
```asm
old: # 18 bytes
movd %xmm0, %eax
andl $2147483647, %eax
cmpl $2139095040, %eax
setl %al
retq
new: # 15 bytes
movd %xmm0, %eax
addl %eax, %eax
cmpl $-16777216, %eax
setb %al
retq
```
Aarch64
```asm
old:
fmov w9, s0
mov w8, #2139095040
and w9, w9, #0x7fffffff
cmp w9, w8
cset w0, lt
ret
new:
fmov w8, s0
ubfx w8, w8, #23, #8
cmp w8, #255
cset w0, lo
ret
```
See the issue for more information.
Added:
Modified:
llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
llvm/test/CodeGen/AArch64/is_fpclass-bfloat.ll
llvm/test/CodeGen/AArch64/is_fpclass.ll
llvm/test/CodeGen/AMDGPU/llvm.is.fpclass.bf16.ll
llvm/test/CodeGen/AMDGPU/llvm.is.fpclass.f16.ll
llvm/test/CodeGen/AMDGPU/r600.llvm.is.fpclass.ll
llvm/test/CodeGen/PowerPC/fp-classify.ll
llvm/test/CodeGen/RISCV/float-intrinsics.ll
llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vfclass.ll
llvm/test/CodeGen/RISCV/rvv/vfclass-sdnode.ll
llvm/test/CodeGen/X86/is_fpclass.ll
llvm/test/CodeGen/X86/isel-fpclass.ll
Removed:
################################################################################
diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index f739ae7bfb7b2..0b18a3a4d7493 100644
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -9575,8 +9575,19 @@ SDValue TargetLowering::expandIS_FPCLASS(EVT ResultVT, SDValue Op,
; // Detect finite numbers of f80 by checking individual classes because
// they have
diff erent settings of the explicit integer bit.
else if ((Test & fcFinite) == fcFinite) {
- // finite(V) ==> abs(V) < exp_mask
- PartialRes = DAG.getSetCC(DL, ResultVT, AbsV, ExpMaskV, ISD::SETLT);
+ // finite(V) ==> (a << 1) < (inf << 1)
+ //
+ // See https://github.com/llvm/llvm-project/issues/169270, this is slightly
+ // shorter than the `finite(V) ==> abs(V) < exp_mask` formula used before.
+
+ assert(APFloat::isIEEELikeFP(OperandVT.getFltSemantics()) &&
+ "finite check requires IEEE-like FP");
+
+ SDValue One = DAG.getShiftAmountConstant(1, IntVT, DL);
+ SDValue TwiceOp = DAG.getNode(ISD::SHL, DL, IntVT, OpAsInt, One);
+ SDValue TwiceInf = DAG.getNode(ISD::SHL, DL, IntVT, ExpMaskV, One);
+
+ PartialRes = DAG.getSetCC(DL, ResultVT, TwiceOp, TwiceInf, ISD::SETULT);
Test &= ~fcFinite;
} else if ((Test & fcFinite) == fcPosFinite) {
// finite(V) && V > 0 ==> V < exp_mask
diff --git a/llvm/test/CodeGen/AArch64/is_fpclass-bfloat.ll b/llvm/test/CodeGen/AArch64/is_fpclass-bfloat.ll
index 2f38481f9572e..bfdec5d0f0769 100644
--- a/llvm/test/CodeGen/AArch64/is_fpclass-bfloat.ll
+++ b/llvm/test/CodeGen/AArch64/is_fpclass-bfloat.ll
@@ -6,19 +6,17 @@ define i1 @isfinite_bf16(bfloat %x) {
; CHECK-LABEL: isfinite_bf16:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: // kill: def $h0 killed $h0 def $s0
-; CHECK-NEXT: fmov w9, s0
-; CHECK-NEXT: mov w8, #32640 // =0x7f80
-; CHECK-NEXT: and w9, w9, #0x7fff
-; CHECK-NEXT: cmp w9, w8
-; CHECK-NEXT: cset w0, lt
+; CHECK-NEXT: fmov w8, s0
+; CHECK-NEXT: ubfx w8, w8, #7, #8
+; CHECK-NEXT: cmp w8, #255
+; CHECK-NEXT: cset w0, lo
; CHECK-NEXT: ret
;
; CHECK-NOFP-LABEL: isfinite_bf16:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov w8, #32640 // =0x7f80
-; CHECK-NOFP-NEXT: and w9, w0, #0x7fff
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: cset w0, lt
+; CHECK-NOFP-NEXT: ubfx w8, w0, #7, #8
+; CHECK-NOFP-NEXT: cmp w8, #255
+; CHECK-NOFP-NEXT: cset w0, lo
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call i1 @llvm.is.fpclass.bf16(bfloat %x, i32 504) ; 0x1f8 = "finite"
@@ -29,19 +27,17 @@ define i1 @not_isfinite_bf16(bfloat %x) {
; CHECK-LABEL: not_isfinite_bf16:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: // kill: def $h0 killed $h0 def $s0
-; CHECK-NEXT: fmov w9, s0
-; CHECK-NEXT: mov w8, #32639 // =0x7f7f
-; CHECK-NEXT: and w9, w9, #0x7fff
-; CHECK-NEXT: cmp w9, w8
-; CHECK-NEXT: cset w0, gt
+; CHECK-NEXT: fmov w8, s0
+; CHECK-NEXT: ubfx w8, w8, #7, #8
+; CHECK-NEXT: cmp w8, #254
+; CHECK-NEXT: cset w0, hi
; CHECK-NEXT: ret
;
; CHECK-NOFP-LABEL: not_isfinite_bf16:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov w8, #32639 // =0x7f7f
-; CHECK-NOFP-NEXT: and w9, w0, #0x7fff
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: cset w0, gt
+; CHECK-NOFP-NEXT: ubfx w8, w0, #7, #8
+; CHECK-NOFP-NEXT: cmp w8, #254
+; CHECK-NOFP-NEXT: cset w0, hi
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call i1 @llvm.is.fpclass.bf16(bfloat %x, i32 519) ; ~0x1f8 = "~finite"
@@ -52,27 +48,25 @@ entry:
define <4 x i1> @isfinite_v4bf16(<4 x bfloat> %x) {
; CHECK-LABEL: isfinite_v4bf16:
; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: mov w8, #32640 // =0x7f80
-; CHECK-NEXT: bic v0.4h, #128, lsl #8
-; CHECK-NEXT: dup v1.4h, w8
-; CHECK-NEXT: cmgt v0.4h, v1.4h, v0.4h
+; CHECK-NEXT: movi d1, #0xff00ff00ff00ff00
+; CHECK-NEXT: add v0.4h, v0.4h, v0.4h
+; CHECK-NEXT: cmhi v0.4h, v1.4h, v0.4h
; CHECK-NEXT: ret
;
; CHECK-NOFP-LABEL: isfinite_v4bf16:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov w8, #32640 // =0x7f80
-; CHECK-NOFP-NEXT: and w9, w0, #0x7fff
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: and w9, w1, #0x7fff
-; CHECK-NOFP-NEXT: cset w0, lt
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: and w9, w2, #0x7fff
-; CHECK-NOFP-NEXT: cset w1, lt
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: and w9, w3, #0x7fff
-; CHECK-NOFP-NEXT: cset w2, lt
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: cset w3, lt
+; CHECK-NOFP-NEXT: ubfx w8, w0, #7, #8
+; CHECK-NOFP-NEXT: ubfx w9, w1, #7, #8
+; CHECK-NOFP-NEXT: cmp w8, #255
+; CHECK-NOFP-NEXT: ubfx w8, w2, #7, #8
+; CHECK-NOFP-NEXT: cset w0, lo
+; CHECK-NOFP-NEXT: cmp w9, #255
+; CHECK-NOFP-NEXT: ubfx w9, w3, #7, #8
+; CHECK-NOFP-NEXT: cset w1, lo
+; CHECK-NOFP-NEXT: cmp w8, #255
+; CHECK-NOFP-NEXT: cset w2, lo
+; CHECK-NOFP-NEXT: cmp w9, #255
+; CHECK-NOFP-NEXT: cset w3, lo
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call <4 x i1> @llvm.is.fpclass.v4bf16(<4 x bfloat> %x, i32 504) ; 0x1f8 = "finite"
@@ -82,27 +76,25 @@ entry:
define <4 x i1> @not_isfinite_v4bf16(<4 x bfloat> %x) {
; CHECK-LABEL: not_isfinite_v4bf16:
; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: mov w8, #32640 // =0x7f80
-; CHECK-NEXT: bic v0.4h, #128, lsl #8
-; CHECK-NEXT: dup v1.4h, w8
-; CHECK-NEXT: cmge v0.4h, v0.4h, v1.4h
+; CHECK-NEXT: movi d1, #0xff00ff00ff00ff00
+; CHECK-NEXT: add v0.4h, v0.4h, v0.4h
+; CHECK-NEXT: cmhs v0.4h, v0.4h, v1.4h
; CHECK-NEXT: ret
;
; CHECK-NOFP-LABEL: not_isfinite_v4bf16:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov w8, #32639 // =0x7f7f
-; CHECK-NOFP-NEXT: and w9, w0, #0x7fff
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: and w9, w1, #0x7fff
-; CHECK-NOFP-NEXT: cset w0, gt
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: and w9, w2, #0x7fff
-; CHECK-NOFP-NEXT: cset w1, gt
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: and w9, w3, #0x7fff
-; CHECK-NOFP-NEXT: cset w2, gt
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: cset w3, gt
+; CHECK-NOFP-NEXT: ubfx w8, w0, #7, #8
+; CHECK-NOFP-NEXT: ubfx w9, w1, #7, #8
+; CHECK-NOFP-NEXT: cmp w8, #254
+; CHECK-NOFP-NEXT: ubfx w8, w2, #7, #8
+; CHECK-NOFP-NEXT: cset w0, hi
+; CHECK-NOFP-NEXT: cmp w9, #254
+; CHECK-NOFP-NEXT: ubfx w9, w3, #7, #8
+; CHECK-NOFP-NEXT: cset w1, hi
+; CHECK-NOFP-NEXT: cmp w8, #254
+; CHECK-NOFP-NEXT: cset w2, hi
+; CHECK-NOFP-NEXT: cmp w9, #254
+; CHECK-NOFP-NEXT: cset w3, hi
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call <4 x i1> @llvm.is.fpclass.v4bf16(<4 x bfloat> %x, i32 519) ; ~0x1f8 = "~finite"
diff --git a/llvm/test/CodeGen/AArch64/is_fpclass.ll b/llvm/test/CodeGen/AArch64/is_fpclass.ll
index 27fd1243d83c7..71a8f42a7b4db 100644
--- a/llvm/test/CodeGen/AArch64/is_fpclass.ll
+++ b/llvm/test/CodeGen/AArch64/is_fpclass.ll
@@ -7,11 +7,10 @@ define i1 @isfinite_h(half %x) {
; CHECK-SD-LABEL: isfinite_h:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: // kill: def $h0 killed $h0 def $s0
-; CHECK-SD-NEXT: fmov w9, s0
-; CHECK-SD-NEXT: mov w8, #31744 // =0x7c00
-; CHECK-SD-NEXT: and w9, w9, #0x7fff
-; CHECK-SD-NEXT: cmp w9, w8
-; CHECK-SD-NEXT: cset w0, lt
+; CHECK-SD-NEXT: fmov w8, s0
+; CHECK-SD-NEXT: ubfx w8, w8, #10, #5
+; CHECK-SD-NEXT: cmp w8, #31
+; CHECK-SD-NEXT: cset w0, lo
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: isfinite_h:
@@ -26,10 +25,9 @@ define i1 @isfinite_h(half %x) {
;
; CHECK-NOFP-LABEL: isfinite_h:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov w8, #31744 // =0x7c00
-; CHECK-NOFP-NEXT: and w9, w0, #0x7fff
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: cset w0, lt
+; CHECK-NOFP-NEXT: ubfx w8, w0, #10, #5
+; CHECK-NOFP-NEXT: cmp w8, #31
+; CHECK-NOFP-NEXT: cset w0, lo
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call i1 @llvm.is.fpclass.f16(half %x, i32 504) ; 0x1f8 = "finite"
@@ -40,11 +38,10 @@ define i1 @not_isfinite_h(half %x) {
; CHECK-SD-LABEL: not_isfinite_h:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: // kill: def $h0 killed $h0 def $s0
-; CHECK-SD-NEXT: fmov w9, s0
-; CHECK-SD-NEXT: mov w8, #31743 // =0x7bff
-; CHECK-SD-NEXT: and w9, w9, #0x7fff
-; CHECK-SD-NEXT: cmp w9, w8
-; CHECK-SD-NEXT: cset w0, gt
+; CHECK-SD-NEXT: fmov w8, s0
+; CHECK-SD-NEXT: ubfx w8, w8, #10, #5
+; CHECK-SD-NEXT: cmp w8, #30
+; CHECK-SD-NEXT: cset w0, hi
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: not_isfinite_h:
@@ -61,10 +58,9 @@ define i1 @not_isfinite_h(half %x) {
;
; CHECK-NOFP-LABEL: not_isfinite_h:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov w8, #31743 // =0x7bff
-; CHECK-NOFP-NEXT: and w9, w0, #0x7fff
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: cset w0, gt
+; CHECK-NOFP-NEXT: ubfx w8, w0, #10, #5
+; CHECK-NOFP-NEXT: cmp w8, #30
+; CHECK-NOFP-NEXT: cset w0, hi
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call i1 @llvm.is.fpclass.f16(half %x, i32 519) ; ~0x1f8 = "~finite"
@@ -74,11 +70,10 @@ entry:
define i1 @isfinite_f(float %x) {
; CHECK-SD-LABEL: isfinite_f:
; CHECK-SD: // %bb.0: // %entry
-; CHECK-SD-NEXT: fmov w9, s0
-; CHECK-SD-NEXT: mov w8, #2139095040 // =0x7f800000
-; CHECK-SD-NEXT: and w9, w9, #0x7fffffff
-; CHECK-SD-NEXT: cmp w9, w8
-; CHECK-SD-NEXT: cset w0, lt
+; CHECK-SD-NEXT: fmov w8, s0
+; CHECK-SD-NEXT: ubfx w8, w8, #23, #8
+; CHECK-SD-NEXT: cmp w8, #255
+; CHECK-SD-NEXT: cset w0, lo
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: isfinite_f:
@@ -92,10 +87,9 @@ define i1 @isfinite_f(float %x) {
;
; CHECK-NOFP-LABEL: isfinite_f:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov w8, #2139095040 // =0x7f800000
-; CHECK-NOFP-NEXT: and w9, w0, #0x7fffffff
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: cset w0, lt
+; CHECK-NOFP-NEXT: ubfx w8, w0, #23, #8
+; CHECK-NOFP-NEXT: cmp w8, #255
+; CHECK-NOFP-NEXT: cset w0, lo
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 504) ; 0x1f8 = "finite"
@@ -105,11 +99,10 @@ entry:
define i1 @not_isfinite_f(float %x) {
; CHECK-SD-LABEL: not_isfinite_f:
; CHECK-SD: // %bb.0: // %entry
-; CHECK-SD-NEXT: fmov w9, s0
-; CHECK-SD-NEXT: mov w8, #2139095039 // =0x7f7fffff
-; CHECK-SD-NEXT: and w9, w9, #0x7fffffff
-; CHECK-SD-NEXT: cmp w9, w8
-; CHECK-SD-NEXT: cset w0, gt
+; CHECK-SD-NEXT: fmov w8, s0
+; CHECK-SD-NEXT: ubfx w8, w8, #23, #8
+; CHECK-SD-NEXT: cmp w8, #254
+; CHECK-SD-NEXT: cset w0, hi
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: not_isfinite_f:
@@ -125,10 +118,9 @@ define i1 @not_isfinite_f(float %x) {
;
; CHECK-NOFP-LABEL: not_isfinite_f:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov w8, #2139095039 // =0x7f7fffff
-; CHECK-NOFP-NEXT: and w9, w0, #0x7fffffff
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: cset w0, gt
+; CHECK-NOFP-NEXT: ubfx w8, w0, #23, #8
+; CHECK-NOFP-NEXT: cmp w8, #254
+; CHECK-NOFP-NEXT: cset w0, hi
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 519) ; ~0x1f8 = "~finite"
@@ -139,11 +131,10 @@ entry:
define i1 @isfinite_f_strictfp(float %x) strictfp {
; CHECK-SD-LABEL: isfinite_f_strictfp:
; CHECK-SD: // %bb.0: // %entry
-; CHECK-SD-NEXT: fmov w9, s0
-; CHECK-SD-NEXT: mov w8, #2139095040 // =0x7f800000
-; CHECK-SD-NEXT: and w9, w9, #0x7fffffff
-; CHECK-SD-NEXT: cmp w9, w8
-; CHECK-SD-NEXT: cset w0, lt
+; CHECK-SD-NEXT: fmov w8, s0
+; CHECK-SD-NEXT: ubfx w8, w8, #23, #8
+; CHECK-SD-NEXT: cmp w8, #255
+; CHECK-SD-NEXT: cset w0, lo
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: isfinite_f_strictfp:
@@ -157,10 +148,9 @@ define i1 @isfinite_f_strictfp(float %x) strictfp {
;
; CHECK-NOFP-LABEL: isfinite_f_strictfp:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov w8, #2139095040 // =0x7f800000
-; CHECK-NOFP-NEXT: and w9, w0, #0x7fffffff
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: cset w0, lt
+; CHECK-NOFP-NEXT: ubfx w8, w0, #23, #8
+; CHECK-NOFP-NEXT: cmp w8, #255
+; CHECK-NOFP-NEXT: cset w0, lo
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 504) strictfp ; 0x1f8 = "finite"
@@ -170,11 +160,10 @@ entry:
define i1 @not_isfinite_f_strictfp(float %x) strictfp {
; CHECK-SD-LABEL: not_isfinite_f_strictfp:
; CHECK-SD: // %bb.0: // %entry
-; CHECK-SD-NEXT: fmov w9, s0
-; CHECK-SD-NEXT: mov w8, #2139095039 // =0x7f7fffff
-; CHECK-SD-NEXT: and w9, w9, #0x7fffffff
-; CHECK-SD-NEXT: cmp w9, w8
-; CHECK-SD-NEXT: cset w0, gt
+; CHECK-SD-NEXT: fmov w8, s0
+; CHECK-SD-NEXT: ubfx w8, w8, #23, #8
+; CHECK-SD-NEXT: cmp w8, #254
+; CHECK-SD-NEXT: cset w0, hi
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: not_isfinite_f_strictfp:
@@ -190,10 +179,9 @@ define i1 @not_isfinite_f_strictfp(float %x) strictfp {
;
; CHECK-NOFP-LABEL: not_isfinite_f_strictfp:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov w8, #2139095039 // =0x7f7fffff
-; CHECK-NOFP-NEXT: and w9, w0, #0x7fffffff
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: cset w0, gt
+; CHECK-NOFP-NEXT: ubfx w8, w0, #23, #8
+; CHECK-NOFP-NEXT: cmp w8, #254
+; CHECK-NOFP-NEXT: cset w0, hi
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 519) strictfp ; ~0x1f8 = ~"finite"
@@ -204,11 +192,10 @@ entry:
define i1 @isfinite_d(double %x) {
; CHECK-SD-LABEL: isfinite_d:
; CHECK-SD: // %bb.0: // %entry
-; CHECK-SD-NEXT: fmov x9, d0
-; CHECK-SD-NEXT: mov x8, #9218868437227405312 // =0x7ff0000000000000
-; CHECK-SD-NEXT: and x9, x9, #0x7fffffffffffffff
-; CHECK-SD-NEXT: cmp x9, x8
-; CHECK-SD-NEXT: cset w0, lt
+; CHECK-SD-NEXT: fmov x8, d0
+; CHECK-SD-NEXT: ubfx x8, x8, #52, #11
+; CHECK-SD-NEXT: cmp x8, #2047
+; CHECK-SD-NEXT: cset w0, lo
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: isfinite_d:
@@ -222,10 +209,9 @@ define i1 @isfinite_d(double %x) {
;
; CHECK-NOFP-LABEL: isfinite_d:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov x8, #9218868437227405312 // =0x7ff0000000000000
-; CHECK-NOFP-NEXT: and x9, x0, #0x7fffffffffffffff
-; CHECK-NOFP-NEXT: cmp x9, x8
-; CHECK-NOFP-NEXT: cset w0, lt
+; CHECK-NOFP-NEXT: ubfx x8, x0, #52, #11
+; CHECK-NOFP-NEXT: cmp x8, #2047
+; CHECK-NOFP-NEXT: cset w0, lo
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call i1 @llvm.is.fpclass.f64(double %x, i32 504) ; 0x1f8 = "finite"
@@ -235,11 +221,10 @@ entry:
define i1 @not_isfinite_d(double %x) {
; CHECK-SD-LABEL: not_isfinite_d:
; CHECK-SD: // %bb.0: // %entry
-; CHECK-SD-NEXT: fmov x9, d0
-; CHECK-SD-NEXT: mov x8, #9218868437227405311 // =0x7fefffffffffffff
-; CHECK-SD-NEXT: and x9, x9, #0x7fffffffffffffff
-; CHECK-SD-NEXT: cmp x9, x8
-; CHECK-SD-NEXT: cset w0, gt
+; CHECK-SD-NEXT: fmov x8, d0
+; CHECK-SD-NEXT: ubfx x8, x8, #52, #11
+; CHECK-SD-NEXT: cmp x8, #2046
+; CHECK-SD-NEXT: cset w0, hi
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: not_isfinite_d:
@@ -255,10 +240,9 @@ define i1 @not_isfinite_d(double %x) {
;
; CHECK-NOFP-LABEL: not_isfinite_d:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov x8, #9218868437227405311 // =0x7fefffffffffffff
-; CHECK-NOFP-NEXT: and x9, x0, #0x7fffffffffffffff
-; CHECK-NOFP-NEXT: cmp x9, x8
-; CHECK-NOFP-NEXT: cset w0, gt
+; CHECK-NOFP-NEXT: ubfx x8, x0, #52, #11
+; CHECK-NOFP-NEXT: cmp x8, #2046
+; CHECK-NOFP-NEXT: cset w0, hi
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call i1 @llvm.is.fpclass.f64(double %x, i32 519) ; ~0x1f8 = "~finite"
@@ -268,11 +252,10 @@ entry:
define i1 @isfinite_d_strictfp(double %x) {
; CHECK-SD-LABEL: isfinite_d_strictfp:
; CHECK-SD: // %bb.0: // %entry
-; CHECK-SD-NEXT: fmov x9, d0
-; CHECK-SD-NEXT: mov x8, #9218868437227405312 // =0x7ff0000000000000
-; CHECK-SD-NEXT: and x9, x9, #0x7fffffffffffffff
-; CHECK-SD-NEXT: cmp x9, x8
-; CHECK-SD-NEXT: cset w0, lt
+; CHECK-SD-NEXT: fmov x8, d0
+; CHECK-SD-NEXT: ubfx x8, x8, #52, #11
+; CHECK-SD-NEXT: cmp x8, #2047
+; CHECK-SD-NEXT: cset w0, lo
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: isfinite_d_strictfp:
@@ -286,10 +269,9 @@ define i1 @isfinite_d_strictfp(double %x) {
;
; CHECK-NOFP-LABEL: isfinite_d_strictfp:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov x8, #9218868437227405312 // =0x7ff0000000000000
-; CHECK-NOFP-NEXT: and x9, x0, #0x7fffffffffffffff
-; CHECK-NOFP-NEXT: cmp x9, x8
-; CHECK-NOFP-NEXT: cset w0, lt
+; CHECK-NOFP-NEXT: ubfx x8, x0, #52, #11
+; CHECK-NOFP-NEXT: cmp x8, #2047
+; CHECK-NOFP-NEXT: cset w0, lo
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call i1 @llvm.is.fpclass.f64(double %x, i32 504) strictfp ; 0x1f8 = "finite"
@@ -299,11 +281,10 @@ entry:
define i1 @not_isfinite_d_strictfp(double %x) {
; CHECK-SD-LABEL: not_isfinite_d_strictfp:
; CHECK-SD: // %bb.0: // %entry
-; CHECK-SD-NEXT: fmov x9, d0
-; CHECK-SD-NEXT: mov x8, #9218868437227405311 // =0x7fefffffffffffff
-; CHECK-SD-NEXT: and x9, x9, #0x7fffffffffffffff
-; CHECK-SD-NEXT: cmp x9, x8
-; CHECK-SD-NEXT: cset w0, gt
+; CHECK-SD-NEXT: fmov x8, d0
+; CHECK-SD-NEXT: ubfx x8, x8, #52, #11
+; CHECK-SD-NEXT: cmp x8, #2046
+; CHECK-SD-NEXT: cset w0, hi
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: not_isfinite_d_strictfp:
@@ -319,10 +300,9 @@ define i1 @not_isfinite_d_strictfp(double %x) {
;
; CHECK-NOFP-LABEL: not_isfinite_d_strictfp:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov x8, #9218868437227405311 // =0x7fefffffffffffff
-; CHECK-NOFP-NEXT: and x9, x0, #0x7fffffffffffffff
-; CHECK-NOFP-NEXT: cmp x9, x8
-; CHECK-NOFP-NEXT: cset w0, gt
+; CHECK-NOFP-NEXT: ubfx x8, x0, #52, #11
+; CHECK-NOFP-NEXT: cmp x8, #2046
+; CHECK-NOFP-NEXT: cset w0, hi
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call i1 @llvm.is.fpclass.f64(double %x, i32 519) strictfp ; ~0x1f8 = "~finite"
@@ -334,12 +314,11 @@ define i1 @isfinite_f128(fp128 %x) {
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: str q0, [sp, #-16]!
; CHECK-SD-NEXT: .cfi_def_cfa_offset 16
-; CHECK-SD-NEXT: ldr x9, [sp, #8]
-; CHECK-SD-NEXT: mov x8, #9223090561878065152 // =0x7fff000000000000
-; CHECK-SD-NEXT: and x9, x9, #0x7fffffffffffffff
-; CHECK-SD-NEXT: cmp x9, x8
-; CHECK-SD-NEXT: cset w0, lt
-; CHECK-SD-NEXT: add sp, sp, #16
+; CHECK-SD-NEXT: ldp x8, x9, [sp], #16
+; CHECK-SD-NEXT: extr x8, x9, x8, #63
+; CHECK-SD-NEXT: mov x9, #-562949953421312 // =0xfffe000000000000
+; CHECK-SD-NEXT: cmp x8, x9
+; CHECK-SD-NEXT: cset w0, lo
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: isfinite_f128:
@@ -354,10 +333,10 @@ define i1 @isfinite_f128(fp128 %x) {
;
; CHECK-NOFP-LABEL: isfinite_f128:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov x8, #9223090561878065152 // =0x7fff000000000000
-; CHECK-NOFP-NEXT: and x9, x1, #0x7fffffffffffffff
-; CHECK-NOFP-NEXT: cmp x9, x8
-; CHECK-NOFP-NEXT: cset w0, lt
+; CHECK-NOFP-NEXT: extr x8, x1, x0, #63
+; CHECK-NOFP-NEXT: mov x9, #-562949953421312 // =0xfffe000000000000
+; CHECK-NOFP-NEXT: cmp x8, x9
+; CHECK-NOFP-NEXT: cset w0, lo
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call i1 @llvm.is.fpclass.f128(fp128 %x, i32 504) ; 0x1f8 = "finite"
@@ -369,12 +348,11 @@ define i1 @not_isfinite_f128(fp128 %x) {
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: str q0, [sp, #-16]!
; CHECK-SD-NEXT: .cfi_def_cfa_offset 16
-; CHECK-SD-NEXT: ldr x9, [sp, #8]
-; CHECK-SD-NEXT: mov x8, #9223090561878065151 // =0x7ffeffffffffffff
-; CHECK-SD-NEXT: and x9, x9, #0x7fffffffffffffff
-; CHECK-SD-NEXT: cmp x9, x8
-; CHECK-SD-NEXT: cset w0, gt
-; CHECK-SD-NEXT: add sp, sp, #16
+; CHECK-SD-NEXT: ldp x8, x9, [sp], #16
+; CHECK-SD-NEXT: extr x8, x9, x8, #63
+; CHECK-SD-NEXT: mov x9, #-562949953421313 // =0xfffdffffffffffff
+; CHECK-SD-NEXT: cmp x8, x9
+; CHECK-SD-NEXT: cset w0, hi
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: not_isfinite_f128:
@@ -397,10 +375,10 @@ define i1 @not_isfinite_f128(fp128 %x) {
;
; CHECK-NOFP-LABEL: not_isfinite_f128:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov x8, #9223090561878065151 // =0x7ffeffffffffffff
-; CHECK-NOFP-NEXT: and x9, x1, #0x7fffffffffffffff
-; CHECK-NOFP-NEXT: cmp x9, x8
-; CHECK-NOFP-NEXT: cset w0, gt
+; CHECK-NOFP-NEXT: extr x8, x1, x0, #63
+; CHECK-NOFP-NEXT: mov x9, #-562949953421313 // =0xfffdffffffffffff
+; CHECK-NOFP-NEXT: cmp x8, x9
+; CHECK-NOFP-NEXT: cset w0, hi
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call i1 @llvm.is.fpclass.f128(fp128 %x, i32 519) ; ~0x1f8 = "~finite"
@@ -410,9 +388,9 @@ entry:
define <4 x i1> @isfinite_v4h(<4 x half> %x) {
; CHECK-SD-LABEL: isfinite_v4h:
; CHECK-SD: // %bb.0: // %entry
-; CHECK-SD-NEXT: movi v1.4h, #124, lsl #8
-; CHECK-SD-NEXT: bic v0.4h, #128, lsl #8
-; CHECK-SD-NEXT: cmgt v0.4h, v1.4h, v0.4h
+; CHECK-SD-NEXT: movi v1.4h, #248, lsl #8
+; CHECK-SD-NEXT: add v0.4h, v0.4h, v0.4h
+; CHECK-SD-NEXT: cmhi v0.4h, v1.4h, v0.4h
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: isfinite_v4h:
@@ -430,19 +408,18 @@ define <4 x i1> @isfinite_v4h(<4 x half> %x) {
;
; CHECK-NOFP-LABEL: isfinite_v4h:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov w8, #31744 // =0x7c00
-; CHECK-NOFP-NEXT: and w9, w0, #0x7fff
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: and w9, w1, #0x7fff
-; CHECK-NOFP-NEXT: cset w0, lt
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: and w9, w2, #0x7fff
-; CHECK-NOFP-NEXT: cset w1, lt
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: and w9, w3, #0x7fff
-; CHECK-NOFP-NEXT: cset w2, lt
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: cset w3, lt
+; CHECK-NOFP-NEXT: ubfx w8, w0, #10, #5
+; CHECK-NOFP-NEXT: ubfx w9, w1, #10, #5
+; CHECK-NOFP-NEXT: cmp w8, #31
+; CHECK-NOFP-NEXT: ubfx w8, w2, #10, #5
+; CHECK-NOFP-NEXT: cset w0, lo
+; CHECK-NOFP-NEXT: cmp w9, #31
+; CHECK-NOFP-NEXT: ubfx w9, w3, #10, #5
+; CHECK-NOFP-NEXT: cset w1, lo
+; CHECK-NOFP-NEXT: cmp w8, #31
+; CHECK-NOFP-NEXT: cset w2, lo
+; CHECK-NOFP-NEXT: cmp w9, #31
+; CHECK-NOFP-NEXT: cset w3, lo
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call <4 x i1> @llvm.is.fpclass.v4f16(<4 x half> %x, i32 504) ; 0x1f8 = "finite"
@@ -452,9 +429,9 @@ entry:
define <4 x i1> @not_isfinite_v4h(<4 x half> %x) {
; CHECK-SD-LABEL: not_isfinite_v4h:
; CHECK-SD: // %bb.0: // %entry
-; CHECK-SD-NEXT: movi v1.4h, #124, lsl #8
-; CHECK-SD-NEXT: bic v0.4h, #128, lsl #8
-; CHECK-SD-NEXT: cmge v0.4h, v0.4h, v1.4h
+; CHECK-SD-NEXT: movi v1.4h, #248, lsl #8
+; CHECK-SD-NEXT: add v0.4h, v0.4h, v0.4h
+; CHECK-SD-NEXT: cmhs v0.4h, v0.4h, v1.4h
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: not_isfinite_v4h:
@@ -474,19 +451,18 @@ define <4 x i1> @not_isfinite_v4h(<4 x half> %x) {
;
; CHECK-NOFP-LABEL: not_isfinite_v4h:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov w8, #31743 // =0x7bff
-; CHECK-NOFP-NEXT: and w9, w0, #0x7fff
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: and w9, w1, #0x7fff
-; CHECK-NOFP-NEXT: cset w0, gt
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: and w9, w2, #0x7fff
-; CHECK-NOFP-NEXT: cset w1, gt
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: and w9, w3, #0x7fff
-; CHECK-NOFP-NEXT: cset w2, gt
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: cset w3, gt
+; CHECK-NOFP-NEXT: ubfx w8, w0, #10, #5
+; CHECK-NOFP-NEXT: ubfx w9, w1, #10, #5
+; CHECK-NOFP-NEXT: cmp w8, #30
+; CHECK-NOFP-NEXT: ubfx w8, w2, #10, #5
+; CHECK-NOFP-NEXT: cset w0, hi
+; CHECK-NOFP-NEXT: cmp w9, #30
+; CHECK-NOFP-NEXT: ubfx w9, w3, #10, #5
+; CHECK-NOFP-NEXT: cset w1, hi
+; CHECK-NOFP-NEXT: cmp w8, #30
+; CHECK-NOFP-NEXT: cset w2, hi
+; CHECK-NOFP-NEXT: cmp w9, #30
+; CHECK-NOFP-NEXT: cset w3, hi
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call <4 x i1> @llvm.is.fpclass.v4f16(<4 x half> %x, i32 519) ; ~0x1f8 = "~finite"
@@ -522,19 +498,18 @@ define <4 x i1> @isfinite_v4f(<4 x float> %x) {
;
; CHECK-NOFP-LABEL: isfinite_v4f:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov w8, #2139095040 // =0x7f800000
-; CHECK-NOFP-NEXT: and w9, w0, #0x7fffffff
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: and w9, w1, #0x7fffffff
-; CHECK-NOFP-NEXT: cset w0, lt
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: and w9, w2, #0x7fffffff
-; CHECK-NOFP-NEXT: cset w1, lt
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: and w9, w3, #0x7fffffff
-; CHECK-NOFP-NEXT: cset w2, lt
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: cset w3, lt
+; CHECK-NOFP-NEXT: ubfx w8, w0, #23, #8
+; CHECK-NOFP-NEXT: ubfx w9, w1, #23, #8
+; CHECK-NOFP-NEXT: cmp w8, #255
+; CHECK-NOFP-NEXT: ubfx w8, w2, #23, #8
+; CHECK-NOFP-NEXT: cset w0, lo
+; CHECK-NOFP-NEXT: cmp w9, #255
+; CHECK-NOFP-NEXT: ubfx w9, w3, #23, #8
+; CHECK-NOFP-NEXT: cset w1, lo
+; CHECK-NOFP-NEXT: cmp w8, #255
+; CHECK-NOFP-NEXT: cset w2, lo
+; CHECK-NOFP-NEXT: cmp w9, #255
+; CHECK-NOFP-NEXT: cset w3, lo
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call <4 x i1> @llvm.is.fpclass.v4f32(<4 x float> %x, i32 504) ; 0x1f8 = "finite"
@@ -544,10 +519,9 @@ entry:
define <4 x i1> @not_isfinite_v4f(<4 x float> %x) {
; CHECK-SD-LABEL: not_isfinite_v4f:
; CHECK-SD: // %bb.0: // %entry
-; CHECK-SD-NEXT: mvni v1.4s, #127, msl #16
-; CHECK-SD-NEXT: bic v0.4s, #128, lsl #24
-; CHECK-SD-NEXT: fneg v1.4s, v1.4s
-; CHECK-SD-NEXT: cmge v0.4s, v0.4s, v1.4s
+; CHECK-SD-NEXT: movi v1.2d, #0xff000000ff000000
+; CHECK-SD-NEXT: add v0.4s, v0.4s, v0.4s
+; CHECK-SD-NEXT: cmhs v0.4s, v0.4s, v1.4s
; CHECK-SD-NEXT: xtn v0.4h, v0.4s
; CHECK-SD-NEXT: ret
;
@@ -571,19 +545,18 @@ define <4 x i1> @not_isfinite_v4f(<4 x float> %x) {
;
; CHECK-NOFP-LABEL: not_isfinite_v4f:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov w8, #2139095039 // =0x7f7fffff
-; CHECK-NOFP-NEXT: and w9, w0, #0x7fffffff
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: and w9, w1, #0x7fffffff
-; CHECK-NOFP-NEXT: cset w0, gt
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: and w9, w2, #0x7fffffff
-; CHECK-NOFP-NEXT: cset w1, gt
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: and w9, w3, #0x7fffffff
-; CHECK-NOFP-NEXT: cset w2, gt
-; CHECK-NOFP-NEXT: cmp w9, w8
-; CHECK-NOFP-NEXT: cset w3, gt
+; CHECK-NOFP-NEXT: ubfx w8, w0, #23, #8
+; CHECK-NOFP-NEXT: ubfx w9, w1, #23, #8
+; CHECK-NOFP-NEXT: cmp w8, #254
+; CHECK-NOFP-NEXT: ubfx w8, w2, #23, #8
+; CHECK-NOFP-NEXT: cset w0, hi
+; CHECK-NOFP-NEXT: cmp w9, #254
+; CHECK-NOFP-NEXT: ubfx w9, w3, #23, #8
+; CHECK-NOFP-NEXT: cset w1, hi
+; CHECK-NOFP-NEXT: cmp w8, #254
+; CHECK-NOFP-NEXT: cset w2, hi
+; CHECK-NOFP-NEXT: cmp w9, #254
+; CHECK-NOFP-NEXT: cset w3, hi
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call <4 x i1> @llvm.is.fpclass.v4f32(<4 x float> %x, i32 519) ; ~0x1f8 = "~finite"
@@ -615,13 +588,12 @@ define <2 x i1> @isfinite_v2d(<2 x double> %x) {
;
; CHECK-NOFP-LABEL: isfinite_v2d:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov x8, #9218868437227405312 // =0x7ff0000000000000
-; CHECK-NOFP-NEXT: and x9, x0, #0x7fffffffffffffff
-; CHECK-NOFP-NEXT: cmp x9, x8
-; CHECK-NOFP-NEXT: and x9, x1, #0x7fffffffffffffff
-; CHECK-NOFP-NEXT: cset w0, lt
-; CHECK-NOFP-NEXT: cmp x9, x8
-; CHECK-NOFP-NEXT: cset w1, lt
+; CHECK-NOFP-NEXT: ubfx x8, x0, #52, #11
+; CHECK-NOFP-NEXT: ubfx x9, x1, #52, #11
+; CHECK-NOFP-NEXT: cmp x8, #2047
+; CHECK-NOFP-NEXT: cset w0, lo
+; CHECK-NOFP-NEXT: cmp x9, #2047
+; CHECK-NOFP-NEXT: cset w1, lo
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call <2 x i1> @llvm.is.fpclass.v2f64(<2 x double> %x, i32 504) ; 0x1f8 = "finite"
@@ -631,12 +603,10 @@ entry:
define <2 x i1> @not_isfinite_v2d(<2 x double> %x) {
; CHECK-SD-LABEL: not_isfinite_v2d:
; CHECK-SD: // %bb.0: // %entry
-; CHECK-SD-NEXT: movi v1.2d, #0xffffffffffffffff
-; CHECK-SD-NEXT: mov x8, #9218868437227405312 // =0x7ff0000000000000
-; CHECK-SD-NEXT: fneg v1.2d, v1.2d
-; CHECK-SD-NEXT: and v0.16b, v0.16b, v1.16b
+; CHECK-SD-NEXT: mov x8, #-9007199254740992 // =0xffe0000000000000
+; CHECK-SD-NEXT: add v0.2d, v0.2d, v0.2d
; CHECK-SD-NEXT: dup v1.2d, x8
-; CHECK-SD-NEXT: cmge v0.2d, v0.2d, v1.2d
+; CHECK-SD-NEXT: cmhs v0.2d, v0.2d, v1.2d
; CHECK-SD-NEXT: xtn v0.2s, v0.2d
; CHECK-SD-NEXT: ret
;
@@ -655,13 +625,12 @@ define <2 x i1> @not_isfinite_v2d(<2 x double> %x) {
;
; CHECK-NOFP-LABEL: not_isfinite_v2d:
; CHECK-NOFP: // %bb.0: // %entry
-; CHECK-NOFP-NEXT: mov x8, #9218868437227405311 // =0x7fefffffffffffff
-; CHECK-NOFP-NEXT: and x9, x0, #0x7fffffffffffffff
-; CHECK-NOFP-NEXT: cmp x9, x8
-; CHECK-NOFP-NEXT: and x9, x1, #0x7fffffffffffffff
-; CHECK-NOFP-NEXT: cset w0, gt
-; CHECK-NOFP-NEXT: cmp x9, x8
-; CHECK-NOFP-NEXT: cset w1, gt
+; CHECK-NOFP-NEXT: ubfx x8, x0, #52, #11
+; CHECK-NOFP-NEXT: ubfx x9, x1, #52, #11
+; CHECK-NOFP-NEXT: cmp x8, #2046
+; CHECK-NOFP-NEXT: cset w0, hi
+; CHECK-NOFP-NEXT: cmp x9, #2046
+; CHECK-NOFP-NEXT: cset w1, hi
; CHECK-NOFP-NEXT: ret
entry:
%0 = tail call <2 x i1> @llvm.is.fpclass.v2f64(<2 x double> %x, i32 519) ; ~0x1f8 = "~finite"
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.is.fpclass.bf16.ll b/llvm/test/CodeGen/AMDGPU/llvm.is.fpclass.bf16.ll
index d80fb6d8ea108..52e9474f46777 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.is.fpclass.bf16.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.is.fpclass.bf16.ll
@@ -1316,51 +1316,52 @@ define i1 @isfinite_bf16(bfloat %x) nounwind {
; GFX7CHECK-LABEL: isfinite_bf16:
; GFX7CHECK: ; %bb.0:
; GFX7CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7CHECK-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX7CHECK-NEXT: s_movk_i32 s4, 0x7f80
-; GFX7CHECK-NEXT: v_cmp_gt_i32_e32 vcc, s4, v0
+; GFX7CHECK-NEXT: v_lshlrev_b32_e32 v0, 1, v0
+; GFX7CHECK-NEXT: v_and_b32_e32 v0, 0xfffe, v0
+; GFX7CHECK-NEXT: s_mov_b32 s4, 0xff00
+; GFX7CHECK-NEXT: v_cmp_gt_u32_e32 vcc, s4, v0
; GFX7CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GFX7CHECK-NEXT: s_setpc_b64 s[30:31]
;
; GFX8CHECK-LABEL: isfinite_bf16:
; GFX8CHECK: ; %bb.0:
; GFX8CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8CHECK-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX8CHECK-NEXT: s_movk_i32 s4, 0x7f80
-; GFX8CHECK-NEXT: v_cmp_gt_i16_e32 vcc, s4, v0
+; GFX8CHECK-NEXT: v_lshlrev_b16_e32 v0, 1, v0
+; GFX8CHECK-NEXT: s_movk_i32 s4, 0xff00
+; GFX8CHECK-NEXT: v_cmp_gt_u16_e32 vcc, s4, v0
; GFX8CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GFX8CHECK-NEXT: s_setpc_b64 s[30:31]
;
; GFX9CHECK-LABEL: isfinite_bf16:
; GFX9CHECK: ; %bb.0:
; GFX9CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9CHECK-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX9CHECK-NEXT: s_movk_i32 s4, 0x7f80
-; GFX9CHECK-NEXT: v_cmp_gt_i16_e32 vcc, s4, v0
+; GFX9CHECK-NEXT: v_lshlrev_b16_e32 v0, 1, v0
+; GFX9CHECK-NEXT: s_movk_i32 s4, 0xff00
+; GFX9CHECK-NEXT: v_cmp_gt_u16_e32 vcc, s4, v0
; GFX9CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GFX9CHECK-NEXT: s_setpc_b64 s[30:31]
;
; GFX10CHECK-LABEL: isfinite_bf16:
; GFX10CHECK: ; %bb.0:
; GFX10CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10CHECK-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX10CHECK-NEXT: v_cmp_gt_i16_e32 vcc_lo, 0x7f80, v0
+; GFX10CHECK-NEXT: v_lshlrev_b16 v0, 1, v0
+; GFX10CHECK-NEXT: v_cmp_gt_u16_e32 vcc_lo, 0xff00, v0
; GFX10CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
; GFX10CHECK-NEXT: s_setpc_b64 s[30:31]
;
; GFX11SELDAG-TRUE16-LABEL: isfinite_bf16:
; GFX11SELDAG-TRUE16: ; %bb.0:
; GFX11SELDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11SELDAG-TRUE16-NEXT: v_and_b16 v0.l, 0x7fff, v0.l
-; GFX11SELDAG-TRUE16-NEXT: v_cmp_gt_i16_e32 vcc_lo, 0x7f80, v0.l
+; GFX11SELDAG-TRUE16-NEXT: v_lshlrev_b16 v0.l, 1, v0.l
+; GFX11SELDAG-TRUE16-NEXT: v_cmp_gt_u16_e32 vcc_lo, 0xff00, v0.l
; GFX11SELDAG-TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
; GFX11SELDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11SELDAG-FAKE16-LABEL: isfinite_bf16:
; GFX11SELDAG-FAKE16: ; %bb.0:
; GFX11SELDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11SELDAG-FAKE16-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX11SELDAG-FAKE16-NEXT: v_cmp_gt_i16_e32 vcc_lo, 0x7f80, v0
+; GFX11SELDAG-FAKE16-NEXT: v_lshlrev_b16 v0, 1, v0
+; GFX11SELDAG-FAKE16-NEXT: v_cmp_gt_u16_e32 vcc_lo, 0xff00, v0
; GFX11SELDAG-FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
; GFX11SELDAG-FAKE16-NEXT: s_setpc_b64 s[30:31]
%1 = call i1 @llvm.is.fpclass.bf16(bfloat %x, i32 504) ; 0x1f8 = "finite"
@@ -3077,51 +3078,52 @@ define i1 @isinf_or_nan_bf16(bfloat %x) {
; GFX7CHECK-LABEL: isinf_or_nan_bf16:
; GFX7CHECK: ; %bb.0: ; %entry
; GFX7CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7CHECK-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX7CHECK-NEXT: s_movk_i32 s4, 0x7f7f
-; GFX7CHECK-NEXT: v_cmp_lt_i32_e32 vcc, s4, v0
+; GFX7CHECK-NEXT: v_lshlrev_b32_e32 v0, 1, v0
+; GFX7CHECK-NEXT: v_and_b32_e32 v0, 0xfffe, v0
+; GFX7CHECK-NEXT: s_mov_b32 s4, 0xfeff
+; GFX7CHECK-NEXT: v_cmp_lt_u32_e32 vcc, s4, v0
; GFX7CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GFX7CHECK-NEXT: s_setpc_b64 s[30:31]
;
; GFX8CHECK-LABEL: isinf_or_nan_bf16:
; GFX8CHECK: ; %bb.0: ; %entry
; GFX8CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8CHECK-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX8CHECK-NEXT: s_movk_i32 s4, 0x7f7f
-; GFX8CHECK-NEXT: v_cmp_lt_i16_e32 vcc, s4, v0
+; GFX8CHECK-NEXT: v_lshlrev_b16_e32 v0, 1, v0
+; GFX8CHECK-NEXT: s_movk_i32 s4, 0xfeff
+; GFX8CHECK-NEXT: v_cmp_lt_u16_e32 vcc, s4, v0
; GFX8CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GFX8CHECK-NEXT: s_setpc_b64 s[30:31]
;
; GFX9CHECK-LABEL: isinf_or_nan_bf16:
; GFX9CHECK: ; %bb.0: ; %entry
; GFX9CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9CHECK-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX9CHECK-NEXT: s_movk_i32 s4, 0x7f7f
-; GFX9CHECK-NEXT: v_cmp_lt_i16_e32 vcc, s4, v0
+; GFX9CHECK-NEXT: v_lshlrev_b16_e32 v0, 1, v0
+; GFX9CHECK-NEXT: s_movk_i32 s4, 0xfeff
+; GFX9CHECK-NEXT: v_cmp_lt_u16_e32 vcc, s4, v0
; GFX9CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GFX9CHECK-NEXT: s_setpc_b64 s[30:31]
;
; GFX10CHECK-LABEL: isinf_or_nan_bf16:
; GFX10CHECK: ; %bb.0: ; %entry
; GFX10CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10CHECK-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX10CHECK-NEXT: v_cmp_lt_i16_e32 vcc_lo, 0x7f7f, v0
+; GFX10CHECK-NEXT: v_lshlrev_b16 v0, 1, v0
+; GFX10CHECK-NEXT: v_cmp_lt_u16_e32 vcc_lo, 0xfeff, v0
; GFX10CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
; GFX10CHECK-NEXT: s_setpc_b64 s[30:31]
;
; GFX11SELDAG-TRUE16-LABEL: isinf_or_nan_bf16:
; GFX11SELDAG-TRUE16: ; %bb.0: ; %entry
; GFX11SELDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11SELDAG-TRUE16-NEXT: v_and_b16 v0.l, 0x7fff, v0.l
-; GFX11SELDAG-TRUE16-NEXT: v_cmp_lt_i16_e32 vcc_lo, 0x7f7f, v0.l
+; GFX11SELDAG-TRUE16-NEXT: v_lshlrev_b16 v0.l, 1, v0.l
+; GFX11SELDAG-TRUE16-NEXT: v_cmp_lt_u16_e32 vcc_lo, 0xfeff, v0.l
; GFX11SELDAG-TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
; GFX11SELDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11SELDAG-FAKE16-LABEL: isinf_or_nan_bf16:
; GFX11SELDAG-FAKE16: ; %bb.0: ; %entry
; GFX11SELDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11SELDAG-FAKE16-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX11SELDAG-FAKE16-NEXT: v_cmp_lt_i16_e32 vcc_lo, 0x7f7f, v0
+; GFX11SELDAG-FAKE16-NEXT: v_lshlrev_b16 v0, 1, v0
+; GFX11SELDAG-FAKE16-NEXT: v_cmp_lt_u16_e32 vcc_lo, 0xfeff, v0
; GFX11SELDAG-FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
; GFX11SELDAG-FAKE16-NEXT: s_setpc_b64 s[30:31]
entry:
@@ -3133,51 +3135,52 @@ define i1 @not_isinf_or_nan_bf16(bfloat %x) {
; GFX7CHECK-LABEL: not_isinf_or_nan_bf16:
; GFX7CHECK: ; %bb.0: ; %entry
; GFX7CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7CHECK-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX7CHECK-NEXT: s_movk_i32 s4, 0x7f80
-; GFX7CHECK-NEXT: v_cmp_gt_i32_e32 vcc, s4, v0
+; GFX7CHECK-NEXT: v_lshlrev_b32_e32 v0, 1, v0
+; GFX7CHECK-NEXT: v_and_b32_e32 v0, 0xfffe, v0
+; GFX7CHECK-NEXT: s_mov_b32 s4, 0xff00
+; GFX7CHECK-NEXT: v_cmp_gt_u32_e32 vcc, s4, v0
; GFX7CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GFX7CHECK-NEXT: s_setpc_b64 s[30:31]
;
; GFX8CHECK-LABEL: not_isinf_or_nan_bf16:
; GFX8CHECK: ; %bb.0: ; %entry
; GFX8CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8CHECK-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX8CHECK-NEXT: s_movk_i32 s4, 0x7f80
-; GFX8CHECK-NEXT: v_cmp_gt_i16_e32 vcc, s4, v0
+; GFX8CHECK-NEXT: v_lshlrev_b16_e32 v0, 1, v0
+; GFX8CHECK-NEXT: s_movk_i32 s4, 0xff00
+; GFX8CHECK-NEXT: v_cmp_gt_u16_e32 vcc, s4, v0
; GFX8CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GFX8CHECK-NEXT: s_setpc_b64 s[30:31]
;
; GFX9CHECK-LABEL: not_isinf_or_nan_bf16:
; GFX9CHECK: ; %bb.0: ; %entry
; GFX9CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9CHECK-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX9CHECK-NEXT: s_movk_i32 s4, 0x7f80
-; GFX9CHECK-NEXT: v_cmp_gt_i16_e32 vcc, s4, v0
+; GFX9CHECK-NEXT: v_lshlrev_b16_e32 v0, 1, v0
+; GFX9CHECK-NEXT: s_movk_i32 s4, 0xff00
+; GFX9CHECK-NEXT: v_cmp_gt_u16_e32 vcc, s4, v0
; GFX9CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GFX9CHECK-NEXT: s_setpc_b64 s[30:31]
;
; GFX10CHECK-LABEL: not_isinf_or_nan_bf16:
; GFX10CHECK: ; %bb.0: ; %entry
; GFX10CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10CHECK-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX10CHECK-NEXT: v_cmp_gt_i16_e32 vcc_lo, 0x7f80, v0
+; GFX10CHECK-NEXT: v_lshlrev_b16 v0, 1, v0
+; GFX10CHECK-NEXT: v_cmp_gt_u16_e32 vcc_lo, 0xff00, v0
; GFX10CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
; GFX10CHECK-NEXT: s_setpc_b64 s[30:31]
;
; GFX11SELDAG-TRUE16-LABEL: not_isinf_or_nan_bf16:
; GFX11SELDAG-TRUE16: ; %bb.0: ; %entry
; GFX11SELDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11SELDAG-TRUE16-NEXT: v_and_b16 v0.l, 0x7fff, v0.l
-; GFX11SELDAG-TRUE16-NEXT: v_cmp_gt_i16_e32 vcc_lo, 0x7f80, v0.l
+; GFX11SELDAG-TRUE16-NEXT: v_lshlrev_b16 v0.l, 1, v0.l
+; GFX11SELDAG-TRUE16-NEXT: v_cmp_gt_u16_e32 vcc_lo, 0xff00, v0.l
; GFX11SELDAG-TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
; GFX11SELDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11SELDAG-FAKE16-LABEL: not_isinf_or_nan_bf16:
; GFX11SELDAG-FAKE16: ; %bb.0: ; %entry
; GFX11SELDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11SELDAG-FAKE16-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX11SELDAG-FAKE16-NEXT: v_cmp_gt_i16_e32 vcc_lo, 0x7f80, v0
+; GFX11SELDAG-FAKE16-NEXT: v_lshlrev_b16 v0, 1, v0
+; GFX11SELDAG-FAKE16-NEXT: v_cmp_gt_u16_e32 vcc_lo, 0xff00, v0
; GFX11SELDAG-FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
; GFX11SELDAG-FAKE16-NEXT: s_setpc_b64 s[30:31]
entry:
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.is.fpclass.f16.ll b/llvm/test/CodeGen/AMDGPU/llvm.is.fpclass.f16.ll
index 76958e63d36c9..b3224dd565fe5 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.is.fpclass.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.is.fpclass.f16.ll
@@ -1977,9 +1977,10 @@ define i1 @isfinite_f16(half %x) nounwind {
; GFX7SELDAG-LABEL: isfinite_f16:
; GFX7SELDAG: ; %bb.0:
; GFX7SELDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7SELDAG-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX7SELDAG-NEXT: s_movk_i32 s4, 0x7c00
-; GFX7SELDAG-NEXT: v_cmp_gt_i32_e32 vcc, s4, v0
+; GFX7SELDAG-NEXT: v_lshlrev_b32_e32 v0, 1, v0
+; GFX7SELDAG-NEXT: v_and_b32_e32 v0, 0xfffe, v0
+; GFX7SELDAG-NEXT: s_mov_b32 s4, 0xf800
+; GFX7SELDAG-NEXT: v_cmp_gt_u32_e32 vcc, s4, v0
; GFX7SELDAG-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GFX7SELDAG-NEXT: s_setpc_b64 s[30:31]
;
@@ -4103,9 +4104,10 @@ define i1 @isinf_or_nan_f16(half %x) {
; GFX7SELDAG-LABEL: isinf_or_nan_f16:
; GFX7SELDAG: ; %bb.0: ; %entry
; GFX7SELDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7SELDAG-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX7SELDAG-NEXT: s_movk_i32 s4, 0x7bff
-; GFX7SELDAG-NEXT: v_cmp_lt_i32_e32 vcc, s4, v0
+; GFX7SELDAG-NEXT: v_lshlrev_b32_e32 v0, 1, v0
+; GFX7SELDAG-NEXT: v_and_b32_e32 v0, 0xfffe, v0
+; GFX7SELDAG-NEXT: s_mov_b32 s4, 0xf7ff
+; GFX7SELDAG-NEXT: v_cmp_lt_u32_e32 vcc, s4, v0
; GFX7SELDAG-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GFX7SELDAG-NEXT: s_setpc_b64 s[30:31]
;
@@ -4180,9 +4182,10 @@ define i1 @not_isinf_or_nan_f16(half %x) {
; GFX7SELDAG-LABEL: not_isinf_or_nan_f16:
; GFX7SELDAG: ; %bb.0: ; %entry
; GFX7SELDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7SELDAG-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX7SELDAG-NEXT: s_movk_i32 s4, 0x7c00
-; GFX7SELDAG-NEXT: v_cmp_gt_i32_e32 vcc, s4, v0
+; GFX7SELDAG-NEXT: v_lshlrev_b32_e32 v0, 1, v0
+; GFX7SELDAG-NEXT: v_and_b32_e32 v0, 0xfffe, v0
+; GFX7SELDAG-NEXT: s_mov_b32 s4, 0xf800
+; GFX7SELDAG-NEXT: v_cmp_gt_u32_e32 vcc, s4, v0
; GFX7SELDAG-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GFX7SELDAG-NEXT: s_setpc_b64 s[30:31]
;
diff --git a/llvm/test/CodeGen/AMDGPU/r600.llvm.is.fpclass.ll b/llvm/test/CodeGen/AMDGPU/r600.llvm.is.fpclass.ll
index 3116094fe1dfd..08039833e2a66 100644
--- a/llvm/test/CodeGen/AMDGPU/r600.llvm.is.fpclass.ll
+++ b/llvm/test/CodeGen/AMDGPU/r600.llvm.is.fpclass.ll
@@ -25,15 +25,14 @@ define amdgpu_kernel void @isnan_f32(ptr addrspace(1) %out, float %x) {
define amdgpu_kernel void @issue135083_f32(ptr addrspace(1) %out, float %x) {
; CM-LABEL: issue135083_f32:
; CM: ; %bb.0:
-; CM-NEXT: ALU 6, @4, KC0[CB0:0-32], KC1[]
+; CM-NEXT: ALU 5, @4, KC0[CB0:0-32], KC1[]
; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T0.X, T1.X
; CM-NEXT: CF_END
; CM-NEXT: PAD
; CM-NEXT: ALU clause starting at 4:
-; CM-NEXT: AND_INT * T0.W, KC0[2].Z, literal.x,
-; CM-NEXT: 2147483647(nan), 0(0.000000e+00)
-; CM-NEXT: SETGT_INT * T0.W, literal.x, PV.W,
-; CM-NEXT: 2139095040(INF), 0(0.000000e+00)
+; CM-NEXT: LSHL * T0.W, KC0[2].Z, 1,
+; CM-NEXT: SETGT_UINT * T0.W, literal.x, PV.W,
+; CM-NEXT: -16777216(-1.701412e+38), 0(0.000000e+00)
; CM-NEXT: AND_INT * T0.X, PV.W, 1,
; CM-NEXT: LSHR * T1.X, KC0[2].Y, literal.x,
; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00)
@@ -46,19 +45,18 @@ define amdgpu_kernel void @issue135083_f32(ptr addrspace(1) %out, float %x) {
define amdgpu_kernel void @issue135083_v2f32(ptr addrspace(1) %out, <2 x float> %x) {
; CM-LABEL: issue135083_v2f32:
; CM: ; %bb.0:
-; CM-NEXT: ALU 10, @4, KC0[CB0:0-32], KC1[]
+; CM-NEXT: ALU 9, @4, KC0[CB0:0-32], KC1[]
; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T0, T1.X
; CM-NEXT: CF_END
; CM-NEXT: PAD
; CM-NEXT: ALU clause starting at 4:
-; CM-NEXT: AND_INT * T0.W, KC0[3].X, literal.x,
-; CM-NEXT: 2147483647(nan), 0(0.000000e+00)
-; CM-NEXT: AND_INT T0.Z, KC0[2].W, literal.x,
-; CM-NEXT: SETGT_INT * T0.W, PV.W, literal.y,
-; CM-NEXT: 2147483647(nan), 2139095039(3.402823e+38)
+; CM-NEXT: LSHL * T0.W, KC0[3].X, 1,
+; CM-NEXT: LSHL T0.Z, KC0[2].W, 1,
+; CM-NEXT: SETGT_UINT * T0.W, PV.W, literal.x,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
; CM-NEXT: CNDE_INT T0.Y, PV.W, 1, 0.0,
-; CM-NEXT: SETGT_INT * T0.W, PV.Z, literal.x,
-; CM-NEXT: 2139095039(3.402823e+38), 0(0.000000e+00)
+; CM-NEXT: SETGT_UINT * T0.W, PV.Z, literal.x,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
; CM-NEXT: CNDE_INT * T0.X, PV.W, 1, 0.0,
; CM-NEXT: LSHR * T1.X, KC0[2].Y, literal.x,
; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00)
@@ -71,26 +69,24 @@ define amdgpu_kernel void @issue135083_v2f32(ptr addrspace(1) %out, <2 x float>
define amdgpu_kernel void @issue135083_v3f32(ptr addrspace(1) %out, <3 x float> %x) {
; CM-LABEL: issue135083_v3f32:
; CM: ; %bb.0:
-; CM-NEXT: ALU 17, @4, KC0[CB0:0-32], KC1[]
+; CM-NEXT: ALU 15, @4, KC0[CB0:0-32], KC1[]
; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T2, T3.X
; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T0.X, T1.X
; CM-NEXT: CF_END
; CM-NEXT: ALU clause starting at 4:
-; CM-NEXT: AND_INT * T0.W, KC0[3].W, literal.x,
-; CM-NEXT: 2147483647(nan), 0(0.000000e+00)
-; CM-NEXT: AND_INT T0.Z, KC0[3].Z, literal.x,
-; CM-NEXT: SETGT_INT * T0.W, PV.W, literal.y,
-; CM-NEXT: 2147483647(nan), 2139095039(3.402823e+38)
+; CM-NEXT: LSHL * T0.W, KC0[3].W, 1,
+; CM-NEXT: LSHL T0.Z, KC0[3].Z, 1,
+; CM-NEXT: SETGT_UINT * T0.W, PV.W, literal.x,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
; CM-NEXT: CNDE_INT T0.X, PV.W, 1, 0.0,
-; CM-NEXT: AND_INT T0.Y, KC0[3].Y, literal.x,
-; CM-NEXT: SETGT_INT T0.Z, PV.Z, literal.y,
-; CM-NEXT: ADD_INT * T0.W, KC0[2].Y, literal.z,
-; CM-NEXT: 2147483647(nan), 2139095039(3.402823e+38)
-; CM-NEXT: 8(1.121039e-44), 0(0.000000e+00)
+; CM-NEXT: LSHL T0.Y, KC0[3].Y, 1,
+; CM-NEXT: SETGT_UINT T0.Z, PV.Z, literal.x,
+; CM-NEXT: ADD_INT * T0.W, KC0[2].Y, literal.y,
+; CM-NEXT: -16777217(-1.701412e+38), 8(1.121039e-44)
; CM-NEXT: LSHR T1.X, PV.W, literal.x,
; CM-NEXT: CNDE_INT T2.Y, PV.Z, 1, 0.0,
-; CM-NEXT: SETGT_INT * T0.W, PV.Y, literal.y,
-; CM-NEXT: 2(2.802597e-45), 2139095039(3.402823e+38)
+; CM-NEXT: SETGT_UINT * T0.W, PV.Y, literal.y,
+; CM-NEXT: 2(2.802597e-45), -16777217(-1.701412e+38)
; CM-NEXT: CNDE_INT * T2.X, PV.W, 1, 0.0,
; CM-NEXT: LSHR * T3.X, KC0[2].Y, literal.x,
; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00)
@@ -103,27 +99,26 @@ define amdgpu_kernel void @issue135083_v3f32(ptr addrspace(1) %out, <3 x float>
define amdgpu_kernel void @issue135083_v4f32(ptr addrspace(1) %out, <4 x float> %x) {
; CM-LABEL: issue135083_v4f32:
; CM: ; %bb.0:
-; CM-NEXT: ALU 18, @4, KC0[CB0:0-32], KC1[]
+; CM-NEXT: ALU 17, @4, KC0[CB0:0-32], KC1[]
; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T0, T1.X
; CM-NEXT: CF_END
; CM-NEXT: PAD
; CM-NEXT: ALU clause starting at 4:
-; CM-NEXT: AND_INT * T0.W, KC0[4].X, literal.x,
-; CM-NEXT: 2147483647(nan), 0(0.000000e+00)
-; CM-NEXT: SETGT_INT T0.Z, PV.W, literal.x,
-; CM-NEXT: AND_INT * T0.W, KC0[3].W, literal.y,
-; CM-NEXT: 2139095039(3.402823e+38), 2147483647(nan)
-; CM-NEXT: AND_INT T0.Y, KC0[3].Z, literal.x,
-; CM-NEXT: SETGT_INT T1.Z, PV.W, literal.y,
+; CM-NEXT: LSHL * T0.W, KC0[4].X, 1,
+; CM-NEXT: SETGT_UINT T0.Z, PV.W, literal.x,
+; CM-NEXT: LSHL * T0.W, KC0[3].W, 1,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
+; CM-NEXT: LSHL T0.Y, KC0[3].Z, 1,
+; CM-NEXT: SETGT_UINT T1.Z, PV.W, literal.x,
; CM-NEXT: CNDE_INT * T0.W, PV.Z, 1, 0.0,
-; CM-NEXT: 2147483647(nan), 2139095039(3.402823e+38)
-; CM-NEXT: AND_INT T1.Y, KC0[3].Y, literal.x,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
+; CM-NEXT: LSHL T1.Y, KC0[3].Y, 1,
; CM-NEXT: CNDE_INT T0.Z, PV.Z, 1, 0.0,
-; CM-NEXT: SETGT_INT * T1.W, PV.Y, literal.y,
-; CM-NEXT: 2147483647(nan), 2139095039(3.402823e+38)
+; CM-NEXT: SETGT_UINT * T1.W, PV.Y, literal.x,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
; CM-NEXT: CNDE_INT T0.Y, PV.W, 1, 0.0,
-; CM-NEXT: SETGT_INT * T1.W, PV.Y, literal.x,
-; CM-NEXT: 2139095039(3.402823e+38), 0(0.000000e+00)
+; CM-NEXT: SETGT_UINT * T1.W, PV.Y, literal.x,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
; CM-NEXT: CNDE_INT * T0.X, PV.W, 1, 0.0,
; CM-NEXT: LSHR * T1.X, KC0[2].Y, literal.x,
; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00)
@@ -136,32 +131,31 @@ define amdgpu_kernel void @issue135083_v4f32(ptr addrspace(1) %out, <4 x float>
define amdgpu_kernel void @issue135083_v5f32(ptr addrspace(1) %out, <5 x float> %x) {
; CM-LABEL: issue135083_v5f32:
; CM: ; %bb.0:
-; CM-NEXT: ALU 23, @4, KC0[CB0:0-32], KC1[]
+; CM-NEXT: ALU 22, @4, KC0[CB0:0-32], KC1[]
; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T1, T3.X
; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T2.X, T0.X
; CM-NEXT: CF_END
; CM-NEXT: ALU clause starting at 4:
-; CM-NEXT: AND_INT T0.Z, KC0[4].Y, literal.x,
-; CM-NEXT: AND_INT * T0.W, KC0[5].X, literal.x,
-; CM-NEXT: 2147483647(nan), 0(0.000000e+00)
-; CM-NEXT: SETGT_INT T0.Y, PV.W, literal.x,
-; CM-NEXT: AND_INT T1.Z, KC0[5].Y, literal.y,
-; CM-NEXT: AND_INT * T0.W, KC0[4].W, literal.y,
-; CM-NEXT: 2139095039(3.402823e+38), 2147483647(nan)
-; CM-NEXT: SETGT_INT T0.X, PV.W, literal.x,
-; CM-NEXT: AND_INT T1.Y, KC0[4].Z, literal.y,
-; CM-NEXT: SETGT_INT T1.Z, PV.Z, literal.x,
+; CM-NEXT: LSHL T0.Z, KC0[4].Y, 1,
+; CM-NEXT: LSHL * T0.W, KC0[5].X, 1,
+; CM-NEXT: SETGT_UINT T0.Y, PV.W, literal.x,
+; CM-NEXT: LSHL T1.Z, KC0[5].Y, 1,
+; CM-NEXT: LSHL * T0.W, KC0[4].W, 1,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
+; CM-NEXT: SETGT_UINT T0.X, PV.W, literal.x,
+; CM-NEXT: LSHL T1.Y, KC0[4].Z, 1,
+; CM-NEXT: SETGT_UINT T1.Z, PV.Z, literal.x,
; CM-NEXT: CNDE_INT * T1.W, PV.Y, 1, 0.0,
-; CM-NEXT: 2139095039(3.402823e+38), 2147483647(nan)
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
; CM-NEXT: CNDE_INT T2.X, PV.Z, 1, 0.0,
-; CM-NEXT: SETGT_INT T0.Y, PV.Y, literal.x,
+; CM-NEXT: SETGT_UINT T0.Y, PV.Y, literal.x,
; CM-NEXT: CNDE_INT T1.Z, PV.X, 1, 0.0,
; CM-NEXT: ADD_INT * T0.W, KC0[2].Y, literal.y,
-; CM-NEXT: 2139095039(3.402823e+38), 16(2.242078e-44)
+; CM-NEXT: -16777217(-1.701412e+38), 16(2.242078e-44)
; CM-NEXT: LSHR T0.X, PV.W, literal.x,
; CM-NEXT: CNDE_INT T1.Y, PV.Y, 1, 0.0,
-; CM-NEXT: SETGT_INT * T0.W, T0.Z, literal.y,
-; CM-NEXT: 2(2.802597e-45), 2139095039(3.402823e+38)
+; CM-NEXT: SETGT_UINT * T0.W, T0.Z, literal.y,
+; CM-NEXT: 2(2.802597e-45), -16777217(-1.701412e+38)
; CM-NEXT: CNDE_INT * T1.X, PV.W, 1, 0.0,
; CM-NEXT: LSHR * T3.X, KC0[2].Y, literal.x,
; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00)
@@ -174,36 +168,34 @@ define amdgpu_kernel void @issue135083_v5f32(ptr addrspace(1) %out, <5 x float>
define amdgpu_kernel void @issue135083_v6f32(ptr addrspace(1) %out, <6 x float> %x) {
; CM-LABEL: issue135083_v6f32:
; CM: ; %bb.0:
-; CM-NEXT: ALU 27, @4, KC0[CB0:0-32], KC1[]
+; CM-NEXT: ALU 25, @4, KC0[CB0:0-32], KC1[]
; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T1, T3.X
; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T2, T0.X
; CM-NEXT: CF_END
; CM-NEXT: ALU clause starting at 4:
-; CM-NEXT: AND_INT T0.Z, KC0[4].Y, literal.x,
-; CM-NEXT: AND_INT * T0.W, KC0[5].Z, literal.x,
-; CM-NEXT: 2147483647(nan), 0(0.000000e+00)
-; CM-NEXT: AND_INT T1.Z, KC0[5].X, literal.x,
-; CM-NEXT: AND_INT * T1.W, KC0[4].W, literal.x,
-; CM-NEXT: 2147483647(nan), 0(0.000000e+00)
-; CM-NEXT: SETGT_INT T0.X, PV.W, literal.x,
-; CM-NEXT: SETGT_INT T0.Y, PV.Z, literal.x,
-; CM-NEXT: AND_INT T1.Z, KC0[5].Y, literal.y,
-; CM-NEXT: SETGT_INT * T0.W, T0.W, literal.x,
-; CM-NEXT: 2139095039(3.402823e+38), 2147483647(nan)
-; CM-NEXT: AND_INT T1.X, KC0[4].Z, literal.x,
+; CM-NEXT: LSHL T0.Z, KC0[4].Y, 1,
+; CM-NEXT: LSHL * T0.W, KC0[5].Z, 1,
+; CM-NEXT: LSHL T1.Z, KC0[5].X, 1,
+; CM-NEXT: LSHL * T1.W, KC0[4].W, 1,
+; CM-NEXT: SETGT_UINT T0.X, PV.W, literal.x,
+; CM-NEXT: SETGT_UINT T0.Y, PV.Z, literal.x,
+; CM-NEXT: LSHL T1.Z, KC0[5].Y, 1,
+; CM-NEXT: SETGT_UINT * T0.W, T0.W, literal.x,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
+; CM-NEXT: LSHL T1.X, KC0[4].Z, 1,
; CM-NEXT: CNDE_INT T2.Y, PV.W, 1, 0.0,
-; CM-NEXT: SETGT_INT T1.Z, PV.Z, literal.y,
+; CM-NEXT: SETGT_UINT T1.Z, PV.Z, literal.x,
; CM-NEXT: CNDE_INT * T1.W, PV.Y, 1, 0.0,
-; CM-NEXT: 2147483647(nan), 2139095039(3.402823e+38)
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
; CM-NEXT: CNDE_INT T2.X, PV.Z, 1, 0.0,
-; CM-NEXT: SETGT_INT T0.Y, PV.X, literal.x,
+; CM-NEXT: SETGT_UINT T0.Y, PV.X, literal.x,
; CM-NEXT: CNDE_INT T1.Z, T0.X, 1, 0.0,
; CM-NEXT: ADD_INT * T0.W, KC0[2].Y, literal.y,
-; CM-NEXT: 2139095039(3.402823e+38), 16(2.242078e-44)
+; CM-NEXT: -16777217(-1.701412e+38), 16(2.242078e-44)
; CM-NEXT: LSHR T0.X, PV.W, literal.x,
; CM-NEXT: CNDE_INT T1.Y, PV.Y, 1, 0.0,
-; CM-NEXT: SETGT_INT * T0.W, T0.Z, literal.y,
-; CM-NEXT: 2(2.802597e-45), 2139095039(3.402823e+38)
+; CM-NEXT: SETGT_UINT * T0.W, T0.Z, literal.y,
+; CM-NEXT: 2(2.802597e-45), -16777217(-1.701412e+38)
; CM-NEXT: CNDE_INT * T1.X, PV.W, 1, 0.0,
; CM-NEXT: LSHR * T3.X, KC0[2].Y, literal.x,
; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00)
@@ -216,45 +208,43 @@ define amdgpu_kernel void @issue135083_v6f32(ptr addrspace(1) %out, <6 x float>
define amdgpu_kernel void @issue135083_v7f32(ptr addrspace(1) %out, <7 x float> %x) {
; CM-LABEL: issue135083_v7f32:
; CM: ; %bb.0:
-; CM-NEXT: ALU 34, @6, KC0[CB0:0-32], KC1[]
+; CM-NEXT: ALU 32, @6, KC0[CB0:0-32], KC1[]
; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T4, T5.X
; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T3, T0.X
; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T1.X, T2.X
; CM-NEXT: CF_END
; CM-NEXT: PAD
; CM-NEXT: ALU clause starting at 6:
-; CM-NEXT: AND_INT T0.Z, KC0[4].Y, literal.x,
-; CM-NEXT: AND_INT * T0.W, KC0[4].W, literal.x,
-; CM-NEXT: 2147483647(nan), 0(0.000000e+00)
-; CM-NEXT: SETGT_INT T0.Y, PV.W, literal.x,
-; CM-NEXT: AND_INT T1.Z, KC0[5].W, literal.y,
-; CM-NEXT: AND_INT * T0.W, KC0[5].X, literal.y,
-; CM-NEXT: 2139095039(3.402823e+38), 2147483647(nan)
-; CM-NEXT: AND_INT T0.X, KC0[4].Z, literal.x,
-; CM-NEXT: SETGT_INT T1.Y, PV.W, literal.y,
-; CM-NEXT: AND_INT T2.Z, KC0[5].Z, literal.x,
-; CM-NEXT: SETGT_INT * T0.W, PV.Z, literal.y,
-; CM-NEXT: 2147483647(nan), 2139095039(3.402823e+38)
+; CM-NEXT: LSHL T0.Z, KC0[4].Y, 1,
+; CM-NEXT: LSHL * T0.W, KC0[4].W, 1,
+; CM-NEXT: SETGT_UINT T0.Y, PV.W, literal.x,
+; CM-NEXT: LSHL T1.Z, KC0[5].W, 1,
+; CM-NEXT: LSHL * T0.W, KC0[5].X, 1,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
+; CM-NEXT: LSHL T0.X, KC0[4].Z, 1,
+; CM-NEXT: SETGT_UINT T1.Y, PV.W, literal.x,
+; CM-NEXT: LSHL T2.Z, KC0[5].Z, 1,
+; CM-NEXT: SETGT_UINT * T0.W, PV.Z, literal.x,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
; CM-NEXT: CNDE_INT T1.X, PV.W, 1, 0.0,
-; CM-NEXT: AND_INT T2.Y, KC0[5].Y, literal.x,
-; CM-NEXT: SETGT_INT T1.Z, PV.Z, literal.y,
-; CM-NEXT: ADD_INT * T0.W, KC0[2].Y, literal.z,
-; CM-NEXT: 2147483647(nan), 2139095039(3.402823e+38)
-; CM-NEXT: 24(3.363116e-44), 0(0.000000e+00)
+; CM-NEXT: LSHL T2.Y, KC0[5].Y, 1,
+; CM-NEXT: SETGT_UINT T1.Z, PV.Z, literal.x,
+; CM-NEXT: ADD_INT * T0.W, KC0[2].Y, literal.y,
+; CM-NEXT: -16777217(-1.701412e+38), 24(3.363116e-44)
; CM-NEXT: LSHR T2.X, PV.W, literal.x,
; CM-NEXT: CNDE_INT T3.Y, PV.Z, 1, 0.0,
-; CM-NEXT: SETGT_INT T1.Z, PV.Y, literal.y,
+; CM-NEXT: SETGT_UINT T1.Z, PV.Y, literal.y,
; CM-NEXT: CNDE_INT * T4.W, T1.Y, 1, 0.0,
-; CM-NEXT: 2(2.802597e-45), 2139095039(3.402823e+38)
+; CM-NEXT: 2(2.802597e-45), -16777217(-1.701412e+38)
; CM-NEXT: CNDE_INT T3.X, PV.Z, 1, 0.0,
-; CM-NEXT: SETGT_INT T1.Y, T0.X, literal.x,
+; CM-NEXT: SETGT_UINT T1.Y, T0.X, literal.x,
; CM-NEXT: CNDE_INT T4.Z, T0.Y, 1, 0.0,
; CM-NEXT: ADD_INT * T0.W, KC0[2].Y, literal.y,
-; CM-NEXT: 2139095039(3.402823e+38), 16(2.242078e-44)
+; CM-NEXT: -16777217(-1.701412e+38), 16(2.242078e-44)
; CM-NEXT: LSHR T0.X, PV.W, literal.x,
; CM-NEXT: CNDE_INT T4.Y, PV.Y, 1, 0.0,
-; CM-NEXT: SETGT_INT * T0.W, T0.Z, literal.y,
-; CM-NEXT: 2(2.802597e-45), 2139095039(3.402823e+38)
+; CM-NEXT: SETGT_UINT * T0.W, T0.Z, literal.y,
+; CM-NEXT: 2(2.802597e-45), -16777217(-1.701412e+38)
; CM-NEXT: CNDE_INT * T4.X, PV.W, 1, 0.0,
; CM-NEXT: LSHR * T5.X, KC0[2].Y, literal.x,
; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00)
@@ -267,43 +257,42 @@ define amdgpu_kernel void @issue135083_v7f32(ptr addrspace(1) %out, <7 x float>
define amdgpu_kernel void @issue135083_v8f32(ptr addrspace(1) %out, <8 x float> %x) {
; CM-LABEL: issue135083_v8f32:
; CM: ; %bb.0:
-; CM-NEXT: ALU 34, @4, KC0[CB0:0-32], KC1[]
+; CM-NEXT: ALU 33, @4, KC0[CB0:0-32], KC1[]
; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T0, T3.X
; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T1, T2.X
; CM-NEXT: CF_END
; CM-NEXT: ALU clause starting at 4:
-; CM-NEXT: AND_INT T0.Z, KC0[6].X, literal.x,
-; CM-NEXT: AND_INT * T0.W, KC0[4].W, literal.x,
-; CM-NEXT: 2147483647(nan), 0(0.000000e+00)
-; CM-NEXT: AND_INT T0.X, KC0[4].Y, literal.x,
-; CM-NEXT: SETGT_INT T0.Y, PV.W, literal.y,
-; CM-NEXT: SETGT_INT T0.Z, PV.Z, literal.y,
-; CM-NEXT: AND_INT * T0.W, KC0[5].W, literal.x,
-; CM-NEXT: 2147483647(nan), 2139095039(3.402823e+38)
-; CM-NEXT: AND_INT T1.X, KC0[5].Z, literal.x,
-; CM-NEXT: SETGT_INT T1.Y, PV.W, literal.y,
-; CM-NEXT: AND_INT T1.Z, KC0[5].X, literal.x,
+; CM-NEXT: LSHL T0.Z, KC0[6].X, 1,
+; CM-NEXT: LSHL * T0.W, KC0[4].W, 1,
+; CM-NEXT: LSHL T0.X, KC0[4].Y, 1,
+; CM-NEXT: SETGT_UINT T0.Y, PV.W, literal.x,
+; CM-NEXT: SETGT_UINT T0.Z, PV.Z, literal.x,
+; CM-NEXT: LSHL * T0.W, KC0[5].W, 1,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
+; CM-NEXT: LSHL T1.X, KC0[5].Z, 1,
+; CM-NEXT: SETGT_UINT T1.Y, PV.W, literal.x,
+; CM-NEXT: LSHL T1.Z, KC0[5].X, 1,
; CM-NEXT: CNDE_INT * T1.W, PV.Z, 1, 0.0,
-; CM-NEXT: 2147483647(nan), 2139095039(3.402823e+38)
-; CM-NEXT: SETGT_INT T2.X, PV.Z, literal.x,
-; CM-NEXT: AND_INT T2.Y, KC0[5].Y, literal.y,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
+; CM-NEXT: SETGT_UINT T2.X, PV.Z, literal.x,
+; CM-NEXT: LSHL T2.Y, KC0[5].Y, 1,
; CM-NEXT: CNDE_INT T1.Z, PV.Y, 1, 0.0,
-; CM-NEXT: SETGT_INT * T0.W, PV.X, literal.x,
-; CM-NEXT: 2139095039(3.402823e+38), 2147483647(nan)
-; CM-NEXT: AND_INT T3.X, KC0[4].Z, literal.x,
+; CM-NEXT: SETGT_UINT * T0.W, PV.X, literal.x,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
+; CM-NEXT: LSHL T3.X, KC0[4].Z, 1,
; CM-NEXT: CNDE_INT T1.Y, PV.W, 1, 0.0,
-; CM-NEXT: SETGT_INT T0.Z, PV.Y, literal.y,
+; CM-NEXT: SETGT_UINT T0.Z, PV.Y, literal.x,
; CM-NEXT: CNDE_INT * T0.W, PV.X, 1, 0.0,
-; CM-NEXT: 2147483647(nan), 2139095039(3.402823e+38)
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
; CM-NEXT: CNDE_INT T1.X, PV.Z, 1, 0.0,
-; CM-NEXT: SETGT_INT T2.Y, PV.X, literal.x,
+; CM-NEXT: SETGT_UINT T2.Y, PV.X, literal.x,
; CM-NEXT: CNDE_INT T0.Z, T0.Y, 1, 0.0,
; CM-NEXT: ADD_INT * T2.W, KC0[2].Y, literal.y,
-; CM-NEXT: 2139095039(3.402823e+38), 16(2.242078e-44)
+; CM-NEXT: -16777217(-1.701412e+38), 16(2.242078e-44)
; CM-NEXT: LSHR T2.X, PV.W, literal.x,
; CM-NEXT: CNDE_INT T0.Y, PV.Y, 1, 0.0,
-; CM-NEXT: SETGT_INT * T2.W, T0.X, literal.y,
-; CM-NEXT: 2(2.802597e-45), 2139095039(3.402823e+38)
+; CM-NEXT: SETGT_UINT * T2.W, T0.X, literal.y,
+; CM-NEXT: 2(2.802597e-45), -16777217(-1.701412e+38)
; CM-NEXT: CNDE_INT * T0.X, PV.W, 1, 0.0,
; CM-NEXT: LSHR * T3.X, KC0[2].Y, literal.x,
; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00)
@@ -316,80 +305,79 @@ define amdgpu_kernel void @issue135083_v8f32(ptr addrspace(1) %out, <8 x float>
define amdgpu_kernel void @issue135083_v16f32(ptr addrspace(1) %out, <16 x float> %x) {
; CM-LABEL: issue135083_v16f32:
; CM: ; %bb.0:
-; CM-NEXT: ALU 69, @6, KC0[CB0:0-32], KC1[]
+; CM-NEXT: ALU 68, @6, KC0[CB0:0-32], KC1[]
; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T1, T7.X
; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T2, T0.X
; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T3, T6.X
; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T4, T5.X
; CM-NEXT: CF_END
; CM-NEXT: ALU clause starting at 6:
-; CM-NEXT: AND_INT T0.Z, KC0[6].Y, literal.x,
-; CM-NEXT: AND_INT * T0.W, KC0[6].W, literal.x,
-; CM-NEXT: 2147483647(nan), 0(0.000000e+00)
-; CM-NEXT: SETGT_INT T0.X, PV.W, literal.x,
-; CM-NEXT: AND_INT T0.Y, KC0[6].Z, literal.y,
-; CM-NEXT: AND_INT T1.Z, KC0[7].Y, literal.y,
-; CM-NEXT: AND_INT * T0.W, KC0[7].X, literal.y,
-; CM-NEXT: 2139095039(3.402823e+38), 2147483647(nan)
-; CM-NEXT: SETGT_INT T1.X, PV.W, literal.x,
-; CM-NEXT: AND_INT T1.Y, KC0[7].Z, literal.y,
-; CM-NEXT: AND_INT T2.Z, KC0[8].X, literal.y,
-; CM-NEXT: AND_INT * T0.W, KC0[7].W, literal.y,
-; CM-NEXT: 2139095039(3.402823e+38), 2147483647(nan)
-; CM-NEXT: SETGT_INT T2.X, PV.W, literal.x,
-; CM-NEXT: SETGT_INT T2.Y, PV.Z, literal.x,
-; CM-NEXT: AND_INT T2.Z, KC0[10].X, literal.y,
-; CM-NEXT: AND_INT * T0.W, KC0[8].W, literal.y,
-; CM-NEXT: 2139095039(3.402823e+38), 2147483647(nan)
-; CM-NEXT: AND_INT T3.X, KC0[8].Y, literal.x,
-; CM-NEXT: SETGT_INT T3.Y, PV.W, literal.y,
-; CM-NEXT: SETGT_INT T2.Z, PV.Z, literal.y,
-; CM-NEXT: AND_INT * T0.W, KC0[9].W, literal.x,
-; CM-NEXT: 2147483647(nan), 2139095039(3.402823e+38)
-; CM-NEXT: AND_INT T4.X, KC0[9].Z, literal.x,
-; CM-NEXT: SETGT_INT T4.Y, PV.W, literal.y,
-; CM-NEXT: AND_INT T3.Z, KC0[9].X, literal.x,
+; CM-NEXT: LSHL T0.Z, KC0[6].Y, 1,
+; CM-NEXT: LSHL * T0.W, KC0[6].W, 1,
+; CM-NEXT: SETGT_UINT T0.X, PV.W, literal.x,
+; CM-NEXT: LSHL T0.Y, KC0[6].Z, 1,
+; CM-NEXT: LSHL T1.Z, KC0[7].Y, 1,
+; CM-NEXT: LSHL * T0.W, KC0[7].X, 1,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
+; CM-NEXT: SETGT_UINT T1.X, PV.W, literal.x,
+; CM-NEXT: LSHL T1.Y, KC0[7].Z, 1,
+; CM-NEXT: LSHL T2.Z, KC0[8].X, 1,
+; CM-NEXT: LSHL * T0.W, KC0[7].W, 1,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
+; CM-NEXT: SETGT_UINT T2.X, PV.W, literal.x,
+; CM-NEXT: SETGT_UINT T2.Y, PV.Z, literal.x,
+; CM-NEXT: LSHL T2.Z, KC0[10].X, 1,
+; CM-NEXT: LSHL * T0.W, KC0[8].W, 1,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
+; CM-NEXT: LSHL T3.X, KC0[8].Y, 1,
+; CM-NEXT: SETGT_UINT T3.Y, PV.W, literal.x,
+; CM-NEXT: SETGT_UINT T2.Z, PV.Z, literal.x,
+; CM-NEXT: LSHL * T0.W, KC0[9].W, 1,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
+; CM-NEXT: LSHL T4.X, KC0[9].Z, 1,
+; CM-NEXT: SETGT_UINT T4.Y, PV.W, literal.x,
+; CM-NEXT: LSHL T3.Z, KC0[9].X, 1,
; CM-NEXT: CNDE_INT * T4.W, PV.Z, 1, 0.0,
-; CM-NEXT: 2147483647(nan), 2139095039(3.402823e+38)
-; CM-NEXT: SETGT_INT T5.X, PV.Z, literal.x,
-; CM-NEXT: AND_INT T5.Y, KC0[9].Y, literal.y,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
+; CM-NEXT: SETGT_UINT T5.X, PV.Z, literal.x,
+; CM-NEXT: LSHL T5.Y, KC0[9].Y, 1,
; CM-NEXT: CNDE_INT T4.Z, PV.Y, 1, 0.0,
-; CM-NEXT: SETGT_INT * T0.W, PV.X, literal.x,
-; CM-NEXT: 2139095039(3.402823e+38), 2147483647(nan)
-; CM-NEXT: AND_INT T6.X, KC0[8].Z, literal.x,
+; CM-NEXT: SETGT_UINT * T0.W, PV.X, literal.x,
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
+; CM-NEXT: LSHL T6.X, KC0[8].Z, 1,
; CM-NEXT: CNDE_INT T4.Y, PV.W, 1, 0.0,
-; CM-NEXT: SETGT_INT T2.Z, PV.Y, literal.y,
+; CM-NEXT: SETGT_UINT T2.Z, PV.Y, literal.x,
; CM-NEXT: CNDE_INT * T3.W, PV.X, 1, 0.0,
-; CM-NEXT: 2147483647(nan), 2139095039(3.402823e+38)
+; CM-NEXT: -16777217(-1.701412e+38), 0(0.000000e+00)
; CM-NEXT: CNDE_INT T4.X, PV.Z, 1, 0.0,
-; CM-NEXT: SETGT_INT T5.Y, PV.X, literal.x,
+; CM-NEXT: SETGT_UINT T5.Y, PV.X, literal.x,
; CM-NEXT: CNDE_INT T3.Z, T3.Y, 1, 0.0,
; CM-NEXT: ADD_INT * T0.W, KC0[2].Y, literal.y,
-; CM-NEXT: 2139095039(3.402823e+38), 48(6.726233e-44)
+; CM-NEXT: -16777217(-1.701412e+38), 48(6.726233e-44)
; CM-NEXT: LSHR T5.X, PV.W, literal.x,
; CM-NEXT: CNDE_INT T3.Y, PV.Y, 1, 0.0,
-; CM-NEXT: SETGT_INT T2.Z, T3.X, literal.y,
+; CM-NEXT: SETGT_UINT T2.Z, T3.X, literal.y,
; CM-NEXT: CNDE_INT * T2.W, T2.Y, 1, 0.0,
-; CM-NEXT: 2(2.802597e-45), 2139095039(3.402823e+38)
+; CM-NEXT: 2(2.802597e-45), -16777217(-1.701412e+38)
; CM-NEXT: CNDE_INT T3.X, PV.Z, 1, 0.0,
-; CM-NEXT: SETGT_INT T1.Y, T1.Y, literal.x,
+; CM-NEXT: SETGT_UINT T1.Y, T1.Y, literal.x,
; CM-NEXT: CNDE_INT T2.Z, T2.X, 1, 0.0,
; CM-NEXT: ADD_INT * T0.W, KC0[2].Y, literal.y,
-; CM-NEXT: 2139095039(3.402823e+38), 32(4.484155e-44)
+; CM-NEXT: -16777217(-1.701412e+38), 32(4.484155e-44)
; CM-NEXT: LSHR T6.X, PV.W, literal.x,
; CM-NEXT: CNDE_INT T2.Y, PV.Y, 1, 0.0,
-; CM-NEXT: SETGT_INT T1.Z, T1.Z, literal.y,
+; CM-NEXT: SETGT_UINT T1.Z, T1.Z, literal.y,
; CM-NEXT: CNDE_INT * T1.W, T1.X, 1, 0.0,
-; CM-NEXT: 2(2.802597e-45), 2139095039(3.402823e+38)
+; CM-NEXT: 2(2.802597e-45), -16777217(-1.701412e+38)
; CM-NEXT: CNDE_INT T2.X, PV.Z, 1, 0.0,
-; CM-NEXT: SETGT_INT T0.Y, T0.Y, literal.x,
+; CM-NEXT: SETGT_UINT T0.Y, T0.Y, literal.x,
; CM-NEXT: CNDE_INT T1.Z, T0.X, 1, 0.0,
; CM-NEXT: ADD_INT * T0.W, KC0[2].Y, literal.y,
-; CM-NEXT: 2139095039(3.402823e+38), 16(2.242078e-44)
+; CM-NEXT: -16777217(-1.701412e+38), 16(2.242078e-44)
; CM-NEXT: LSHR T0.X, PV.W, literal.x,
; CM-NEXT: CNDE_INT T1.Y, PV.Y, 1, 0.0,
-; CM-NEXT: SETGT_INT * T0.W, T0.Z, literal.y,
-; CM-NEXT: 2(2.802597e-45), 2139095039(3.402823e+38)
+; CM-NEXT: SETGT_UINT * T0.W, T0.Z, literal.y,
+; CM-NEXT: 2(2.802597e-45), -16777217(-1.701412e+38)
; CM-NEXT: CNDE_INT * T1.X, PV.W, 1, 0.0,
; CM-NEXT: LSHR * T7.X, KC0[2].Y, literal.x,
; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00)
diff --git a/llvm/test/CodeGen/PowerPC/fp-classify.ll b/llvm/test/CodeGen/PowerPC/fp-classify.ll
index dc9853ff2e301..22b098dfc0df0 100644
--- a/llvm/test/CodeGen/PowerPC/fp-classify.ll
+++ b/llvm/test/CodeGen/PowerPC/fp-classify.ll
@@ -88,11 +88,9 @@ define zeroext i1 @abs_isinfornanf(float %x) {
; P8-LABEL: abs_isinfornanf:
; P8: # %bb.0: # %entry
; P8-NEXT: xscvdpspn 0, 1
-; P8-NEXT: lis 4, 32639
-; P8-NEXT: ori 4, 4, 65535
; P8-NEXT: mffprwz 3, 0
-; P8-NEXT: clrlwi 3, 3, 1
-; P8-NEXT: sub 3, 4, 3
+; P8-NEXT: rlwinm 3, 3, 9, 24, 31
+; P8-NEXT: subfic 3, 3, 254
; P8-NEXT: rldicl 3, 3, 1, 63
; P8-NEXT: blr
;
@@ -113,14 +111,11 @@ define zeroext i1 @abs_isinfornan(double %x) {
; P8-LABEL: abs_isinfornan:
; P8: # %bb.0: # %entry
; P8-NEXT: mffprd 3, 1
-; P8-NEXT: li 4, -33
-; P8-NEXT: rldicl 4, 4, 47, 1
-; P8-NEXT: sradi 5, 4, 63
-; P8-NEXT: clrldi 3, 3, 1
-; P8-NEXT: rldicl 6, 3, 1, 63
-; P8-NEXT: subc 3, 4, 3
-; P8-NEXT: adde 3, 6, 5
-; P8-NEXT: xori 3, 3, 1
+; P8-NEXT: li 4, 2046
+; P8-NEXT: rldicl 3, 3, 12, 53
+; P8-NEXT: subfic 3, 3, 2046
+; P8-NEXT: subfe 3, 4, 4
+; P8-NEXT: neg 3, 3
; P8-NEXT: blr
;
; P9-LABEL: abs_isinfornan:
@@ -141,16 +136,13 @@ define zeroext i1 @abs_isinfornanq(fp128 %x) {
; P8: # %bb.0: # %entry
; P8-NEXT: xxswapd 0, 34
; P8-NEXT: addi 3, 1, -16
-; P8-NEXT: li 4, -3
+; P8-NEXT: li 4, 32766
; P8-NEXT: stxvd2x 0, 0, 3
-; P8-NEXT: rldicl 4, 4, 47, 1
-; P8-NEXT: ld 3, -8(1)
-; P8-NEXT: sradi 5, 4, 63
-; P8-NEXT: clrldi 3, 3, 1
-; P8-NEXT: rldicl 6, 3, 1, 63
-; P8-NEXT: subc 3, 4, 3
-; P8-NEXT: adde 3, 6, 5
-; P8-NEXT: xori 3, 3, 1
+; P8-NEXT: lhz 3, -2(1)
+; P8-NEXT: clrldi 3, 3, 49
+; P8-NEXT: subfic 3, 3, 32766
+; P8-NEXT: subfe 3, 4, 4
+; P8-NEXT: neg 3, 3
; P8-NEXT: blr
;
; P9-LABEL: abs_isinfornanq:
diff --git a/llvm/test/CodeGen/RISCV/float-intrinsics.ll b/llvm/test/CodeGen/RISCV/float-intrinsics.ll
index 069e20da7b908..51d4921c61c95 100644
--- a/llvm/test/CodeGen/RISCV/float-intrinsics.ll
+++ b/llvm/test/CodeGen/RISCV/float-intrinsics.ll
@@ -2224,17 +2224,15 @@ define i1 @isfinite_fpclass(float %x) {
; RV32I-LABEL: isfinite_fpclass:
; RV32I: # %bb.0:
; RV32I-NEXT: slli a0, a0, 1
-; RV32I-NEXT: srli a0, a0, 1
-; RV32I-NEXT: lui a1, 522240
-; RV32I-NEXT: slt a0, a0, a1
+; RV32I-NEXT: srli a0, a0, 24
+; RV32I-NEXT: sltiu a0, a0, 255
; RV32I-NEXT: ret
;
; RV64I-LABEL: isfinite_fpclass:
; RV64I: # %bb.0:
; RV64I-NEXT: slli a0, a0, 33
-; RV64I-NEXT: srli a0, a0, 33
-; RV64I-NEXT: lui a1, 522240
-; RV64I-NEXT: slt a0, a0, a1
+; RV64I-NEXT: srli a0, a0, 56
+; RV64I-NEXT: sltiu a0, a0, 255
; RV64I-NEXT: ret
%1 = call i1 @llvm.is.fpclass.f32(float %x, i32 504) ; 0x1f8 = "finite"
ret i1 %1
@@ -2389,19 +2387,17 @@ define i1 @isnotfinite_fpclass(float %x) {
; RV32I-LABEL: isnotfinite_fpclass:
; RV32I: # %bb.0:
; RV32I-NEXT: slli a0, a0, 1
-; RV32I-NEXT: lui a1, 522240
-; RV32I-NEXT: srli a0, a0, 1
-; RV32I-NEXT: addi a1, a1, -1
-; RV32I-NEXT: slt a0, a1, a0
+; RV32I-NEXT: srli a0, a0, 24
+; RV32I-NEXT: sltiu a0, a0, 255
+; RV32I-NEXT: xori a0, a0, 1
; RV32I-NEXT: ret
;
; RV64I-LABEL: isnotfinite_fpclass:
; RV64I: # %bb.0:
; RV64I-NEXT: slli a0, a0, 33
-; RV64I-NEXT: lui a1, 522240
-; RV64I-NEXT: srli a0, a0, 33
-; RV64I-NEXT: addi a1, a1, -1
-; RV64I-NEXT: slt a0, a1, a0
+; RV64I-NEXT: srli a0, a0, 56
+; RV64I-NEXT: sltiu a0, a0, 255
+; RV64I-NEXT: xori a0, a0, 1
; RV64I-NEXT: ret
%1 = call i1 @llvm.is.fpclass.f32(float %x, i32 519) ; ox207 = "inf|nan"
ret i1 %1
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vfclass.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vfclass.ll
index de3ed5fbaa030..7648c89b9c075 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vfclass.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vfclass.ll
@@ -148,12 +148,10 @@ define <8 x i1> @iszero_v8bf16(<8 x bfloat> %x) {
define <8 x i1> @isfinite_v8bf16(<8 x bfloat> %x) {
; ZVFBFMIN-LABEL: isfinite_v8bf16:
; ZVFBFMIN: # %bb.0:
-; ZVFBFMIN-NEXT: lui a0, 8
-; ZVFBFMIN-NEXT: addi a1, a0, -1
; ZVFBFMIN-NEXT: vsetivli zero, 8, e16, m1, ta, ma
-; ZVFBFMIN-NEXT: vand.vx v8, v8, a1
-; ZVFBFMIN-NEXT: addi a0, a0, -128
-; ZVFBFMIN-NEXT: vmslt.vx v0, v8, a0
+; ZVFBFMIN-NEXT: vadd.vv v8, v8, v8
+; ZVFBFMIN-NEXT: li a0, -256
+; ZVFBFMIN-NEXT: vmsltu.vx v0, v8, a0
; ZVFBFMIN-NEXT: ret
;
; ZVFBFA-LABEL: isfinite_v8bf16:
diff --git a/llvm/test/CodeGen/RISCV/rvv/vfclass-sdnode.ll b/llvm/test/CodeGen/RISCV/rvv/vfclass-sdnode.ll
index 7828e713c02cc..6905b52612c01 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vfclass-sdnode.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vfclass-sdnode.ll
@@ -13,10 +13,10 @@
; RUN: --check-prefixes=CHECK,ZVFBFMIN,ZVFHMIN
; RUN: llc -mtriple=riscv32 -mattr=+d,+zfhmin,+zvfhmin,+experimental-zvfbfa,+v \
; RUN: -target-abi=ilp32d -verify-machineinstrs < %s | FileCheck %s \
-; RUN: --check-prefixes=CHECK,ZVFBFA,ZVFHMIN
+; RUN: --check-prefixes=CHECK,ZVFBFA
; RUN: llc -mtriple=riscv64 -mattr=+d,+zfhmin,+zvfhmin,+experimental-zvfbfa,+v \
; RUN: -target-abi=lp64d -verify-machineinstrs < %s | FileCheck %s \
-; RUN: --check-prefixes=CHECK,ZVFBFA,ZVFHMIN
+; RUN: --check-prefixes=CHECK,ZVFBFA
define <vscale x 2 x i1> @isnan_nxv2bf16(<vscale x 2 x bfloat> %x) {
; ZVFBFMIN-LABEL: isnan_nxv2bf16:
@@ -158,12 +158,10 @@ define <vscale x 8 x i1> @iszero_nxv8bf16(<vscale x 8 x bfloat> %x) {
define <vscale x 8 x i1> @isfinite_nxv8bf16(<vscale x 8 x bfloat> %x) {
; ZVFBFMIN-LABEL: isfinite_nxv8bf16:
; ZVFBFMIN: # %bb.0:
-; ZVFBFMIN-NEXT: lui a0, 8
-; ZVFBFMIN-NEXT: addi a1, a0, -1
-; ZVFBFMIN-NEXT: vsetvli a2, zero, e16, m2, ta, ma
-; ZVFBFMIN-NEXT: vand.vx v8, v8, a1
-; ZVFBFMIN-NEXT: addi a0, a0, -128
-; ZVFBFMIN-NEXT: vmslt.vx v0, v8, a0
+; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m2, ta, ma
+; ZVFBFMIN-NEXT: vadd.vv v8, v8, v8
+; ZVFBFMIN-NEXT: li a0, -256
+; ZVFBFMIN-NEXT: vmsltu.vx v0, v8, a0
; ZVFBFMIN-NEXT: ret
;
; ZVFBFA-LABEL: isfinite_nxv8bf16:
@@ -236,6 +234,17 @@ define <vscale x 2 x i1> @isnan_nxv2f16(<vscale x 2 x half> %x) {
; ZVFHMIN-NEXT: slli a0, a0, 10
; ZVFHMIN-NEXT: vmsgt.vx v0, v8, a0
; ZVFHMIN-NEXT: ret
+;
+; ZVFBFA-LABEL: isnan_nxv2f16:
+; ZVFBFA: # %bb.0:
+; ZVFBFA-NEXT: lui a0, 8
+; ZVFBFA-NEXT: addi a0, a0, -1
+; ZVFBFA-NEXT: vsetvli a1, zero, e16, mf2, ta, ma
+; ZVFBFA-NEXT: vand.vx v8, v8, a0
+; ZVFBFA-NEXT: li a0, 31
+; ZVFBFA-NEXT: slli a0, a0, 10
+; ZVFBFA-NEXT: vmsgt.vx v0, v8, a0
+; ZVFBFA-NEXT: ret
%1 = call <vscale x 2 x i1> @llvm.is.fpclass.nxv2f16(<vscale x 2 x half> %x, i32 3) ; nan
ret <vscale x 2 x i1> %1
}
diff --git a/llvm/test/CodeGen/X86/is_fpclass.ll b/llvm/test/CodeGen/X86/is_fpclass.ll
index 70062b245f36d..9b9732f433ace 100644
--- a/llvm/test/CodeGen/X86/is_fpclass.ll
+++ b/llvm/test/CodeGen/X86/is_fpclass.ll
@@ -240,18 +240,18 @@ entry:
define i1 @isfinite_f(float %x) {
; X86-LABEL: isfinite_f:
; X86: # %bb.0: # %entry
-; X86-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF
-; X86-NEXT: andl {{[0-9]+}}(%esp), %eax
-; X86-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X86-NEXT: setl %al
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: addl %eax, %eax
+; X86-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X86-NEXT: setb %al
; X86-NEXT: retl
;
; X64-LABEL: isfinite_f:
; X64: # %bb.0: # %entry
; X64-NEXT: movd %xmm0, %eax
-; X64-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
-; X64-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X64-NEXT: setl %al
+; X64-NEXT: addl %eax, %eax
+; X64-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X64-NEXT: setb %al
; X64-NEXT: retq
entry:
%0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 504) ; 0x1f8 = "finite"
@@ -261,18 +261,18 @@ entry:
define i1 @not_isfinite_f(float %x) {
; X86-LABEL: not_isfinite_f:
; X86: # %bb.0: # %entry
-; X86-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF
-; X86-NEXT: andl {{[0-9]+}}(%esp), %eax
-; X86-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X86-NEXT: setge %al
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: addl %eax, %eax
+; X86-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X86-NEXT: setae %al
; X86-NEXT: retl
;
; X64-LABEL: not_isfinite_f:
; X64: # %bb.0: # %entry
; X64-NEXT: movd %xmm0, %eax
-; X64-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
-; X64-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X64-NEXT: setge %al
+; X64-NEXT: addl %eax, %eax
+; X64-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X64-NEXT: setae %al
; X64-NEXT: retq
entry:
%0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 519) ; ~0x1f8 = "~finite"
@@ -1017,18 +1017,18 @@ entry:
define i1 @isfinite_f_strictfp(float %x) strictfp {
; X86-LABEL: isfinite_f_strictfp:
; X86: # %bb.0: # %entry
-; X86-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF
-; X86-NEXT: andl {{[0-9]+}}(%esp), %eax
-; X86-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X86-NEXT: setl %al
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: addl %eax, %eax
+; X86-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X86-NEXT: setb %al
; X86-NEXT: retl
;
; X64-LABEL: isfinite_f_strictfp:
; X64: # %bb.0: # %entry
; X64-NEXT: movd %xmm0, %eax
-; X64-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
-; X64-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X64-NEXT: setl %al
+; X64-NEXT: addl %eax, %eax
+; X64-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X64-NEXT: setb %al
; X64-NEXT: retq
entry:
%0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 504) strictfp ; 0x1f8 = "finite"
@@ -1038,18 +1038,18 @@ entry:
define i1 @not_isfinite_f_strictfp(float %x) strictfp {
; X86-LABEL: not_isfinite_f_strictfp:
; X86: # %bb.0: # %entry
-; X86-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF
-; X86-NEXT: andl {{[0-9]+}}(%esp), %eax
-; X86-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X86-NEXT: setge %al
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: addl %eax, %eax
+; X86-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X86-NEXT: setae %al
; X86-NEXT: retl
;
; X64-LABEL: not_isfinite_f_strictfp:
; X64: # %bb.0: # %entry
; X64-NEXT: movd %xmm0, %eax
-; X64-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
-; X64-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X64-NEXT: setge %al
+; X64-NEXT: addl %eax, %eax
+; X64-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X64-NEXT: setae %al
; X64-NEXT: retq
entry:
%0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 519) strictfp ; ~0x1f8 = ~"finite"
@@ -1150,31 +1150,21 @@ entry:
define i1 @isfinite_d(double %x) {
; X86-LABEL: isfinite_d:
; X86: # %bb.0: # %entry
-; X86-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF
-; X86-NEXT: andl {{[0-9]+}}(%esp), %eax
-; X86-NEXT: cmpl $2146435072, %eax # imm = 0x7FF00000
-; X86-NEXT: setl %al
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: shrl $20, %eax
+; X86-NEXT: andl $2047, %eax # imm = 0x7FF
+; X86-NEXT: cmpl $2047, %eax # imm = 0x7FF
+; X86-NEXT: setb %al
; X86-NEXT: retl
;
-; X64-GENERIC-LABEL: isfinite_d:
-; X64-GENERIC: # %bb.0: # %entry
-; X64-GENERIC-NEXT: movq %xmm0, %rax
-; X64-GENERIC-NEXT: movabsq $9223372036854775807, %rcx # imm = 0x7FFFFFFFFFFFFFFF
-; X64-GENERIC-NEXT: andq %rax, %rcx
-; X64-GENERIC-NEXT: movabsq $9218868437227405312, %rax # imm = 0x7FF0000000000000
-; X64-GENERIC-NEXT: cmpq %rax, %rcx
-; X64-GENERIC-NEXT: setl %al
-; X64-GENERIC-NEXT: retq
-;
-; X64-NDD-LABEL: isfinite_d:
-; X64-NDD: # %bb.0: # %entry
-; X64-NDD-NEXT: movq %xmm0, %rax
-; X64-NDD-NEXT: movabsq $9223372036854775807, %rcx # imm = 0x7FFFFFFFFFFFFFFF
-; X64-NDD-NEXT: andq %rcx, %rax
-; X64-NDD-NEXT: movabsq $9218868437227405312, %rcx # imm = 0x7FF0000000000000
-; X64-NDD-NEXT: cmpq %rcx, %rax
-; X64-NDD-NEXT: setl %al
-; X64-NDD-NEXT: retq
+; X64-LABEL: isfinite_d:
+; X64: # %bb.0: # %entry
+; X64-NEXT: movq %xmm0, %rax
+; X64-NEXT: shrq $52, %rax
+; X64-NEXT: andl $2047, %eax # imm = 0x7FF
+; X64-NEXT: cmpl $2047, %eax # imm = 0x7FF
+; X64-NEXT: setb %al
+; X64-NEXT: retq
entry:
%0 = tail call i1 @llvm.is.fpclass.f64(double %x, i32 504) ; 0x1f8 = "finite"
ret i1 %0
@@ -2032,18 +2022,18 @@ entry:
define i1 @isinf_or_nan_f(float %x) {
; X86-LABEL: isinf_or_nan_f:
; X86: # %bb.0: # %entry
-; X86-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF
-; X86-NEXT: andl {{[0-9]+}}(%esp), %eax
-; X86-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X86-NEXT: setge %al
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: addl %eax, %eax
+; X86-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X86-NEXT: setae %al
; X86-NEXT: retl
;
; X64-LABEL: isinf_or_nan_f:
; X64: # %bb.0: # %entry
; X64-NEXT: movd %xmm0, %eax
-; X64-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
-; X64-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X64-NEXT: setge %al
+; X64-NEXT: addl %eax, %eax
+; X64-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X64-NEXT: setae %al
; X64-NEXT: retq
entry:
%0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 519) ; 0x204|0x3 = "inf|nan"
@@ -2053,18 +2043,18 @@ entry:
define i1 @not_isinf_or_nan_f(float %x) {
; X86-LABEL: not_isinf_or_nan_f:
; X86: # %bb.0: # %entry
-; X86-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF
-; X86-NEXT: andl {{[0-9]+}}(%esp), %eax
-; X86-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X86-NEXT: setl %al
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: addl %eax, %eax
+; X86-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X86-NEXT: setb %al
; X86-NEXT: retl
;
; X64-LABEL: not_isinf_or_nan_f:
; X64: # %bb.0: # %entry
; X64-NEXT: movd %xmm0, %eax
-; X64-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
-; X64-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X64-NEXT: setl %al
+; X64-NEXT: addl %eax, %eax
+; X64-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X64-NEXT: setb %al
; X64-NEXT: retq
entry:
%0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 504) ; ~(0x204|0x3) = "~(inf|nan)"
@@ -2277,10 +2267,11 @@ define i1 @not_is_plus_inf_or_snan_f(float %x) {
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: cmpl $-8388608, %eax # imm = 0xFF800000
; X86-NEXT: sete %cl
-; X86-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
-; X86-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X86-NEXT: setl %dl
+; X86-NEXT: leal (%eax,%eax), %edx
+; X86-NEXT: cmpl $-16777216, %edx # imm = 0xFF000000
+; X86-NEXT: setb %dl
; X86-NEXT: orb %cl, %dl
+; X86-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
; X86-NEXT: cmpl $2143289344, %eax # imm = 0x7FC00000
; X86-NEXT: setge %al
; X86-NEXT: orb %dl, %al
@@ -2291,10 +2282,11 @@ define i1 @not_is_plus_inf_or_snan_f(float %x) {
; X64-GENERIC-NEXT: movd %xmm0, %eax
; X64-GENERIC-NEXT: cmpl $-8388608, %eax # imm = 0xFF800000
; X64-GENERIC-NEXT: sete %cl
-; X64-GENERIC-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
-; X64-GENERIC-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X64-GENERIC-NEXT: setl %dl
+; X64-GENERIC-NEXT: leal (%rax,%rax), %edx
+; X64-GENERIC-NEXT: cmpl $-16777216, %edx # imm = 0xFF000000
+; X64-GENERIC-NEXT: setb %dl
; X64-GENERIC-NEXT: orb %cl, %dl
+; X64-GENERIC-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
; X64-GENERIC-NEXT: cmpl $2143289344, %eax # imm = 0x7FC00000
; X64-GENERIC-NEXT: setge %al
; X64-GENERIC-NEXT: orb %dl, %al
@@ -2305,10 +2297,11 @@ define i1 @not_is_plus_inf_or_snan_f(float %x) {
; X64-NDD-NEXT: movd %xmm0, %eax
; X64-NDD-NEXT: cmpl $-8388608, %eax # imm = 0xFF800000
; X64-NDD-NEXT: sete %cl
-; X64-NDD-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
-; X64-NDD-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X64-NDD-NEXT: setl %dl
+; X64-NDD-NEXT: leal (%rax,%rax), %edx
+; X64-NDD-NEXT: cmpl $-16777216, %edx # imm = 0xFF000000
+; X64-NDD-NEXT: setb %dl
; X64-NDD-NEXT: orb %dl, %cl
+; X64-NDD-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
; X64-NDD-NEXT: cmpl $2143289344, %eax # imm = 0x7FC00000
; X64-NDD-NEXT: setge %al
; X64-NDD-NEXT: orb %cl, %al
@@ -2326,14 +2319,15 @@ define i1 @not_is_plus_inf_or_qnan_f(float %x) {
; X86-NEXT: cmpl $2143289344, %ecx # imm = 0x7FC00000
; X86-NEXT: setl %dl
; X86-NEXT: cmpl $2139095041, %ecx # imm = 0x7F800001
-; X86-NEXT: setge %dh
-; X86-NEXT: andb %dl, %dh
+; X86-NEXT: setge %cl
+; X86-NEXT: andb %dl, %cl
; X86-NEXT: cmpl $-8388608, %eax # imm = 0xFF800000
; X86-NEXT: sete %dl
-; X86-NEXT: cmpl $2139095040, %ecx # imm = 0x7F800000
-; X86-NEXT: setl %al
+; X86-NEXT: addl %eax, %eax
+; X86-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X86-NEXT: setb %al
; X86-NEXT: orb %dl, %al
-; X86-NEXT: orb %dh, %al
+; X86-NEXT: orb %cl, %al
; X86-NEXT: retl
;
; X64-GENERIC-LABEL: not_is_plus_inf_or_qnan_f:
@@ -2344,14 +2338,15 @@ define i1 @not_is_plus_inf_or_qnan_f(float %x) {
; X64-GENERIC-NEXT: cmpl $2143289344, %ecx # imm = 0x7FC00000
; X64-GENERIC-NEXT: setl %dl
; X64-GENERIC-NEXT: cmpl $2139095041, %ecx # imm = 0x7F800001
-; X64-GENERIC-NEXT: setge %sil
-; X64-GENERIC-NEXT: andb %dl, %sil
+; X64-GENERIC-NEXT: setge %cl
+; X64-GENERIC-NEXT: andb %dl, %cl
; X64-GENERIC-NEXT: cmpl $-8388608, %eax # imm = 0xFF800000
; X64-GENERIC-NEXT: sete %dl
-; X64-GENERIC-NEXT: cmpl $2139095040, %ecx # imm = 0x7F800000
-; X64-GENERIC-NEXT: setl %al
+; X64-GENERIC-NEXT: addl %eax, %eax
+; X64-GENERIC-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X64-GENERIC-NEXT: setb %al
; X64-GENERIC-NEXT: orb %dl, %al
-; X64-GENERIC-NEXT: orb %sil, %al
+; X64-GENERIC-NEXT: orb %cl, %al
; X64-GENERIC-NEXT: retq
;
; X64-NDD-LABEL: not_is_plus_inf_or_qnan_f:
@@ -2361,14 +2356,15 @@ define i1 @not_is_plus_inf_or_qnan_f(float %x) {
; X64-NDD-NEXT: cmpl $2143289344, %ecx # imm = 0x7FC00000
; X64-NDD-NEXT: setl %dl
; X64-NDD-NEXT: cmpl $2139095041, %ecx # imm = 0x7F800001
-; X64-NDD-NEXT: setge %sil
-; X64-NDD-NEXT: andb %sil, %dl
+; X64-NDD-NEXT: setge %cl
+; X64-NDD-NEXT: andb %dl, %cl
; X64-NDD-NEXT: cmpl $-8388608, %eax # imm = 0xFF800000
-; X64-NDD-NEXT: sete %al
-; X64-NDD-NEXT: cmpl $2139095040, %ecx # imm = 0x7F800000
-; X64-NDD-NEXT: setl %cl
-; X64-NDD-NEXT: orb %cl, %al
+; X64-NDD-NEXT: sete %dl
+; X64-NDD-NEXT: addl %eax, %eax
+; X64-NDD-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X64-NDD-NEXT: setb %al
; X64-NDD-NEXT: orb %dl, %al
+; X64-NDD-NEXT: orb %cl, %al
; X64-NDD-NEXT: retq
%class = tail call i1 @llvm.is.fpclass.f32(float %x, i32 509) ; ~(+inf|qnan)
ret i1 %class
@@ -2454,10 +2450,11 @@ define i1 @not_is_minus_inf_or_snan_f(float %x) {
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
; X86-NEXT: sete %cl
-; X86-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
-; X86-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X86-NEXT: setl %dl
+; X86-NEXT: leal (%eax,%eax), %edx
+; X86-NEXT: cmpl $-16777216, %edx # imm = 0xFF000000
+; X86-NEXT: setb %dl
; X86-NEXT: orb %cl, %dl
+; X86-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
; X86-NEXT: cmpl $2143289344, %eax # imm = 0x7FC00000
; X86-NEXT: setge %al
; X86-NEXT: orb %dl, %al
@@ -2468,10 +2465,11 @@ define i1 @not_is_minus_inf_or_snan_f(float %x) {
; X64-GENERIC-NEXT: movd %xmm0, %eax
; X64-GENERIC-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
; X64-GENERIC-NEXT: sete %cl
-; X64-GENERIC-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
-; X64-GENERIC-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X64-GENERIC-NEXT: setl %dl
+; X64-GENERIC-NEXT: leal (%rax,%rax), %edx
+; X64-GENERIC-NEXT: cmpl $-16777216, %edx # imm = 0xFF000000
+; X64-GENERIC-NEXT: setb %dl
; X64-GENERIC-NEXT: orb %cl, %dl
+; X64-GENERIC-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
; X64-GENERIC-NEXT: cmpl $2143289344, %eax # imm = 0x7FC00000
; X64-GENERIC-NEXT: setge %al
; X64-GENERIC-NEXT: orb %dl, %al
@@ -2482,10 +2480,11 @@ define i1 @not_is_minus_inf_or_snan_f(float %x) {
; X64-NDD-NEXT: movd %xmm0, %eax
; X64-NDD-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
; X64-NDD-NEXT: sete %cl
-; X64-NDD-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
-; X64-NDD-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X64-NDD-NEXT: setl %dl
+; X64-NDD-NEXT: leal (%rax,%rax), %edx
+; X64-NDD-NEXT: cmpl $-16777216, %edx # imm = 0xFF000000
+; X64-NDD-NEXT: setb %dl
; X64-NDD-NEXT: orb %dl, %cl
+; X64-NDD-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
; X64-NDD-NEXT: cmpl $2143289344, %eax # imm = 0x7FC00000
; X64-NDD-NEXT: setge %al
; X64-NDD-NEXT: orb %cl, %al
@@ -2503,14 +2502,15 @@ define i1 @not_is_minus_inf_or_qnan_f(float %x) {
; X86-NEXT: cmpl $2143289344, %ecx # imm = 0x7FC00000
; X86-NEXT: setl %dl
; X86-NEXT: cmpl $2139095041, %ecx # imm = 0x7F800001
-; X86-NEXT: setge %dh
-; X86-NEXT: andb %dl, %dh
+; X86-NEXT: setge %cl
+; X86-NEXT: andb %dl, %cl
; X86-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
; X86-NEXT: sete %dl
-; X86-NEXT: cmpl $2139095040, %ecx # imm = 0x7F800000
-; X86-NEXT: setl %al
+; X86-NEXT: addl %eax, %eax
+; X86-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X86-NEXT: setb %al
; X86-NEXT: orb %dl, %al
-; X86-NEXT: orb %dh, %al
+; X86-NEXT: orb %cl, %al
; X86-NEXT: retl
;
; X64-GENERIC-LABEL: not_is_minus_inf_or_qnan_f:
@@ -2521,14 +2521,15 @@ define i1 @not_is_minus_inf_or_qnan_f(float %x) {
; X64-GENERIC-NEXT: cmpl $2143289344, %ecx # imm = 0x7FC00000
; X64-GENERIC-NEXT: setl %dl
; X64-GENERIC-NEXT: cmpl $2139095041, %ecx # imm = 0x7F800001
-; X64-GENERIC-NEXT: setge %sil
-; X64-GENERIC-NEXT: andb %dl, %sil
+; X64-GENERIC-NEXT: setge %cl
+; X64-GENERIC-NEXT: andb %dl, %cl
; X64-GENERIC-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
; X64-GENERIC-NEXT: sete %dl
-; X64-GENERIC-NEXT: cmpl $2139095040, %ecx # imm = 0x7F800000
-; X64-GENERIC-NEXT: setl %al
+; X64-GENERIC-NEXT: addl %eax, %eax
+; X64-GENERIC-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X64-GENERIC-NEXT: setb %al
; X64-GENERIC-NEXT: orb %dl, %al
-; X64-GENERIC-NEXT: orb %sil, %al
+; X64-GENERIC-NEXT: orb %cl, %al
; X64-GENERIC-NEXT: retq
;
; X64-NDD-LABEL: not_is_minus_inf_or_qnan_f:
@@ -2538,14 +2539,15 @@ define i1 @not_is_minus_inf_or_qnan_f(float %x) {
; X64-NDD-NEXT: cmpl $2143289344, %ecx # imm = 0x7FC00000
; X64-NDD-NEXT: setl %dl
; X64-NDD-NEXT: cmpl $2139095041, %ecx # imm = 0x7F800001
-; X64-NDD-NEXT: setge %sil
-; X64-NDD-NEXT: andb %sil, %dl
+; X64-NDD-NEXT: setge %cl
+; X64-NDD-NEXT: andb %dl, %cl
; X64-NDD-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X64-NDD-NEXT: sete %al
-; X64-NDD-NEXT: cmpl $2139095040, %ecx # imm = 0x7F800000
-; X64-NDD-NEXT: setl %cl
-; X64-NDD-NEXT: orb %cl, %al
+; X64-NDD-NEXT: sete %dl
+; X64-NDD-NEXT: addl %eax, %eax
+; X64-NDD-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X64-NDD-NEXT: setb %al
; X64-NDD-NEXT: orb %dl, %al
+; X64-NDD-NEXT: orb %cl, %al
; X64-NDD-NEXT: retq
%class = tail call i1 @llvm.is.fpclass.f32(float %x, i32 1017) ; "-inf|qnan"
ret i1 %class
diff --git a/llvm/test/CodeGen/X86/isel-fpclass.ll b/llvm/test/CodeGen/X86/isel-fpclass.ll
index 0393ae602fd10..cfc3cc2f07d21 100644
--- a/llvm/test/CodeGen/X86/isel-fpclass.ll
+++ b/llvm/test/CodeGen/X86/isel-fpclass.ll
@@ -402,18 +402,18 @@ entry:
define i1 @isfinite_f(float %x) nounwind {
; X86-LABEL: isfinite_f:
; X86: # %bb.0: # %entry
-; X86-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF
-; X86-NEXT: andl {{[0-9]+}}(%esp), %eax
-; X86-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X86-NEXT: setl %al
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: addl %eax, %eax
+; X86-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X86-NEXT: setb %al
; X86-NEXT: retl
;
; X64-LABEL: isfinite_f:
; X64: # %bb.0: # %entry
; X64-NEXT: movd %xmm0, %eax
-; X64-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
-; X64-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X64-NEXT: setl %al
+; X64-NEXT: addl %eax, %eax
+; X64-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X64-NEXT: setb %al
; X64-NEXT: retq
;
; X86-FASTISEL-LABEL: isfinite_f:
@@ -421,10 +421,10 @@ define i1 @isfinite_f(float %x) nounwind {
; X86-FASTISEL-NEXT: pushl %eax
; X86-FASTISEL-NEXT: flds {{[0-9]+}}(%esp)
; X86-FASTISEL-NEXT: fstps (%esp)
-; X86-FASTISEL-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF
-; X86-FASTISEL-NEXT: andl (%esp), %eax
-; X86-FASTISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X86-FASTISEL-NEXT: setl %al
+; X86-FASTISEL-NEXT: movl (%esp), %eax
+; X86-FASTISEL-NEXT: addl %eax, %eax
+; X86-FASTISEL-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X86-FASTISEL-NEXT: setb %al
; X86-FASTISEL-NEXT: popl %ecx
; X86-FASTISEL-NEXT: retl
;
@@ -443,18 +443,18 @@ entry:
define i1 @not_isfinite_f(float %x) nounwind {
; X86-LABEL: not_isfinite_f:
; X86: # %bb.0: # %entry
-; X86-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF
-; X86-NEXT: andl {{[0-9]+}}(%esp), %eax
-; X86-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X86-NEXT: setge %al
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: addl %eax, %eax
+; X86-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X86-NEXT: setae %al
; X86-NEXT: retl
;
; X64-LABEL: not_isfinite_f:
; X64: # %bb.0: # %entry
; X64-NEXT: movd %xmm0, %eax
-; X64-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF
-; X64-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X64-NEXT: setge %al
+; X64-NEXT: addl %eax, %eax
+; X64-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X64-NEXT: setae %al
; X64-NEXT: retq
;
; X86-FASTISEL-LABEL: not_isfinite_f:
@@ -462,10 +462,10 @@ define i1 @not_isfinite_f(float %x) nounwind {
; X86-FASTISEL-NEXT: pushl %eax
; X86-FASTISEL-NEXT: flds {{[0-9]+}}(%esp)
; X86-FASTISEL-NEXT: fstps (%esp)
-; X86-FASTISEL-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF
-; X86-FASTISEL-NEXT: andl (%esp), %eax
-; X86-FASTISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000
-; X86-FASTISEL-NEXT: setge %al
+; X86-FASTISEL-NEXT: movl (%esp), %eax
+; X86-FASTISEL-NEXT: addl %eax, %eax
+; X86-FASTISEL-NEXT: cmpl $-16777216, %eax # imm = 0xFF000000
+; X86-FASTISEL-NEXT: setae %al
; X86-FASTISEL-NEXT: popl %ecx
; X86-FASTISEL-NEXT: retl
;
More information about the llvm-commits
mailing list