Skip to content

[NVPTX] Expand EXTLOAD for v8f16 and v8bf16 #72672

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Nov 17, 2023
Merged

[NVPTX] Expand EXTLOAD for v8f16 and v8bf16 #72672

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -606,6 +606,10 @@ NVPTXTargetLowering::NVPTXTargetLowering(const NVPTXTargetMachine &TM,
setLoadExtAction(ISD::EXTLOAD, MVT::v4f32, MVT::v4bf16, Expand);
setLoadExtAction(ISD::EXTLOAD, MVT::v4f64, MVT::v4bf16, Expand);
setLoadExtAction(ISD::EXTLOAD, MVT::v4f64, MVT::v4f32, Expand);
setLoadExtAction(ISD::EXTLOAD, MVT::v8f32, MVT::v8f16, Expand);
setLoadExtAction(ISD::EXTLOAD, MVT::v8f64, MVT::v8f16, Expand);
setLoadExtAction(ISD::EXTLOAD, MVT::v8f32, MVT::v8bf16, Expand);
setLoadExtAction(ISD::EXTLOAD, MVT::v8f64, MVT::v8bf16, Expand);
// Turn FP truncstore into trunc + store.
// FIXME: vector types should also be expanded
setTruncStoreAction(MVT::f32, MVT::f16, Expand);
Expand Down
20 changes: 20 additions & 0 deletions llvm/test/CodeGen/NVPTX/bf16-instructions.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -207,3 +207,23 @@ define bfloat @test_select_cc_bf16_f64(double %a, double %b, bfloat %c, bfloat %
%r = select i1 %cc, bfloat %c, bfloat %d
ret bfloat %r
}

; CHECK-LABEL: test_extload_bf16x8
; CHECK: ld.shared.v4.b32 {%r
; CHECK: mov.b32 {%rs
; CHECK: mov.b32 {%rs
; CHECK: mov.b32 {%rs
; CHECK: mov.b32 {%rs
; SM80: cvt.f32.bf16 %f{{.*}}, %rs
; SM80: cvt.f32.bf16 %f{{.*}}, %rs
; SM80: cvt.f32.bf16 %f{{.*}}, %rs
; SM80: cvt.f32.bf16 %f{{.*}}, %rs
; SM80: cvt.f32.bf16 %f{{.*}}, %rs
; SM80: cvt.f32.bf16 %f{{.*}}, %rs
; SM80: cvt.f32.bf16 %f{{.*}}, %rs
; SM80: cvt.f32.bf16 %f{{.*}}, %rs
define <8 x float> @test_extload_bf16x8(ptr addrspace(3) noundef %arg) #0 {
%load = load <8 x bfloat>, ptr addrspace(3) %arg, align 16
%res = fpext <8 x bfloat> %load to <8 x float>
ret <8 x float> %res
}
70 changes: 58 additions & 12 deletions llvm/test/CodeGen/NVPTX/vector-loads.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -99,9 +99,20 @@ define void @foo_complex(ptr nocapture readonly align 16 dereferenceable(1342177

; CHECK-LABEL: extv8f16_global_a16(
define void @extv8f16_global_a16(ptr addrspace(1) noalias readonly align 16 %dst, ptr addrspace(1) noalias readonly align 16 %src) #0 {
; CHECK: ld.global.v4.b16 {%f
; CHECK: ld.global.v4.b16 {%f
; CHECK: ld.global.v4.b32 {%r
%v = load <8 x half>, ptr addrspace(1) %src, align 16
; CHECK: mov.b32 {%rs
; CHECK: mov.b32 {%rs
; CHECK: mov.b32 {%rs
; CHECK: mov.b32 {%rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
%ext = fpext <8 x half> %v to <8 x float>
; CHECK: st.global.v4.f32
; CHECK: st.global.v4.f32
Expand All @@ -111,11 +122,23 @@ define void @extv8f16_global_a16(ptr addrspace(1) noalias readonly align 16 %dst

; CHECK-LABEL: extv8f16_global_a4(
define void @extv8f16_global_a4(ptr addrspace(1) noalias readonly align 16 %dst, ptr addrspace(1) noalias readonly align 16 %src) #0 {
; CHECK: ld.global.v2.b16 {%f
; CHECK: ld.global.v2.b16 {%f
; CHECK: ld.global.v2.b16 {%f
; CHECK: ld.global.v2.b16 {%f
; CHECK: ld.global.b32 %r
; CHECK: ld.global.b32 %r
; CHECK: ld.global.b32 %r
; CHECK: ld.global.b32 %r
%v = load <8 x half>, ptr addrspace(1) %src, align 4
; CHECK: mov.b32 {%rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: mov.b32 {%rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: mov.b32 {%rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: mov.b32 {%rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
%ext = fpext <8 x half> %v to <8 x float>
; CHECK: st.global.v4.f32
; CHECK: st.global.v4.f32
Expand All @@ -126,9 +149,20 @@ define void @extv8f16_global_a4(ptr addrspace(1) noalias readonly align 16 %dst,

; CHECK-LABEL: extv8f16_generic_a16(
define void @extv8f16_generic_a16(ptr noalias readonly align 16 %dst, ptr noalias readonly align 16 %src) #0 {
; CHECK: ld.v4.b16 {%f
; CHECK: ld.v4.b16 {%f
; CHECK: ld.v4.b32 {%r
%v = load <8 x half>, ptr %src, align 16
; CHECK: mov.b32 {%rs
; CHECK: mov.b32 {%rs
; CHECK: mov.b32 {%rs
; CHECK: mov.b32 {%rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
%ext = fpext <8 x half> %v to <8 x float>
; CHECK: st.v4.f32
; CHECK: st.v4.f32
Expand All @@ -138,11 +172,23 @@ define void @extv8f16_generic_a16(ptr noalias readonly align 16 %dst, ptr noalia

; CHECK-LABEL: extv8f16_generic_a4(
define void @extv8f16_generic_a4(ptr noalias readonly align 16 %dst, ptr noalias readonly align 16 %src) #0 {
; CHECK: ld.v2.b16 {%f
; CHECK: ld.v2.b16 {%f
; CHECK: ld.v2.b16 {%f
; CHECK: ld.v2.b16 {%f
; CHECK: ld.b32 %r
; CHECK: ld.b32 %r
; CHECK: ld.b32 %r
; CHECK: ld.b32 %r
%v = load <8 x half>, ptr %src, align 4
; CHECK: mov.b32 {%rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: mov.b32 {%rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: mov.b32 {%rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: mov.b32 {%rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
; CHECK: cvt.f32.f16 %f{{.*}}, %rs
%ext = fpext <8 x half> %v to <8 x float>
; CHECK: st.v4.f32
; CHECK: st.v4.f32
Expand Down