[llvm] AMDGPU: Add tests for minimum and maximum intrinsics (PR #90997)
via llvm-commits
llvm-commits at lists.llvm.org
Sat May 4 06:31:25 PDT 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-amdgpu
Author: Matt Arsenault (arsenm)
<details>
<summary>Changes</summary>
Baseline tests for new expansion. I think we can do better and avoid the classes.
---
Patch is 1.43 MiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/90997.diff
6 Files Affected:
- (added) llvm/test/CodeGen/AMDGPU/llvm.maximum.f16.ll (+4993)
- (added) llvm/test/CodeGen/AMDGPU/llvm.maximum.f32.ll (+4551)
- (added) llvm/test/CodeGen/AMDGPU/llvm.maximum.f64.ll (+6487)
- (added) llvm/test/CodeGen/AMDGPU/llvm.minimum.f16.ll (+3967)
- (added) llvm/test/CodeGen/AMDGPU/llvm.minimum.f32.ll (+4551)
- (added) llvm/test/CodeGen/AMDGPU/llvm.minimum.f64.ll (+6487)
``````````diff
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.maximum.f16.ll b/llvm/test/CodeGen/AMDGPU/llvm.maximum.f16.ll
new file mode 100644
index 00000000000000..c49e6a9a9f25cf
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/llvm.maximum.f16.ll
@@ -0,0 +1,4993 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx703 < %s | FileCheck -check-prefixes=GCN,GFX7 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx803 < %s | FileCheck -check-prefixes=GCN,GFX8 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx940 < %s | FileCheck -check-prefixes=GCN,GFX940 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1030 < %s | FileCheck -check-prefixes=GCN,GFX10 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 < %s | FileCheck -check-prefixes=GCN,GFX11 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GCN,GFX12 %s
+
+define half @v_maximum_f16(half %src0, half %src1) {
+; GFX7-LABEL: v_maximum_f16:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: v_cvt_f16_f32_e32 v1, v1
+; GFX7-NEXT: v_cvt_f16_f32_e32 v0, v0
+; GFX7-NEXT: v_mov_b32_e32 v2, 0x7fc00000
+; GFX7-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX7-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX7-NEXT: v_max_f32_e32 v3, v0, v1
+; GFX7-NEXT: v_cmp_o_f32_e32 vcc, v0, v1
+; GFX7-NEXT: v_cndmask_b32_e32 v2, v2, v3, vcc
+; GFX7-NEXT: v_cmp_class_f32_e64 vcc, v0, 64
+; GFX7-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX7-NEXT: v_cmp_class_f32_e64 vcc, v1, 64
+; GFX7-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX7-NEXT: v_cmp_eq_f32_e32 vcc, 0, v2
+; GFX7-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: v_maximum_f16:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX8-NEXT: v_mov_b32_e32 v3, 0x7e00
+; GFX8-NEXT: v_cmp_o_f16_e32 vcc, v0, v1
+; GFX8-NEXT: v_cndmask_b32_e32 v2, v3, v2, vcc
+; GFX8-NEXT: v_cmp_class_f16_e64 vcc, v0, 64
+; GFX8-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX8-NEXT: v_cmp_class_f16_e64 vcc, v1, 64
+; GFX8-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX8-NEXT: v_cmp_eq_f16_e32 vcc, 0, v2
+; GFX8-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_maximum_f16:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX9-NEXT: v_mov_b32_e32 v3, 0x7e00
+; GFX9-NEXT: v_cmp_o_f16_e32 vcc, v0, v1
+; GFX9-NEXT: v_cndmask_b32_e32 v2, v3, v2, vcc
+; GFX9-NEXT: v_cmp_class_f16_e64 vcc, v0, 64
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX9-NEXT: v_cmp_class_f16_e64 vcc, v1, 64
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX9-NEXT: v_cmp_eq_f16_e32 vcc, 0, v2
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX940-LABEL: v_maximum_f16:
+; GFX940: ; %bb.0:
+; GFX940-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX940-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX940-NEXT: v_mov_b32_e32 v3, 0x7e00
+; GFX940-NEXT: v_cmp_o_f16_e32 vcc, v0, v1
+; GFX940-NEXT: s_nop 1
+; GFX940-NEXT: v_cndmask_b32_e32 v2, v3, v2, vcc
+; GFX940-NEXT: v_cmp_class_f16_e64 vcc, v0, 64
+; GFX940-NEXT: s_nop 1
+; GFX940-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX940-NEXT: v_cmp_class_f16_e64 vcc, v1, 64
+; GFX940-NEXT: s_nop 1
+; GFX940-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX940-NEXT: v_cmp_eq_f16_e32 vcc, 0, v2
+; GFX940-NEXT: s_nop 1
+; GFX940-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX940-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_maximum_f16:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX10-NEXT: v_cmp_o_f16_e32 vcc_lo, v0, v1
+; GFX10-NEXT: v_cndmask_b32_e32 v2, 0x7e00, v2, vcc_lo
+; GFX10-NEXT: v_cmp_class_f16_e64 vcc_lo, v0, 64
+; GFX10-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo
+; GFX10-NEXT: v_cmp_class_f16_e64 vcc_lo, v1, 64
+; GFX10-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc_lo
+; GFX10-NEXT: v_cmp_eq_f16_e32 vcc_lo, 0, v2
+; GFX10-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_maximum_f16:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX11-NEXT: v_cmp_o_f16_e32 vcc_lo, v0, v1
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX11-NEXT: v_cndmask_b32_e32 v2, 0x7e00, v2, vcc_lo
+; GFX11-NEXT: v_cmp_class_f16_e64 vcc_lo, v0, 64
+; GFX11-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo
+; GFX11-NEXT: v_cmp_class_f16_e64 vcc_lo, v1, 64
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX11-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc_lo
+; GFX11-NEXT: v_cmp_eq_f16_e32 vcc_lo, 0, v2
+; GFX11-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_maximum_f16:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_maximum_f16 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %op = call half @llvm.maximum.f16(half %src0, half %src1)
+ ret half %op
+}
+
+define half @v_maximum_f16__nnan(half %src0, half %src1) {
+; GFX7-LABEL: v_maximum_f16__nnan:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: v_cvt_f16_f32_e32 v1, v1
+; GFX7-NEXT: v_cvt_f16_f32_e32 v0, v0
+; GFX7-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX7-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX7-NEXT: v_max_f32_e32 v2, v0, v1
+; GFX7-NEXT: v_cmp_class_f32_e64 vcc, v0, 64
+; GFX7-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX7-NEXT: v_cmp_class_f32_e64 vcc, v1, 64
+; GFX7-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX7-NEXT: v_cmp_eq_f32_e32 vcc, 0, v2
+; GFX7-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: v_maximum_f16__nnan:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX8-NEXT: v_cmp_class_f16_e64 vcc, v0, 64
+; GFX8-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX8-NEXT: v_cmp_class_f16_e64 vcc, v1, 64
+; GFX8-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX8-NEXT: v_cmp_eq_f16_e32 vcc, 0, v2
+; GFX8-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_maximum_f16__nnan:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX9-NEXT: v_cmp_class_f16_e64 vcc, v0, 64
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX9-NEXT: v_cmp_class_f16_e64 vcc, v1, 64
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX9-NEXT: v_cmp_eq_f16_e32 vcc, 0, v2
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX940-LABEL: v_maximum_f16__nnan:
+; GFX940: ; %bb.0:
+; GFX940-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX940-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX940-NEXT: v_cmp_class_f16_e64 vcc, v0, 64
+; GFX940-NEXT: s_nop 1
+; GFX940-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX940-NEXT: v_cmp_class_f16_e64 vcc, v1, 64
+; GFX940-NEXT: s_nop 1
+; GFX940-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX940-NEXT: v_cmp_eq_f16_e32 vcc, 0, v2
+; GFX940-NEXT: s_nop 1
+; GFX940-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX940-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_maximum_f16__nnan:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX10-NEXT: v_cmp_class_f16_e64 vcc_lo, v0, 64
+; GFX10-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo
+; GFX10-NEXT: v_cmp_class_f16_e64 vcc_lo, v1, 64
+; GFX10-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc_lo
+; GFX10-NEXT: v_cmp_eq_f16_e32 vcc_lo, 0, v2
+; GFX10-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_maximum_f16__nnan:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX11-NEXT: v_cmp_class_f16_e64 vcc_lo, v0, 64
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX11-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo
+; GFX11-NEXT: v_cmp_class_f16_e64 vcc_lo, v1, 64
+; GFX11-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc_lo
+; GFX11-NEXT: v_cmp_eq_f16_e32 vcc_lo, 0, v2
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX11-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_maximum_f16__nnan:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_maximum_f16 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %op = call nnan half @llvm.maximum.f16(half %src0, half %src1)
+ ret half %op
+}
+
+define half @v_maximum_f16__nsz(half %src0, half %src1) {
+; GFX7-LABEL: v_maximum_f16__nsz:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: v_cvt_f16_f32_e32 v1, v1
+; GFX7-NEXT: v_cvt_f16_f32_e32 v0, v0
+; GFX7-NEXT: v_mov_b32_e32 v2, 0x7fc00000
+; GFX7-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX7-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX7-NEXT: v_max_f32_e32 v3, v0, v1
+; GFX7-NEXT: v_cmp_o_f32_e32 vcc, v0, v1
+; GFX7-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: v_maximum_f16__nsz:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX8-NEXT: v_mov_b32_e32 v3, 0x7e00
+; GFX8-NEXT: v_cmp_o_f16_e32 vcc, v0, v1
+; GFX8-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_maximum_f16__nsz:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX9-NEXT: v_mov_b32_e32 v3, 0x7e00
+; GFX9-NEXT: v_cmp_o_f16_e32 vcc, v0, v1
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX940-LABEL: v_maximum_f16__nsz:
+; GFX940: ; %bb.0:
+; GFX940-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX940-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX940-NEXT: v_mov_b32_e32 v3, 0x7e00
+; GFX940-NEXT: v_cmp_o_f16_e32 vcc, v0, v1
+; GFX940-NEXT: s_nop 1
+; GFX940-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc
+; GFX940-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_maximum_f16__nsz:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX10-NEXT: v_cmp_o_f16_e32 vcc_lo, v0, v1
+; GFX10-NEXT: v_cndmask_b32_e32 v0, 0x7e00, v2, vcc_lo
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_maximum_f16__nsz:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX11-NEXT: v_cmp_o_f16_e32 vcc_lo, v0, v1
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX11-NEXT: v_cndmask_b32_e32 v0, 0x7e00, v2, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_maximum_f16__nsz:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_maximum_f16 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %op = call nsz half @llvm.maximum.f16(half %src0, half %src1)
+ ret half %op
+}
+
+define half @v_maximum_f16__nnan_nsz(half %src0, half %src1) {
+; GFX7-LABEL: v_maximum_f16__nnan_nsz:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: v_cvt_f16_f32_e32 v1, v1
+; GFX7-NEXT: v_cvt_f16_f32_e32 v0, v0
+; GFX7-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX7-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX7-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: v_maximum_f16__nnan_nsz:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_max_f16_e32 v0, v0, v1
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_maximum_f16__nnan_nsz:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_max_f16_e32 v0, v0, v1
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX940-LABEL: v_maximum_f16__nnan_nsz:
+; GFX940: ; %bb.0:
+; GFX940-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX940-NEXT: v_max_f16_e32 v0, v0, v1
+; GFX940-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_maximum_f16__nnan_nsz:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_max_f16_e32 v0, v0, v1
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_maximum_f16__nnan_nsz:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_max_f16_e32 v0, v0, v1
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_maximum_f16__nnan_nsz:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_maximum_f16 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %op = call nnan nsz half @llvm.maximum.f16(half %src0, half %src1)
+ ret half %op
+}
+
+define half @v_maximum_f16__nnan_src0(half %arg0, half %src1) {
+; GFX7-LABEL: v_maximum_f16__nnan_src0:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: v_cvt_f16_f32_e32 v0, v0
+; GFX7-NEXT: v_cvt_f16_f32_e32 v1, v1
+; GFX7-NEXT: v_mov_b32_e32 v2, 0x7fc00000
+; GFX7-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX7-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX7-NEXT: v_add_f32_e32 v0, 1.0, v0
+; GFX7-NEXT: v_max_f32_e32 v3, v0, v1
+; GFX7-NEXT: v_cmp_o_f32_e32 vcc, v0, v1
+; GFX7-NEXT: v_cndmask_b32_e32 v2, v2, v3, vcc
+; GFX7-NEXT: v_cmp_class_f32_e64 vcc, v0, 64
+; GFX7-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX7-NEXT: v_cmp_class_f32_e64 vcc, v1, 64
+; GFX7-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX7-NEXT: v_cmp_eq_f32_e32 vcc, 0, v2
+; GFX7-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: v_maximum_f16__nnan_src0:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_add_f16_e32 v0, 1.0, v0
+; GFX8-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX8-NEXT: v_mov_b32_e32 v3, 0x7e00
+; GFX8-NEXT: v_cmp_o_f16_e32 vcc, v0, v1
+; GFX8-NEXT: v_cndmask_b32_e32 v2, v3, v2, vcc
+; GFX8-NEXT: v_cmp_class_f16_e64 vcc, v0, 64
+; GFX8-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX8-NEXT: v_cmp_class_f16_e64 vcc, v1, 64
+; GFX8-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX8-NEXT: v_cmp_eq_f16_e32 vcc, 0, v2
+; GFX8-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_maximum_f16__nnan_src0:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_f16_e32 v0, 1.0, v0
+; GFX9-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX9-NEXT: v_mov_b32_e32 v3, 0x7e00
+; GFX9-NEXT: v_cmp_o_f16_e32 vcc, v0, v1
+; GFX9-NEXT: v_cndmask_b32_e32 v2, v3, v2, vcc
+; GFX9-NEXT: v_cmp_class_f16_e64 vcc, v0, 64
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX9-NEXT: v_cmp_class_f16_e64 vcc, v1, 64
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX9-NEXT: v_cmp_eq_f16_e32 vcc, 0, v2
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX940-LABEL: v_maximum_f16__nnan_src0:
+; GFX940: ; %bb.0:
+; GFX940-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX940-NEXT: v_add_f16_e32 v0, 1.0, v0
+; GFX940-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX940-NEXT: v_mov_b32_e32 v3, 0x7e00
+; GFX940-NEXT: v_cmp_o_f16_e32 vcc, v0, v1
+; GFX940-NEXT: s_nop 1
+; GFX940-NEXT: v_cndmask_b32_e32 v2, v3, v2, vcc
+; GFX940-NEXT: v_cmp_class_f16_e64 vcc, v0, 64
+; GFX940-NEXT: s_nop 1
+; GFX940-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX940-NEXT: v_cmp_class_f16_e64 vcc, v1, 64
+; GFX940-NEXT: s_nop 1
+; GFX940-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX940-NEXT: v_cmp_eq_f16_e32 vcc, 0, v2
+; GFX940-NEXT: s_nop 1
+; GFX940-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX940-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_maximum_f16__nnan_src0:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_add_f16_e32 v0, 1.0, v0
+; GFX10-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX10-NEXT: v_cmp_o_f16_e32 vcc_lo, v0, v1
+; GFX10-NEXT: v_cndmask_b32_e32 v2, 0x7e00, v2, vcc_lo
+; GFX10-NEXT: v_cmp_class_f16_e64 vcc_lo, v0, 64
+; GFX10-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo
+; GFX10-NEXT: v_cmp_class_f16_e64 vcc_lo, v1, 64
+; GFX10-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc_lo
+; GFX10-NEXT: v_cmp_eq_f16_e32 vcc_lo, 0, v2
+; GFX10-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_maximum_f16__nnan_src0:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_f16_e32 v0, 1.0, v0
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX11-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX11-NEXT: v_cmp_o_f16_e32 vcc_lo, v0, v1
+; GFX11-NEXT: v_cndmask_b32_e32 v2, 0x7e00, v2, vcc_lo
+; GFX11-NEXT: v_cmp_class_f16_e64 vcc_lo, v0, 64
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX11-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo
+; GFX11-NEXT: v_cmp_class_f16_e64 vcc_lo, v1, 64
+; GFX11-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc_lo
+; GFX11-NEXT: v_cmp_eq_f16_e32 vcc_lo, 0, v2
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX11-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_maximum_f16__nnan_src0:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_add_f16_e32 v0, 1.0, v0
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX12-NEXT: v_maximum_f16 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %src0 = fadd nnan half %arg0, 1.0
+ %op = call half @llvm.maximum.f16(half %src0, half %src1)
+ ret half %op
+}
+
+define half @v_maximum_f16__nnan_src1(half %src0, half %arg1) {
+; GFX7-LABEL: v_maximum_f16__nnan_src1:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: v_cvt_f16_f32_e32 v1, v1
+; GFX7-NEXT: v_cvt_f16_f32_e32 v0, v0
+; GFX7-NEXT: v_mov_b32_e32 v2, 0x7fc00000
+; GFX7-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX7-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX7-NEXT: v_add_f32_e32 v1, 1.0, v1
+; GFX7-NEXT: v_max_f32_e32 v3, v0, v1
+; GFX7-NEXT: v_cmp_o_f32_e32 vcc, v0, v1
+; GFX7-NEXT: v_cndmask_b32_e32 v2, v2, v3, vcc
+; GFX7-NEXT: v_cmp_class_f32_e64 vcc, v0, 64
+; GFX7-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX7-NEXT: v_cmp_class_f32_e64 vcc, v1, 64
+; GFX7-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX7-NEXT: v_cmp_eq_f32_e32 vcc, 0, v2
+; GFX7-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: v_maximum_f16__nna...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/90997
More information about the llvm-commits
mailing list