[llvm] [AMDGPU][GFX12] Add tests for {image/flat}_atomic_pk (PR #78683)
Mariusz Sikora via llvm-commits
llvm-commits at lists.llvm.org
Fri Jan 19 01:31:45 PST 2024
https://github.com/mariusz-sikora-at-amd updated https://github.com/llvm/llvm-project/pull/78683
>From 36acef44837d60d4c131fe770fc4b6b0ed92bcd6 Mon Sep 17 00:00:00 2001
From: Mariusz Sikora <mariusz.sikora at amd.com>
Date: Fri, 19 Jan 2024 10:31:14 +0100
Subject: [PATCH] [AMDGPU][GFX12] Add tests for flat_atomic_pk_fadd_16
---
llvm/test/MC/AMDGPU/gfx12_asm_vflat.s | 42 +++++++++++++++++++
.../Disassembler/AMDGPU/gfx12_dasm_vflat.txt | 42 +++++++++++++++++++
2 files changed, 84 insertions(+)
diff --git a/llvm/test/MC/AMDGPU/gfx12_asm_vflat.s b/llvm/test/MC/AMDGPU/gfx12_asm_vflat.s
index bac3655d19ec69..c814b42c127552 100644
--- a/llvm/test/MC/AMDGPU/gfx12_asm_vflat.s
+++ b/llvm/test/MC/AMDGPU/gfx12_asm_vflat.s
@@ -432,6 +432,48 @@ flat_atomic_xor_b64 v[1:2], v[0:1], v[2:3] offset:-64 th:TH_ATOMIC_RETURN
flat_atomic_xor_b64 v[1:2], v[0:1], v[2:3] offset:64 th:TH_ATOMIC_RETURN
// GFX12: encoding: [0x7c,0xc0,0x12,0xec,0x01,0x00,0x10,0x01,0x00,0x40,0x00,0x00]
+flat_atomic_pk_add_f16 v1, v[0:1], v2 offset:-8000000 th:TH_ATOMIC_RETURN
+// GFX12: encoding: [0x7c,0x40,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0x00,0xee,0x85]
+
+flat_atomic_pk_add_f16 v1, v[0:1], v2 offset:8000000 th:TH_ATOMIC_RETURN
+// GFX12: encoding: [0x7c,0x40,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0x00,0x12,0x7a]
+
+flat_atomic_pk_add_f16 v1, v[0:1], v2 offset:-64 th:TH_ATOMIC_RETURN
+// GFX12: encoding: [0x7c,0x40,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0xc0,0xff,0xff]
+
+flat_atomic_pk_add_f16 v1, v[0:1], v2 offset:64 th:TH_ATOMIC_RETURN
+// GFX12: encoding: [0x7c,0x40,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0x40,0x00,0x00]
+
+flat_atomic_pk_add_f16 v[0:1], v2 offset:-64
+// GFX12: encoding: [0x7c,0x40,0x16,0xec,0x00,0x00,0x00,0x01,0x00,0xc0,0xff,0xff]
+
+flat_atomic_pk_add_f16 v[0:1], v2 offset:64
+// GFX12: encoding: [0x7c,0x40,0x16,0xec,0x00,0x00,0x00,0x01,0x00,0x40,0x00,0x00]
+
+flat_atomic_pk_add_f16 v[0:1], v2
+// GFX12: encoding: [0x7c,0x40,0x16,0xec,0x00,0x00,0x00,0x01,0x00,0x00,0x00,0x00]
+
+flat_atomic_pk_add_bf16 v1, v[0:1], v2 offset:-8000000 th:TH_ATOMIC_RETURN
+// GFX12: encoding: [0x7c,0x80,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0x00,0xee,0x85]
+
+flat_atomic_pk_add_bf16 v1, v[0:1], v2 offset:8000000 th:TH_ATOMIC_RETURN
+// GFX12: encoding: [0x7c,0x80,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0x00,0x12,0x7a]
+
+flat_atomic_pk_add_bf16 v1, v[0:1], v2 offset:-64 th:TH_ATOMIC_RETURN
+// GFX12: encoding: [0x7c,0x80,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0xc0,0xff,0xff]
+
+flat_atomic_pk_add_bf16 v1, v[0:1], v2 offset:64 th:TH_ATOMIC_RETURN
+// GFX12: encoding: [0x7c,0x80,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0x40,0x00,0x00]
+
+flat_atomic_pk_add_bf16 v[0:1], v2 offset:-64
+// GFX12: encoding: [0x7c,0x80,0x16,0xec,0x00,0x00,0x00,0x01,0x00,0xc0,0xff,0xff]
+
+flat_atomic_pk_add_bf16 v[0:1], v2 offset:64
+// GFX12: encoding: [0x7c,0x80,0x16,0xec,0x00,0x00,0x00,0x01,0x00,0x40,0x00,0x00]
+
+flat_atomic_pk_add_bf16 v[0:1], v2
+// GFX12: encoding: [0x7c,0x80,0x16,0xec,0x00,0x00,0x00,0x01,0x00,0x00,0x00,0x00]
+
flat_load_b128 v[1:4], v[0:1] offset:-64
// GFX12: encoding: [0x7c,0xc0,0x05,0xec,0x01,0x00,0x00,0x00,0x00,0xc0,0xff,0xff]
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vflat.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vflat.txt
index b579297235d85e..181f2bd646069e 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vflat.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vflat.txt
@@ -240,6 +240,48 @@
# GFX12: flat_atomic_xor_b64 v[1:2], v[0:1], v[2:3] offset:64 th:TH_ATOMIC_RETURN ; encoding: [0x7c,0xc0,0x12,0xec,0x01,0x00,0x10,0x01,0x00,0x40,0x00,0x00]
0x7c,0xc0,0x12,0xec,0x01,0x00,0x10,0x01,0x00,0x40,0x00,0x00
+# GFX12: flat_atomic_pk_add_f16 v1, v[0:1], v2 offset:-8000000 th:TH_ATOMIC_RETURN ; encoding: [0x7c,0x40,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0x00,0xee,0x85]
+0x7c,0x40,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0x00,0xee,0x85
+
+# GFX12: flat_atomic_pk_add_f16 v1, v[0:1], v2 offset:8000000 th:TH_ATOMIC_RETURN ; encoding: [0x7c,0x40,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0x00,0x12,0x7a]
+0x7c,0x40,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0x00,0x12,0x7a
+
+# GFX12: flat_atomic_pk_add_f16 v1, v[0:1], v2 offset:-64 th:TH_ATOMIC_RETURN ; encoding: [0x7c,0x40,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0xc0,0xff,0xff]
+0x7c,0x40,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0xc0,0xff,0xff
+
+# GFX12: flat_atomic_pk_add_f16 v1, v[0:1], v2 offset:64 th:TH_ATOMIC_RETURN ; encoding: [0x7c,0x40,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0x40,0x00,0x00]
+0x7c,0x40,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0x40,0x00,0x00
+
+# GFX12: flat_atomic_pk_add_f16 v[0:1], v2 offset:-64 ; encoding: [0x7c,0x40,0x16,0xec,0x00,0x00,0x00,0x01,0x00,0xc0,0xff,0xff]
+0x7c,0x40,0x16,0xec,0x00,0x00,0x00,0x01,0x00,0xc0,0xff,0xff
+
+# GFX12: flat_atomic_pk_add_f16 v[0:1], v2 offset:64 ; encoding: [0x7c,0x40,0x16,0xec,0x00,0x00,0x00,0x01,0x00,0x40,0x00,0x00]
+0x7c,0x40,0x16,0xec,0x00,0x00,0x00,0x01,0x00,0x40,0x00,0x00
+
+# GFX12: flat_atomic_pk_add_f16 v[0:1], v2 ; encoding: [0x7c,0x40,0x16,0xec,0x00,0x00,0x00,0x01,0x00,0x00,0x00,0x00]
+0x7c,0x40,0x16,0xec,0x00,0x00,0x00,0x01,0x00,0x00,0x00,0x00
+
+# GFX12: flat_atomic_pk_add_bf16 v1, v[0:1], v2 offset:-8000000 th:TH_ATOMIC_RETURN ; encoding: [0x7c,0x80,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0x00,0xee,0x85]
+0x7c,0x80,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0x00,0xee,0x85
+
+# GFX12: flat_atomic_pk_add_bf16 v1, v[0:1], v2 offset:8000000 th:TH_ATOMIC_RETURN ; encoding: [0x7c,0x80,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0x00,0x12,0x7a]
+0x7c,0x80,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0x00,0x12,0x7a
+
+# GFX12: flat_atomic_pk_add_bf16 v1, v[0:1], v2 offset:-64 th:TH_ATOMIC_RETURN ; encoding: [0x7c,0x80,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0xc0,0xff,0xff]
+0x7c,0x80,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0xc0,0xff,0xff
+
+# GFX12: flat_atomic_pk_add_bf16 v1, v[0:1], v2 offset:64 th:TH_ATOMIC_RETURN ; encoding: [0x7c,0x80,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0x40,0x00,0x00]
+0x7c,0x80,0x16,0xec,0x01,0x00,0x10,0x01,0x00,0x40,0x00,0x00
+
+# GFX12: flat_atomic_pk_add_bf16 v[0:1], v2 offset:-64 ; encoding: [0x7c,0x80,0x16,0xec,0x00,0x00,0x00,0x01,0x00,0xc0,0xff,0xff]
+0x7c,0x80,0x16,0xec,0x00,0x00,0x00,0x01,0x00,0xc0,0xff,0xff
+
+# GFX12: flat_atomic_pk_add_bf16 v[0:1], v2 offset:64 ; encoding: [0x7c,0x80,0x16,0xec,0x00,0x00,0x00,0x01,0x00,0x40,0x00,0x00]
+0x7c,0x80,0x16,0xec,0x00,0x00,0x00,0x01,0x00,0x40,0x00,0x00
+
+# GFX12: flat_atomic_pk_add_bf16 v[0:1], v2 ; encoding: [0x7c,0x80,0x16,0xec,0x00,0x00,0x00,0x01,0x00,0x00,0x00,0x00]
+0x7c,0x80,0x16,0xec,0x00,0x00,0x00,0x01,0x00,0x00,0x00,0x00
+
# GFX12: flat_load_b128 v[1:4], v[0:1] offset:64 ; encoding: [0x7c,0xc0,0x05,0xec,0x01,0x00,0x00,0x00,0x00,0x40,0x00,0x00]
0x7c,0xc0,0x05,0xec,0x01,0x00,0x00,0x00,0x00,0x40,0x00,0x00
More information about the llvm-commits
mailing list