Skip to content

AMDGPU: Add baseline test for nofpclass on call results #167263

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

Jump to bottom
Merged
merged 1 commit into from
Nov 10, 2025
Merged

AMDGPU: Add baseline test for nofpclass on call results #167263

merged 1 commit into from
Nov 10, 2025

Conversation

@arsenm
Copy link
Contributor

@arsenm arsenm commented Nov 10, 2025

No description provided.

@arsenm arsenm added the llvm:SelectionDAG SelectionDAGISel as well label Nov 10, 2025 — with Graphite App
@arsenm Graphite App
Copy link
Contributor Author

arsenm commented Nov 10, 2025 •
edited
Loading

This stack of pull requests is managed by Graphite. Learn more about stacking.

@arsenm arsenm added the floating-point Floating-point math label Nov 10, 2025 — with Graphite App
@arsenm arsenm mentioned this pull request Nov 10, 2025
Merged
@llvmbot
Copy link
Member

llvmbot commented Nov 10, 2025 •
edited
Loading

@llvm/pr-subscribers-backend-amdgpu

@llvm/pr-subscribers-llvm-selectiondag

Author: Matt Arsenault (arsenm)

Changes

Full diff: https://github.com/llvm/llvm-project/pull/167263.diff

1 Files Affected:

  • (added) llvm/test/CodeGen/AMDGPU/nofpclass-call.ll (+199)
diff --git a/llvm/test/CodeGen/AMDGPU/nofpclass-call.ll b/llvm/test/CodeGen/AMDGPU/nofpclass-call.ll
new file mode 100644
index 0000000000000..1861f02ec8b1c
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/nofpclass-call.ll
@@ -0,0 +1,199 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck %s
+
+; Check that nofpclass attributes on call returns are used in
+; selectiondag.
+
+define internal float @func_f32(ptr addrspace(1) %ptr) {
+; CHECK-LABEL: func_f32:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT:    global_load_dword v0, v[0:1], off glc
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    s_setpc_b64 s[30:31]
+  %ld = load volatile float, ptr addrspace(1) %ptr
+  ret float %ld
+}
+
+define float @call_nofpclass_funcs_f32(ptr addrspace(1) %ptr) {
+; CHECK-LABEL: call_nofpclass_funcs_f32:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT:    s_mov_b32 s18, s33
+; CHECK-NEXT:    s_mov_b32 s33, s32
+; CHECK-NEXT:    s_xor_saveexec_b64 s[16:17], -1
+; CHECK-NEXT:    buffer_store_dword v4, off, s[0:3], s33 ; 4-byte Folded Spill
+; CHECK-NEXT:    s_mov_b64 exec, s[16:17]
+; CHECK-NEXT:    s_addk_i32 s32, 0x400
+; CHECK-NEXT:    v_writelane_b32 v4, s30, 0
+; CHECK-NEXT:    s_getpc_b64 s[16:17]
+; CHECK-NEXT:    s_add_u32 s16, s16, func_f32@rel32@lo+4
+; CHECK-NEXT:    s_addc_u32 s17, s17, func_f32@rel32@hi+12
+; CHECK-NEXT:    v_writelane_b32 v4, s31, 1
+; CHECK-NEXT:    v_mov_b32_e32 v2, v0
+; CHECK-NEXT:    s_swappc_b64 s[30:31], s[16:17]
+; CHECK-NEXT:    v_mov_b32_e32 v3, v0
+; CHECK-NEXT:    v_mov_b32_e32 v0, v2
+; CHECK-NEXT:    s_swappc_b64 s[30:31], s[16:17]
+; CHECK-NEXT:    v_max_f32_e32 v1, v3, v3
+; CHECK-NEXT:    v_max_f32_e32 v0, v0, v0
+; CHECK-NEXT:    v_min_f32_e32 v0, v1, v0
+; CHECK-NEXT:    v_readlane_b32 s31, v4, 1
+; CHECK-NEXT:    v_readlane_b32 s30, v4, 0
+; CHECK-NEXT:    s_mov_b32 s32, s33
+; CHECK-NEXT:    s_xor_saveexec_b64 s[4:5], -1
+; CHECK-NEXT:    buffer_load_dword v4, off, s[0:3], s33 ; 4-byte Folded Reload
+; CHECK-NEXT:    s_mov_b64 exec, s[4:5]
+; CHECK-NEXT:    s_mov_b32 s33, s18
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    s_setpc_b64 s[30:31]
+  %call0 = call nofpclass(nan) float @func_f32(ptr addrspace(1) %ptr)
+  %call1 = call nofpclass(nan) float @func_f32(ptr addrspace(1) %ptr)
+  %min = call float @llvm.minnum.f32(float %call0, float %call1)
+  ret float %min
+}
+
+define internal <2 x float> @func_v2f32(ptr addrspace(1) %ptr) {
+; CHECK-LABEL: func_v2f32:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off glc
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    s_setpc_b64 s[30:31]
+  %ld = load volatile <2 x float>, ptr addrspace(1) %ptr
+  ret <2 x float> %ld
+}
+
+define <2 x float> @call_nofpclass_funcs_v2f32(ptr addrspace(1) %ptr) {
+; CHECK-LABEL: call_nofpclass_funcs_v2f32:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT:    s_mov_b32 s18, s33
+; CHECK-NEXT:    s_mov_b32 s33, s32
+; CHECK-NEXT:    s_xor_saveexec_b64 s[16:17], -1
+; CHECK-NEXT:    buffer_store_dword v6, off, s[0:3], s33 ; 4-byte Folded Spill
+; CHECK-NEXT:    s_mov_b64 exec, s[16:17]
+; CHECK-NEXT:    s_addk_i32 s32, 0x400
+; CHECK-NEXT:    v_writelane_b32 v6, s30, 0
+; CHECK-NEXT:    s_getpc_b64 s[16:17]
+; CHECK-NEXT:    s_add_u32 s16, s16, func_v2f32@rel32@lo+4
+; CHECK-NEXT:    s_addc_u32 s17, s17, func_v2f32@rel32@hi+12
+; CHECK-NEXT:    v_writelane_b32 v6, s31, 1
+; CHECK-NEXT:    v_mov_b32_e32 v2, v1
+; CHECK-NEXT:    v_mov_b32_e32 v3, v0
+; CHECK-NEXT:    s_swappc_b64 s[30:31], s[16:17]
+; CHECK-NEXT:    v_mov_b32_e32 v4, v0
+; CHECK-NEXT:    v_mov_b32_e32 v5, v1
+; CHECK-NEXT:    v_mov_b32_e32 v0, v3
+; CHECK-NEXT:    v_mov_b32_e32 v1, v2
+; CHECK-NEXT:    s_swappc_b64 s[30:31], s[16:17]
+; CHECK-NEXT:    v_max_f32_e32 v2, v4, v4
+; CHECK-NEXT:    v_max_f32_e32 v0, v0, v0
+; CHECK-NEXT:    v_min_f32_e32 v0, v2, v0
+; CHECK-NEXT:    v_max_f32_e32 v2, v5, v5
+; CHECK-NEXT:    v_max_f32_e32 v1, v1, v1
+; CHECK-NEXT:    v_min_f32_e32 v1, v2, v1
+; CHECK-NEXT:    v_readlane_b32 s31, v6, 1
+; CHECK-NEXT:    v_readlane_b32 s30, v6, 0
+; CHECK-NEXT:    s_mov_b32 s32, s33
+; CHECK-NEXT:    s_xor_saveexec_b64 s[4:5], -1
+; CHECK-NEXT:    buffer_load_dword v6, off, s[0:3], s33 ; 4-byte Folded Reload
+; CHECK-NEXT:    s_mov_b64 exec, s[4:5]
+; CHECK-NEXT:    s_mov_b32 s33, s18
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    s_setpc_b64 s[30:31]
+  %call0 = call nofpclass(nan) <2 x float> @func_v2f32(ptr addrspace(1) %ptr)
+  %call1 = call nofpclass(nan) <2 x float> @func_v2f32(ptr addrspace(1) %ptr)
+  %min = call <2 x float> @llvm.minnum.v2f32(<2 x float> %call0, <2 x float> %call1)
+  ret <2 x float> %min
+}
+
+define internal double @func_f64(ptr addrspace(1) %ptr) {
+; CHECK-LABEL: func_f64:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off glc
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    s_setpc_b64 s[30:31]
+  %ld = load volatile double, ptr addrspace(1) %ptr
+  ret double %ld
+}
+
+define double @call_nofpclass_funcs_f64(ptr addrspace(1) %ptr) {
+; CHECK-LABEL: call_nofpclass_funcs_f64:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT:    s_mov_b32 s18, s33
+; CHECK-NEXT:    s_mov_b32 s33, s32
+; CHECK-NEXT:    s_xor_saveexec_b64 s[16:17], -1
+; CHECK-NEXT:    buffer_store_dword v6, off, s[0:3], s33 ; 4-byte Folded Spill
+; CHECK-NEXT:    s_mov_b64 exec, s[16:17]
+; CHECK-NEXT:    s_addk_i32 s32, 0x400
+; CHECK-NEXT:    v_writelane_b32 v6, s30, 0
+; CHECK-NEXT:    s_getpc_b64 s[16:17]
+; CHECK-NEXT:    s_add_u32 s16, s16, func_f64@rel32@lo+4
+; CHECK-NEXT:    s_addc_u32 s17, s17, func_f64@rel32@hi+12
+; CHECK-NEXT:    v_writelane_b32 v6, s31, 1
+; CHECK-NEXT:    v_mov_b32_e32 v4, v1
+; CHECK-NEXT:    v_mov_b32_e32 v5, v0
+; CHECK-NEXT:    s_swappc_b64 s[30:31], s[16:17]
+; CHECK-NEXT:    v_mov_b32_e32 v2, v0
+; CHECK-NEXT:    v_mov_b32_e32 v3, v1
+; CHECK-NEXT:    v_mov_b32_e32 v0, v5
+; CHECK-NEXT:    v_mov_b32_e32 v1, v4
+; CHECK-NEXT:    s_swappc_b64 s[30:31], s[16:17]
+; CHECK-NEXT:    v_max_f64 v[2:3], v[2:3], v[2:3]
+; CHECK-NEXT:    v_max_f64 v[0:1], v[0:1], v[0:1]
+; CHECK-NEXT:    v_readlane_b32 s31, v6, 1
+; CHECK-NEXT:    v_readlane_b32 s30, v6, 0
+; CHECK-NEXT:    s_mov_b32 s32, s33
+; CHECK-NEXT:    v_min_f64 v[0:1], v[2:3], v[0:1]
+; CHECK-NEXT:    s_xor_saveexec_b64 s[4:5], -1
+; CHECK-NEXT:    buffer_load_dword v6, off, s[0:3], s33 ; 4-byte Folded Reload
+; CHECK-NEXT:    s_mov_b64 exec, s[4:5]
+; CHECK-NEXT:    s_mov_b32 s33, s18
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    s_setpc_b64 s[30:31]
+  %call0 = call nofpclass(nan) double @func_f64(ptr addrspace(1) %ptr)
+  %call1 = call nofpclass(nan) double @func_f64(ptr addrspace(1) %ptr)
+  %min = call double @llvm.minnum.f64(double %call0, double %call1)
+  ret double %min
+}
+
+define float @call_nofpclass_intrinsic_f32(float %x, float %y, float %z) {
+; CHECK-LABEL: call_nofpclass_intrinsic_f32:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT:    v_sqrt_f32_e32 v0, v0
+; CHECK-NEXT:    v_sqrt_f32_e32 v1, v1
+; CHECK-NEXT:    v_cmp_lt_f32_e32 vcc, v0, v1
+; CHECK-NEXT:    v_cndmask_b32_e32 v0, v1, v0, vcc
+; CHECK-NEXT:    s_setpc_b64 s[30:31]
+  %call0 = call nofpclass(nan) float @llvm.amdgcn.sqrt.f32(float %x)
+  %call1 = call nofpclass(nan) float @llvm.amdgcn.sqrt.f32(float %y)
+  %lt = fcmp olt float %call0, %call1
+  %min = select nsz i1 %lt, float %call0, float %call1
+  ret float %min
+}
+
+define <2 x half> @call_nofpclass_intrinsic_v2f16(float %x, float %y, float %z, float %w) {
+; CHECK-LABEL: call_nofpclass_intrinsic_v2f16:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT:    v_cvt_pkrtz_f16_f32 v0, v0, v1
+; CHECK-NEXT:    v_cvt_pkrtz_f16_f32 v1, v2, v3
+; CHECK-NEXT:    v_lshrrev_b32_e32 v2, 16, v1
+; CHECK-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
+; CHECK-NEXT:    v_cmp_lt_f16_e32 vcc, v0, v1
+; CHECK-NEXT:    v_cndmask_b32_e32 v0, v1, v0, vcc
+; CHECK-NEXT:    v_cmp_lt_f16_e32 vcc, v3, v2
+; CHECK-NEXT:    v_cndmask_b32_e32 v1, v2, v3, vcc
+; CHECK-NEXT:    s_mov_b32 s4, 0x5040100
+; CHECK-NEXT:    v_perm_b32 v0, v1, v0, s4
+; CHECK-NEXT:    s_setpc_b64 s[30:31]
+  %call0 = call nofpclass(nan) <2 x half> @llvm.amdgcn.cvt.pkrtz(float %x, float %y)
+  %call1 = call nofpclass(nan) <2 x half> @llvm.amdgcn.cvt.pkrtz(float %z, float %w)
+  %lt = fcmp olt <2 x half> %call0, %call1
+  %min = select nsz <2 x i1> %lt, <2 x half> %call0, <2 x half> %call1
+  ret <2 x half> %min
+}

@arsenm arsenm marked this pull request as ready for review November 10, 2025 05:06
This was referenced Nov 10, 2025
Merged
Merged
RKSimon
RKSimon reviewed Nov 10, 2025
View reviewed changes
llvm/test/CodeGen/AMDGPU/nofpclass-call.ll
; CHECK-NEXT: global_load_dword v0, v[0:1], off glc
; CHECK-NEXT: s_waitcnt vmcnt(0)
; CHECK-NEXT: s_setpc_b64 s[30:31]
%ld = load volatile float, ptr addrspace(1) %ptr
Copy link
Collaborator

@RKSimon RKSimon Nov 10, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Whats the purpose of these volatile load tests?

Copy link
Contributor Author

@arsenm arsenm Nov 10, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Just provide a value. Attributor would be able to fold these into one value if it was a return of constant, though we happen to not run it in codegen anymore

@arsenm arsenm merged commit 726c049 into main Nov 10, 2025
14 checks passed
@arsenm arsenm deleted the users/arsenm/amdgpu/add-baseline-test-nofpclass-attr-call branch November 10, 2025 16:04
RKSimon
RKSimon reviewed Nov 10, 2025
View reviewed changes
Copy link
Collaborator

@RKSimon RKSimon left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM - cheers

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@RKSimon RKSimon RKSimon left review comments

@dtcxzyw dtcxzyw Awaiting requested review from dtcxzyw

@jcranmer-intel jcranmer-intel Awaiting requested review from jcranmer-intel

@ro-i ro-i Awaiting requested review from ro-i

@wzssyqa wzssyqa Awaiting requested review from wzssyqa

Assignees

No one assigned

Labels

backend:AMDGPU floating-point Floating-point math llvm:SelectionDAG SelectionDAGISel as well

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@arsenm @llvmbot @RKSimon