summaryrefslogtreecommitdiff
path: root/test/Transforms/InstCombine/amdgcn-intrinsics.ll
diff options
context:
space:
mode:
Diffstat (limited to 'test/Transforms/InstCombine/amdgcn-intrinsics.ll')
-rw-r--r--test/Transforms/InstCombine/amdgcn-intrinsics.ll903
1 files changed, 903 insertions, 0 deletions
diff --git a/test/Transforms/InstCombine/amdgcn-intrinsics.ll b/test/Transforms/InstCombine/amdgcn-intrinsics.ll
index a228968f25bce..deae5502bcdb8 100644
--- a/test/Transforms/InstCombine/amdgcn-intrinsics.ll
+++ b/test/Transforms/InstCombine/amdgcn-intrinsics.ll
@@ -7,6 +7,12 @@
declare float @llvm.amdgcn.rcp.f32(float) nounwind readnone
declare double @llvm.amdgcn.rcp.f64(double) nounwind readnone
+; CHECK-LABEL: @test_constant_fold_rcp_f32_undef
+; CHECK-NEXT: ret float undef
+define float @test_constant_fold_rcp_f32_undef() nounwind {
+ %val = call float @llvm.amdgcn.rcp.f32(float undef) nounwind readnone
+ ret float %val
+}
; CHECK-LABEL: @test_constant_fold_rcp_f32_1
; CHECK-NEXT: ret float 1.000000e+00
@@ -50,6 +56,18 @@ define double @test_constant_fold_rcp_f64_43() nounwind {
ret double %val
}
+; --------------------------------------------------------------------
+; llvm.amdgcn.rsq
+; --------------------------------------------------------------------
+
+declare float @llvm.amdgcn.rsq.f32(float) nounwind readnone
+
+; CHECK-LABEL: @test_constant_fold_rsq_f32_undef
+; CHECK-NEXT: ret float undef
+define float @test_constant_fold_rsq_f32_undef() nounwind {
+ %val = call float @llvm.amdgcn.rsq.f32(float undef) nounwind readnone
+ ret float %val
+}
; --------------------------------------------------------------------
; llvm.amdgcn.frexp.mant
@@ -633,3 +651,888 @@ define float @cos_fabs_fneg_f32(float %x) {
%cos = call float @llvm.amdgcn.cos.f32(float %x.fabs.fneg)
ret float %cos
}
+
+; --------------------------------------------------------------------
+; llvm.amdgcn.cvt.pkrtz
+; --------------------------------------------------------------------
+
+declare <2 x half> @llvm.amdgcn.cvt.pkrtz(float, float) nounwind readnone
+
+; CHECK-LABEL: @vars_lhs_cvt_pkrtz(
+; CHECK: %cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %x, float %y)
+define <2 x half> @vars_lhs_cvt_pkrtz(float %x, float %y) {
+ %cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %x, float %y)
+ ret <2 x half> %cvt
+}
+
+; CHECK-LABEL: @constant_lhs_cvt_pkrtz(
+; CHECK: %cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float 0.000000e+00, float %y)
+define <2 x half> @constant_lhs_cvt_pkrtz(float %y) {
+ %cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float 0.0, float %y)
+ ret <2 x half> %cvt
+}
+
+; CHECK-LABEL: @constant_rhs_cvt_pkrtz(
+; CHECK: %cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %x, float 0.000000e+00)
+define <2 x half> @constant_rhs_cvt_pkrtz(float %x) {
+ %cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %x, float 0.0)
+ ret <2 x half> %cvt
+}
+
+; CHECK-LABEL: @undef_lhs_cvt_pkrtz(
+; CHECK: %cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float undef, float %y)
+define <2 x half> @undef_lhs_cvt_pkrtz(float %y) {
+ %cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float undef, float %y)
+ ret <2 x half> %cvt
+}
+
+; CHECK-LABEL: @undef_rhs_cvt_pkrtz(
+; CHECK: %cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %x, float undef)
+define <2 x half> @undef_rhs_cvt_pkrtz(float %x) {
+ %cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %x, float undef)
+ ret <2 x half> %cvt
+}
+
+; CHECK-LABEL: @undef_cvt_pkrtz(
+; CHECK: ret <2 x half> undef
+define <2 x half> @undef_cvt_pkrtz() {
+ %cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float undef, float undef)
+ ret <2 x half> %cvt
+}
+
+; CHECK-LABEL: @constant_splat0_cvt_pkrtz(
+; CHECK: ret <2 x half> zeroinitializer
+define <2 x half> @constant_splat0_cvt_pkrtz() {
+ %cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float 0.0, float 0.0)
+ ret <2 x half> %cvt
+}
+
+; CHECK-LABEL: @constant_cvt_pkrtz(
+; CHECK: ret <2 x half> <half 0xH4000, half 0xH4400>
+define <2 x half> @constant_cvt_pkrtz() {
+ %cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float 2.0, float 4.0)
+ ret <2 x half> %cvt
+}
+
+; Test constant values where rtz changes result
+; CHECK-LABEL: @constant_rtz_pkrtz(
+; CHECK: ret <2 x half> <half 0xH7BFF, half 0xH7BFF>
+define <2 x half> @constant_rtz_pkrtz() {
+ %cvt = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float 65535.0, float 65535.0)
+ ret <2 x half> %cvt
+}
+
+; --------------------------------------------------------------------
+; llvm.amdgcn.ubfe
+; --------------------------------------------------------------------
+
+declare i32 @llvm.amdgcn.ubfe.i32(i32, i32, i32) nounwind readnone
+declare i64 @llvm.amdgcn.ubfe.i64(i64, i32, i32) nounwind readnone
+
+; CHECK-LABEL: @ubfe_var_i32(
+; CHECK-NEXT: %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 %offset, i32 %width)
+define i32 @ubfe_var_i32(i32 %src, i32 %offset, i32 %width) {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 %offset, i32 %width)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_clear_high_bits_constant_offset_i32(
+; CHECK-NEXT: %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 5, i32 %width)
+define i32 @ubfe_clear_high_bits_constant_offset_i32(i32 %src, i32 %width) {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 133, i32 %width)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_clear_high_bits_constant_width_i32(
+; CHECK-NEXT: %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 %offset, i32 5)
+define i32 @ubfe_clear_high_bits_constant_width_i32(i32 %src, i32 %offset) {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 %offset, i32 133)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_width_0(
+; CHECK-NEXT: ret i32 0
+define i32 @ubfe_width_0(i32 %src, i32 %offset) {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 %offset, i32 0)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_width_31(
+; CHECK: %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 %offset, i32 31)
+define i32 @ubfe_width_31(i32 %src, i32 %offset) {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 %offset, i32 31)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_width_32(
+; CHECK-NEXT: ret i32 0
+define i32 @ubfe_width_32(i32 %src, i32 %offset) {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 %offset, i32 32)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_width_33(
+; CHECK-NEXT: %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 %offset, i32 1)
+define i32 @ubfe_width_33(i32 %src, i32 %offset) {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 %offset, i32 33)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_offset_33(
+; CHECK-NEXT: %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 1, i32 %width)
+define i32 @ubfe_offset_33(i32 %src, i32 %width) {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 33, i32 %width)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_offset_0(
+; CHECK-NEXT: %1 = sub i32 32, %width
+; CHECK-NEXT: %2 = shl i32 %src, %1
+; CHECK-NEXT: %bfe = lshr i32 %2, %1
+; CHECK-NEXT: ret i32 %bfe
+define i32 @ubfe_offset_0(i32 %src, i32 %width) {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 0, i32 %width)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_offset_32(
+; CHECK-NEXT: %1 = sub i32 32, %width
+; CHECK-NEXT: %2 = shl i32 %src, %1
+; CHECK-NEXT: %bfe = lshr i32 %2, %1
+; CHECK-NEXT: ret i32 %bfe
+define i32 @ubfe_offset_32(i32 %src, i32 %width) {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 32, i32 %width)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_offset_31(
+; CHECK-NEXT: %1 = sub i32 32, %width
+; CHECK-NEXT: %2 = shl i32 %src, %1
+; CHECK-NEXT: %bfe = lshr i32 %2, %1
+; CHECK-NEXT: ret i32 %bfe
+define i32 @ubfe_offset_31(i32 %src, i32 %width) {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 32, i32 %width)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_offset_0_width_0(
+; CHECK-NEXT: ret i32 0
+define i32 @ubfe_offset_0_width_0(i32 %src) {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 0, i32 0)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_offset_0_width_3(
+; CHECK-NEXT: and i32 %src, 7
+; CHECK-NEXT: ret
+define i32 @ubfe_offset_0_width_3(i32 %src) {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 0, i32 3)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_offset_3_width_1(
+; CHECK-NEXT: %1 = lshr i32 %src, 3
+; CHECK-NEXT: and i32 %1, 1
+; CHECK-NEXT: ret i32
+define i32 @ubfe_offset_3_width_1(i32 %src) {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 3, i32 1)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_offset_3_width_4(
+; CHECK-NEXT: %1 = lshr i32 %src, 3
+; CHECK-NEXT: and i32 %1, 15
+; CHECK-NEXT: ret i32
+define i32 @ubfe_offset_3_width_4(i32 %src) {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 3, i32 4)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_0_0_0(
+; CHECK-NEXT: ret i32 0
+define i32 @ubfe_0_0_0() {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 0, i32 0, i32 0)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_neg1_5_7(
+; CHECK-NEXT: ret i32 127
+define i32 @ubfe_neg1_5_7() {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 -1, i32 5, i32 7)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_undef_src_i32(
+; CHECK-NEXT: ret i32 undef
+define i32 @ubfe_undef_src_i32(i32 %offset, i32 %width) {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 undef, i32 %offset, i32 %width)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_undef_offset_i32(
+; CHECK: %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 undef, i32 %width)
+define i32 @ubfe_undef_offset_i32(i32 %src, i32 %width) {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 undef, i32 %width)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_undef_width_i32(
+; CHECK: %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 %offset, i32 undef)
+define i32 @ubfe_undef_width_i32(i32 %src, i32 %offset) {
+ %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %src, i32 %offset, i32 undef)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @ubfe_offset_33_width_4_i64(
+; CHECK-NEXT: %1 = lshr i64 %src, 33
+; CHECK-NEXT: %bfe = and i64 %1, 15
+define i64 @ubfe_offset_33_width_4_i64(i64 %src) {
+ %bfe = call i64 @llvm.amdgcn.ubfe.i64(i64 %src, i32 33, i32 4)
+ ret i64 %bfe
+}
+
+; CHECK-LABEL: @ubfe_offset_0_i64(
+; CHECK-NEXT: %1 = sub i32 64, %width
+; CHECK-NEXT: %2 = zext i32 %1 to i64
+; CHECK-NEXT: %3 = shl i64 %src, %2
+; CHECK-NEXT: %bfe = lshr i64 %3, %2
+; CHECK-NEXT: ret i64 %bfe
+define i64 @ubfe_offset_0_i64(i64 %src, i32 %width) {
+ %bfe = call i64 @llvm.amdgcn.ubfe.i64(i64 %src, i32 0, i32 %width)
+ ret i64 %bfe
+}
+
+; CHECK-LABEL: @ubfe_offset_32_width_32_i64(
+; CHECK-NEXT: %bfe = lshr i64 %src, 32
+; CHECK-NEXT: ret i64 %bfe
+define i64 @ubfe_offset_32_width_32_i64(i64 %src) {
+ %bfe = call i64 @llvm.amdgcn.ubfe.i64(i64 %src, i32 32, i32 32)
+ ret i64 %bfe
+}
+
+; --------------------------------------------------------------------
+; llvm.amdgcn.sbfe
+; --------------------------------------------------------------------
+
+declare i32 @llvm.amdgcn.sbfe.i32(i32, i32, i32) nounwind readnone
+declare i64 @llvm.amdgcn.sbfe.i64(i64, i32, i32) nounwind readnone
+
+; CHECK-LABEL: @sbfe_offset_31(
+; CHECK-NEXT: %1 = sub i32 32, %width
+; CHECK-NEXT: %2 = shl i32 %src, %1
+; CHECK-NEXT: %bfe = ashr i32 %2, %1
+; CHECK-NEXT: ret i32 %bfe
+define i32 @sbfe_offset_31(i32 %src, i32 %width) {
+ %bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 %src, i32 32, i32 %width)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @sbfe_neg1_5_7(
+; CHECK-NEXT: ret i32 -1
+define i32 @sbfe_neg1_5_7() {
+ %bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 -1, i32 5, i32 7)
+ ret i32 %bfe
+}
+
+; CHECK-LABEL: @sbfe_offset_32_width_32_i64(
+; CHECK-NEXT: %bfe = ashr i64 %src, 32
+; CHECK-NEXT: ret i64 %bfe
+define i64 @sbfe_offset_32_width_32_i64(i64 %src) {
+ %bfe = call i64 @llvm.amdgcn.sbfe.i64(i64 %src, i32 32, i32 32)
+ ret i64 %bfe
+}
+
+; --------------------------------------------------------------------
+; llvm.amdgcn.exp
+; --------------------------------------------------------------------
+
+declare void @llvm.amdgcn.exp.f32(i32, i32, float, float, float, float, i1, i1) nounwind inaccessiblememonly
+
+; Make sure no crashing on invalid variable params
+; CHECK-LABEL: @exp_invalid_inputs(
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 %en, float 1.000000e+00, float 2.000000e+00, float 5.000000e-01, float 4.000000e+00, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 %tgt, i32 15, float 1.000000e+00, float 2.000000e+00, float 5.000000e-01, float 4.000000e+00, i1 true, i1 false)
+define void @exp_invalid_inputs(i32 %tgt, i32 %en) {
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 %en, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+ call void @llvm.amdgcn.exp.f32(i32 %tgt, i32 15, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+ ret void
+}
+
+; CHECK-LABEL: @exp_disabled_inputs_to_undef(
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 1, float 1.000000e+00, float undef, float undef, float undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float undef, float 2.000000e+00, float undef, float undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 4, float undef, float undef, float 5.000000e-01, float undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 8, float undef, float undef, float undef, float 4.000000e+00, i1 true, i1 false)
+
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 1, float %x, float undef, float undef, float undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float undef, float %y, float undef, float undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 4, float undef, float undef, float %z, float undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 8, float undef, float undef, float undef, float %w, i1 true, i1 false)
+
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 0, float undef, float undef, float undef, float undef, i1 true, i1 false)
+
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 3, float 1.000000e+00, float 2.000000e+00, float undef, float undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 5, float 1.000000e+00, float undef, float 5.000000e-01, float undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 9, float 1.000000e+00, float undef, float undef, float 4.000000e+00, i1 false, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float 1.000000e+00, float 2.000000e+00, float 5.000000e-01, float 4.000000e+00, i1 false, i1 false)
+define void @exp_disabled_inputs_to_undef(float %x, float %y, float %z, float %w) {
+ ; enable src0..src3 constants
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 1, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 4, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 8, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+
+ ; enable src0..src3 variables
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 1, float %x, float %y, float %z, float %w, i1 true, i1 false)
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float %x, float %y, float %z, float %w, i1 true, i1 false)
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 4, float %x, float %y, float %z, float %w, i1 true, i1 false)
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 8, float %x, float %y, float %z, float %w, i1 true, i1 false)
+
+ ; enable none
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 0, float %x, float %y, float %z, float %w, i1 true, i1 false)
+
+ ; enable different source combinations
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 3, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 5, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 9, float 1.0, float 2.0, float 0.5, float 4.0, i1 false, i1 false)
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float 1.0, float 2.0, float 0.5, float 4.0, i1 false, i1 false)
+
+ ret void
+}
+
+; --------------------------------------------------------------------
+; llvm.amdgcn.exp.compr
+; --------------------------------------------------------------------
+
+declare void @llvm.amdgcn.exp.compr.v2f16(i32, i32, <2 x half>, <2 x half>, i1, i1) nounwind inaccessiblememonly
+
+; CHECK-LABEL: @exp_compr_invalid_inputs(
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 %en, <2 x half> <half 0xH3C00, half 0xH4000>, <2 x half> <half 0xH3800, half 0xH4400>, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 %tgt, i32 5, <2 x half> <half 0xH3C00, half 0xH4000>, <2 x half> <half 0xH3800, half 0xH4400>, i1 true, i1 false)
+define void @exp_compr_invalid_inputs(i32 %tgt, i32 %en) {
+ call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 %en, <2 x half> <half 1.0, half 2.0>, <2 x half> <half 0.5, half 4.0>, i1 true, i1 false)
+ call void @llvm.amdgcn.exp.compr.v2f16(i32 %tgt, i32 5, <2 x half> <half 1.0, half 2.0>, <2 x half> <half 0.5, half 4.0>, i1 true, i1 false)
+ ret void
+}
+
+; CHECK-LABEL: @exp_compr_disabled_inputs_to_undef(
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 0, <2 x half> undef, <2 x half> undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 1, <2 x half> <half 0xH3C00, half 0xH4000>, <2 x half> undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 2, <2 x half> <half 0xH3C00, half 0xH4000>, <2 x half> undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 3, <2 x half> <half 0xH3C00, half 0xH4000>, <2 x half> undef, i1 true, i1 false)
+
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 0, <2 x half> undef, <2 x half> undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 1, <2 x half> %xy, <2 x half> undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 2, <2 x half> %xy, <2 x half> undef, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 3, <2 x half> %xy, <2 x half> undef, i1 true, i1 false)
+
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 12, <2 x half> undef, <2 x half> %zw, i1 true, i1 false)
+; CHECK: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 15, <2 x half> %xy, <2 x half> %zw, i1 true, i1 false)
+define void @exp_compr_disabled_inputs_to_undef(<2 x half> %xy, <2 x half> %zw) {
+ call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 0, <2 x half> <half 1.0, half 2.0>, <2 x half> <half 0.5, half 4.0>, i1 true, i1 false)
+ call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 1, <2 x half> <half 1.0, half 2.0>, <2 x half> <half 0.5, half 4.0>, i1 true, i1 false)
+ call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 2, <2 x half> <half 1.0, half 2.0>, <2 x half> <half 0.5, half 4.0>, i1 true, i1 false)
+ call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 3, <2 x half> <half 1.0, half 2.0>, <2 x half> <half 0.5, half 4.0>, i1 true, i1 false)
+
+ call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 0, <2 x half> %xy, <2 x half> %zw, i1 true, i1 false)
+ call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 1, <2 x half> %xy, <2 x half> %zw, i1 true, i1 false)
+ call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 2, <2 x half> %xy, <2 x half> %zw, i1 true, i1 false)
+ call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 3, <2 x half> %xy, <2 x half> %zw, i1 true, i1 false)
+
+ call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 12, <2 x half> %xy, <2 x half> %zw, i1 true, i1 false)
+ call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 15, <2 x half> %xy, <2 x half> %zw, i1 true, i1 false)
+ ret void
+}
+
+; --------------------------------------------------------------------
+; llvm.amdgcn.fmed3
+; --------------------------------------------------------------------
+
+declare float @llvm.amdgcn.fmed3.f32(float, float, float) nounwind readnone
+
+; CHECK-LABEL: @fmed3_f32(
+; CHECK: %med3 = call float @llvm.amdgcn.fmed3.f32(float %x, float %y, float %z)
+define float @fmed3_f32(float %x, float %y, float %z) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float %x, float %y, float %z)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_canonicalize_x_c0_c1_f32(
+; CHECK: call float @llvm.amdgcn.fmed3.f32(float %x, float 0.000000e+00, float 1.000000e+00)
+define float @fmed3_canonicalize_x_c0_c1_f32(float %x) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float %x, float 0.0, float 1.0)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_canonicalize_c0_x_c1_f32(
+; CHECK: call float @llvm.amdgcn.fmed3.f32(float %x, float 0.000000e+00, float 1.000000e+00)
+define float @fmed3_canonicalize_c0_x_c1_f32(float %x) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float 0.0, float %x, float 1.0)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_canonicalize_c0_c1_x_f32(
+; CHECK: call float @llvm.amdgcn.fmed3.f32(float %x, float 0.000000e+00, float 1.000000e+00)
+define float @fmed3_canonicalize_c0_c1_x_f32(float %x) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float 0.0, float 1.0, float %x)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_canonicalize_x_y_c_f32(
+; CHECK: call float @llvm.amdgcn.fmed3.f32(float %x, float %y, float 1.000000e+00)
+define float @fmed3_canonicalize_x_y_c_f32(float %x, float %y) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float %x, float %y, float 1.0)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_canonicalize_x_c_y_f32(
+; CHECK: %med3 = call float @llvm.amdgcn.fmed3.f32(float %x, float %y, float 1.000000e+00)
+define float @fmed3_canonicalize_x_c_y_f32(float %x, float %y) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float %x, float 1.0, float %y)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_canonicalize_c_x_y_f32(
+; CHECK: call float @llvm.amdgcn.fmed3.f32(float %x, float %y, float 1.000000e+00)
+define float @fmed3_canonicalize_c_x_y_f32(float %x, float %y) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float 1.0, float %x, float %y)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_undef_x_y_f32(
+; CHECK: call float @llvm.minnum.f32(float %x, float %y)
+define float @fmed3_undef_x_y_f32(float %x, float %y) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float undef, float %x, float %y)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_fmf_undef_x_y_f32(
+; CHECK: call nnan float @llvm.minnum.f32(float %x, float %y)
+define float @fmed3_fmf_undef_x_y_f32(float %x, float %y) {
+ %med3 = call nnan float @llvm.amdgcn.fmed3.f32(float undef, float %x, float %y)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_x_undef_y_f32(
+; CHECK: call float @llvm.minnum.f32(float %x, float %y)
+define float @fmed3_x_undef_y_f32(float %x, float %y) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float %x, float undef, float %y)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_x_y_undef_f32(
+; CHECK: call float @llvm.minnum.f32(float %x, float %y)
+define float @fmed3_x_y_undef_f32(float %x, float %y) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float %x, float %y, float undef)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_qnan0_x_y_f32(
+; CHECK: call float @llvm.minnum.f32(float %x, float %y)
+define float @fmed3_qnan0_x_y_f32(float %x, float %y) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float 0x7FF8000000000000, float %x, float %y)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_x_qnan0_y_f32(
+; CHECK: call float @llvm.minnum.f32(float %x, float %y)
+define float @fmed3_x_qnan0_y_f32(float %x, float %y) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float %x, float 0x7FF8000000000000, float %y)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_x_y_qnan0_f32(
+; CHECK: call float @llvm.minnum.f32(float %x, float %y)
+define float @fmed3_x_y_qnan0_f32(float %x, float %y) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float %x, float %y, float 0x7FF8000000000000)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_qnan1_x_y_f32(
+; CHECK: call float @llvm.minnum.f32(float %x, float %y)
+define float @fmed3_qnan1_x_y_f32(float %x, float %y) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float 0x7FF8000100000000, float %x, float %y)
+ ret float %med3
+}
+
+; This can return any of the qnans.
+; CHECK-LABEL: @fmed3_qnan0_qnan1_qnan2_f32(
+; CHECK: ret float 0x7FF8002000000000
+define float @fmed3_qnan0_qnan1_qnan2_f32(float %x, float %y) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float 0x7FF8000100000000, float 0x7FF8002000000000, float 0x7FF8030000000000)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_constant_src0_0_f32(
+; CHECK: ret float 5.000000e-01
+define float @fmed3_constant_src0_0_f32(float %x, float %y) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float 0.5, float -1.0, float 4.0)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_constant_src0_1_f32(
+; CHECK: ret float 5.000000e-01
+define float @fmed3_constant_src0_1_f32(float %x, float %y) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float 0.5, float 4.0, float -1.0)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_constant_src1_0_f32(
+; CHECK: ret float 5.000000e-01
+define float @fmed3_constant_src1_0_f32(float %x, float %y) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float -1.0, float 0.5, float 4.0)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_constant_src1_1_f32(
+; CHECK: ret float 5.000000e-01
+define float @fmed3_constant_src1_1_f32(float %x, float %y) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float 4.0, float 0.5, float -1.0)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_constant_src2_0_f32(
+; CHECK: ret float 5.000000e-01
+define float @fmed3_constant_src2_0_f32(float %x, float %y) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float -1.0, float 4.0, float 0.5)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_constant_src2_1_f32(
+; CHECK: ret float 5.000000e-01
+define float @fmed3_constant_src2_1_f32(float %x, float %y) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float 4.0, float -1.0, float 0.5)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_x_qnan0_qnan1_f32(
+; CHECK: ret float %x
+define float @fmed3_x_qnan0_qnan1_f32(float %x) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float %x, float 0x7FF8001000000000, float 0x7FF8002000000000)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_qnan0_x_qnan1_f32(
+; CHECK: ret float %x
+define float @fmed3_qnan0_x_qnan1_f32(float %x) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float 0x7FF8001000000000, float %x, float 0x7FF8002000000000)
+ ret float %med3
+}
+
+; CHECK-LABEL: @fmed3_qnan0_qnan1_x_f32(
+; CHECK: ret float %x
+define float @fmed3_qnan0_qnan1_x_f32(float %x) {
+ %med3 = call float @llvm.amdgcn.fmed3.f32(float 0x7FF8001000000000, float 0x7FF8002000000000, float %x)
+ ret float %med3
+}
+
+; --------------------------------------------------------------------
+; llvm.amdgcn.icmp
+; --------------------------------------------------------------------
+
+declare i64 @llvm.amdgcn.icmp.i32(i32, i32, i32) nounwind readnone convergent
+declare i64 @llvm.amdgcn.icmp.i64(i64, i64, i32) nounwind readnone convergent
+
+; Make sure there's no crash for invalid input
+; CHECK-LABEL: @invalid_nonconstant_icmp_code(
+; CHECK: call i64 @llvm.amdgcn.icmp.i32(i32 %a, i32 %b, i32 %c)
+define i64 @invalid_nonconstant_icmp_code(i32 %a, i32 %b, i32 %c) {
+ %result = call i64 @llvm.amdgcn.icmp.i32(i32 %a, i32 %b, i32 %c)
+ ret i64 %result
+}
+
+; CHECK-LABEL: @invalid_icmp_code(
+; CHECK: %under = call i64 @llvm.amdgcn.icmp.i32(i32 %a, i32 %b, i32 31)
+; CHECK: %over = call i64 @llvm.amdgcn.icmp.i32(i32 %a, i32 %b, i32 42)
+define i64 @invalid_icmp_code(i32 %a, i32 %b) {
+ %under = call i64 @llvm.amdgcn.icmp.i32(i32 %a, i32 %b, i32 31)
+ %over = call i64 @llvm.amdgcn.icmp.i32(i32 %a, i32 %b, i32 42)
+ %or = or i64 %under, %over
+ ret i64 %or
+}
+
+; CHECK-LABEL: @icmp_constant_inputs_false(
+; CHECK: ret i64 0
+define i64 @icmp_constant_inputs_false() {
+ %result = call i64 @llvm.amdgcn.icmp.i32(i32 9, i32 8, i32 32)
+ ret i64 %result
+}
+
+; CHECK-LABEL: @icmp_constant_inputs_true(
+; CHECK: ret i64 -1
+define i64 @icmp_constant_inputs_true() {
+ %result = call i64 @llvm.amdgcn.icmp.i32(i32 9, i32 8, i32 34)
+ ret i64 %result
+}
+
+; CHECK-LABEL: @icmp_constant_to_rhs_slt(
+; CHECK: %result = call i64 @llvm.amdgcn.icmp.i32(i32 %x, i32 9, i32 38)
+define i64 @icmp_constant_to_rhs_slt(i32 %x) {
+ %result = call i64 @llvm.amdgcn.icmp.i32(i32 9, i32 %x, i32 40)
+ ret i64 %result
+}
+
+; CHECK-LABEL: @fold_icmp_ne_0_zext_icmp_eq_i32(
+; CHECK-NEXT: call i64 @llvm.amdgcn.icmp.i32(i32 %a, i32 %b, i32 32)
+define i64 @fold_icmp_ne_0_zext_icmp_eq_i32(i32 %a, i32 %b) {
+ %cmp = icmp eq i32 %a, %b
+ %zext.cmp = zext i1 %cmp to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %zext.cmp, i32 0, i32 33)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_ne_0_zext_icmp_ne_i32(
+; CHECK-NEXT: call i64 @llvm.amdgcn.icmp.i32(i32 %a, i32 %b, i32 33)
+define i64 @fold_icmp_ne_0_zext_icmp_ne_i32(i32 %a, i32 %b) {
+ %cmp = icmp ne i32 %a, %b
+ %zext.cmp = zext i1 %cmp to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %zext.cmp, i32 0, i32 33)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_ne_0_zext_icmp_sle_i32(
+; CHECK-NEXT: call i64 @llvm.amdgcn.icmp.i32(i32 %a, i32 %b, i32 41)
+define i64 @fold_icmp_ne_0_zext_icmp_sle_i32(i32 %a, i32 %b) {
+ %cmp = icmp sle i32 %a, %b
+ %zext.cmp = zext i1 %cmp to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %zext.cmp, i32 0, i32 33)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_ne_0_zext_icmp_ugt_i64(
+; CHECK-NEXT: call i64 @llvm.amdgcn.icmp.i64(i64 %a, i64 %b, i32 34)
+define i64 @fold_icmp_ne_0_zext_icmp_ugt_i64(i64 %a, i64 %b) {
+ %cmp = icmp ugt i64 %a, %b
+ %zext.cmp = zext i1 %cmp to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %zext.cmp, i32 0, i32 33)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_ne_0_zext_icmp_ult_swap_i64(
+; CHECK-NEXT: call i64 @llvm.amdgcn.icmp.i64(i64 %a, i64 %b, i32 34)
+define i64 @fold_icmp_ne_0_zext_icmp_ult_swap_i64(i64 %a, i64 %b) {
+ %cmp = icmp ugt i64 %a, %b
+ %zext.cmp = zext i1 %cmp to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 0, i32 %zext.cmp, i32 33)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_ne_0_zext_fcmp_oeq_f32(
+; CHECK-NEXT: call i64 @llvm.amdgcn.fcmp.f32(float %a, float %b, i32 1)
+define i64 @fold_icmp_ne_0_zext_fcmp_oeq_f32(float %a, float %b) {
+ %cmp = fcmp oeq float %a, %b
+ %zext.cmp = zext i1 %cmp to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %zext.cmp, i32 0, i32 33)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_ne_0_zext_fcmp_une_f32(
+; CHECK-NEXT: call i64 @llvm.amdgcn.fcmp.f32(float %a, float %b, i32 14)
+define i64 @fold_icmp_ne_0_zext_fcmp_une_f32(float %a, float %b) {
+ %cmp = fcmp une float %a, %b
+ %zext.cmp = zext i1 %cmp to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %zext.cmp, i32 0, i32 33)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_ne_0_zext_fcmp_olt_f64(
+; CHECK-NEXT: call i64 @llvm.amdgcn.fcmp.f64(double %a, double %b, i32 4)
+define i64 @fold_icmp_ne_0_zext_fcmp_olt_f64(double %a, double %b) {
+ %cmp = fcmp olt double %a, %b
+ %zext.cmp = zext i1 %cmp to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %zext.cmp, i32 0, i32 33)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_sext_icmp_ne_0_i32(
+; CHECK: %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %a, i32 %b, i32 32)
+define i64 @fold_icmp_sext_icmp_ne_0_i32(i32 %a, i32 %b) {
+ %cmp = icmp eq i32 %a, %b
+ %sext.cmp = sext i1 %cmp to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %sext.cmp, i32 0, i32 33)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_eq_0_zext_icmp_eq_i32(
+; CHECK-NEXT: call i64 @llvm.amdgcn.icmp.i32(i32 %a, i32 %b, i32 33)
+define i64 @fold_icmp_eq_0_zext_icmp_eq_i32(i32 %a, i32 %b) {
+ %cmp = icmp eq i32 %a, %b
+ %zext.cmp = zext i1 %cmp to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %zext.cmp, i32 0, i32 32)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_eq_0_zext_icmp_slt_i32(
+; CHECK-NEXT: call i64 @llvm.amdgcn.icmp.i32(i32 %a, i32 %b, i32 39)
+define i64 @fold_icmp_eq_0_zext_icmp_slt_i32(i32 %a, i32 %b) {
+ %cmp = icmp slt i32 %a, %b
+ %zext.cmp = zext i1 %cmp to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %zext.cmp, i32 0, i32 32)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_eq_0_zext_fcmp_oeq_f32(
+; CHECK-NEXT: call i64 @llvm.amdgcn.fcmp.f32(float %a, float %b, i32 14)
+define i64 @fold_icmp_eq_0_zext_fcmp_oeq_f32(float %a, float %b) {
+ %cmp = fcmp oeq float %a, %b
+ %zext.cmp = zext i1 %cmp to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %zext.cmp, i32 0, i32 32)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_eq_0_zext_fcmp_ule_f32(
+; CHECK-NEXT: call i64 @llvm.amdgcn.fcmp.f32(float %a, float %b, i32 2)
+define i64 @fold_icmp_eq_0_zext_fcmp_ule_f32(float %a, float %b) {
+ %cmp = fcmp ule float %a, %b
+ %zext.cmp = zext i1 %cmp to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %zext.cmp, i32 0, i32 32)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_eq_0_zext_fcmp_ogt_f32(
+; CHECK-NEXT: call i64 @llvm.amdgcn.fcmp.f32(float %a, float %b, i32 13)
+define i64 @fold_icmp_eq_0_zext_fcmp_ogt_f32(float %a, float %b) {
+ %cmp = fcmp ogt float %a, %b
+ %zext.cmp = zext i1 %cmp to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %zext.cmp, i32 0, i32 32)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_zext_icmp_eq_1_i32(
+; CHECK-NEXT: call i64 @llvm.amdgcn.icmp.i32(i32 %a, i32 %b, i32 32)
+define i64 @fold_icmp_zext_icmp_eq_1_i32(i32 %a, i32 %b) {
+ %cmp = icmp eq i32 %a, %b
+ %zext.cmp = zext i1 %cmp to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %zext.cmp, i32 1, i32 32)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_zext_argi1_eq_1_i32(
+; CHECK: %zext.cond = zext i1 %cond to i32
+; CHECK: call i64 @llvm.amdgcn.icmp.i32(i32 %zext.cond, i32 0, i32 33)
+define i64 @fold_icmp_zext_argi1_eq_1_i32(i1 %cond) {
+ %zext.cond = zext i1 %cond to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %zext.cond, i32 1, i32 32)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_zext_argi1_eq_neg1_i32(
+; CHECK: %zext.cond = zext i1 %cond to i32
+; CHECK: call i64 @llvm.amdgcn.icmp.i32(i32 %zext.cond, i32 -1, i32 32)
+define i64 @fold_icmp_zext_argi1_eq_neg1_i32(i1 %cond) {
+ %zext.cond = zext i1 %cond to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %zext.cond, i32 -1, i32 32)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_sext_argi1_eq_1_i32(
+; CHECK: %sext.cond = sext i1 %cond to i32
+; CHECK: call i64 @llvm.amdgcn.icmp.i32(i32 %sext.cond, i32 1, i32 32)
+define i64 @fold_icmp_sext_argi1_eq_1_i32(i1 %cond) {
+ %sext.cond = sext i1 %cond to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %sext.cond, i32 1, i32 32)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_sext_argi1_eq_neg1_i32(
+; CHECK: %sext.cond = sext i1 %cond to i32
+; CHECK: call i64 @llvm.amdgcn.icmp.i32(i32 %sext.cond, i32 0, i32 33)
+define i64 @fold_icmp_sext_argi1_eq_neg1_i32(i1 %cond) {
+ %sext.cond = sext i1 %cond to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %sext.cond, i32 -1, i32 32)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_sext_argi1_eq_neg1_i64(
+; CHECK: %sext.cond = sext i1 %cond to i64
+; CHECK: call i64 @llvm.amdgcn.icmp.i64(i64 %sext.cond, i64 0, i32 33)
+define i64 @fold_icmp_sext_argi1_eq_neg1_i64(i1 %cond) {
+ %sext.cond = sext i1 %cond to i64
+ %mask = call i64 @llvm.amdgcn.icmp.i64(i64 %sext.cond, i64 -1, i32 32)
+ ret i64 %mask
+}
+
+; TODO: Should be able to fold to false
+; CHECK-LABEL: @fold_icmp_sext_icmp_eq_1_i32(
+; CHECK: %cmp = icmp eq i32 %a, %b
+; CHECK: %sext.cmp = sext i1 %cmp to i32
+; CHECK: %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %sext.cmp, i32 1, i32 32)
+define i64 @fold_icmp_sext_icmp_eq_1_i32(i32 %a, i32 %b) {
+ %cmp = icmp eq i32 %a, %b
+ %sext.cmp = sext i1 %cmp to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %sext.cmp, i32 1, i32 32)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_sext_icmp_eq_neg1_i32(
+; CHECK: call i64 @llvm.amdgcn.icmp.i32(i32 %a, i32 %b, i32 32)
+define i64 @fold_icmp_sext_icmp_eq_neg1_i32(i32 %a, i32 %b) {
+ %cmp = icmp eq i32 %a, %b
+ %sext.cmp = sext i1 %cmp to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %sext.cmp, i32 -1, i32 32)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_icmp_sext_icmp_sge_neg1_i32(
+; CHECK: call i64 @llvm.amdgcn.icmp.i32(i32 %a, i32 %b, i32 39)
+define i64 @fold_icmp_sext_icmp_sge_neg1_i32(i32 %a, i32 %b) {
+ %cmp = icmp sge i32 %a, %b
+ %sext.cmp = sext i1 %cmp to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %sext.cmp, i32 -1, i32 32)
+ ret i64 %mask
+}
+
+; CHECK-LABEL: @fold_not_icmp_ne_0_zext_icmp_sle_i32(
+; CHECK-NEXT: call i64 @llvm.amdgcn.icmp.i32(i32 %a, i32 %b, i32 38)
+define i64 @fold_not_icmp_ne_0_zext_icmp_sle_i32(i32 %a, i32 %b) {
+ %cmp = icmp sle i32 %a, %b
+ %not = xor i1 %cmp, true
+ %zext.cmp = zext i1 %not to i32
+ %mask = call i64 @llvm.amdgcn.icmp.i32(i32 %zext.cmp, i32 0, i32 33)
+ ret i64 %mask
+}
+
+; --------------------------------------------------------------------
+; llvm.amdgcn.fcmp
+; --------------------------------------------------------------------
+
+declare i64 @llvm.amdgcn.fcmp.f32(float, float, i32) nounwind readnone convergent
+
+; Make sure there's no crash for invalid input
+; CHECK-LABEL: @invalid_nonconstant_fcmp_code(
+; CHECK: call i64 @llvm.amdgcn.fcmp.f32(float %a, float %b, i32 %c)
+define i64 @invalid_nonconstant_fcmp_code(float %a, float %b, i32 %c) {
+ %result = call i64 @llvm.amdgcn.fcmp.f32(float %a, float %b, i32 %c)
+ ret i64 %result
+}
+
+; CHECK-LABEL: @invalid_fcmp_code(
+; CHECK: %under = call i64 @llvm.amdgcn.fcmp.f32(float %a, float %b, i32 -1)
+; CHECK: %over = call i64 @llvm.amdgcn.fcmp.f32(float %a, float %b, i32 16)
+define i64 @invalid_fcmp_code(float %a, float %b) {
+ %under = call i64 @llvm.amdgcn.fcmp.f32(float %a, float %b, i32 -1)
+ %over = call i64 @llvm.amdgcn.fcmp.f32(float %a, float %b, i32 16)
+ %or = or i64 %under, %over
+ ret i64 %or
+}
+
+; CHECK-LABEL: @fcmp_constant_inputs_false(
+; CHECK: ret i64 0
+define i64 @fcmp_constant_inputs_false() {
+ %result = call i64 @llvm.amdgcn.fcmp.f32(float 2.0, float 4.0, i32 1)
+ ret i64 %result
+}
+
+; CHECK-LABEL: @fcmp_constant_inputs_true(
+; CHECK: ret i64 -1
+define i64 @fcmp_constant_inputs_true() {
+ %result = call i64 @llvm.amdgcn.fcmp.f32(float 2.0, float 4.0, i32 4)
+ ret i64 %result
+}
+
+; CHECK-LABEL: @fcmp_constant_to_rhs_olt(
+; CHECK: %result = call i64 @llvm.amdgcn.fcmp.f32(float %x, float 4.000000e+00, i32 2)
+define i64 @fcmp_constant_to_rhs_olt(float %x) {
+ %result = call i64 @llvm.amdgcn.fcmp.f32(float 4.0, float %x, i32 4)
+ ret i64 %result
+}