[llvm] [AMDGPU] gfx1250 runlines for global-atomicrmw-fadd.ll. NFC (PR #159817)
via llvm-commits
llvm-commits at lists.llvm.org
Fri Sep 19 10:25:08 PDT 2025
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-amdgpu
Author: Stanislav Mekhanoshin (rampitec)
<details>
<summary>Changes</summary>
---
Patch is 174.36 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/159817.diff
1 Files Affected:
- (modified) llvm/test/CodeGen/AMDGPU/global-atomicrmw-fadd.ll (+2582)
``````````diff
diff --git a/llvm/test/CodeGen/AMDGPU/global-atomicrmw-fadd.ll b/llvm/test/CodeGen/AMDGPU/global-atomicrmw-fadd.ll
index 2cad8eeea33cf..b67a1c513c49f 100644
--- a/llvm/test/CodeGen/AMDGPU/global-atomicrmw-fadd.ll
+++ b/llvm/test/CodeGen/AMDGPU/global-atomicrmw-fadd.ll
@@ -1,4 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1250 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-TRUE16 %s
+; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1250 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-FAKE16 %s
; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-TRUE16 %s
; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-FAKE16 %s
; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx942 < %s | FileCheck -check-prefix=GFX942 %s
@@ -16,6 +18,18 @@
; --------------------------------------------------------------------
define float @global_agent_atomic_fadd_ret_f32__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_agent_atomic_fadd_ret_f32__amdgpu_no_fine_grained_memory:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v0, v[0:1], v2, off th:TH_ATOMIC_RETURN scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_agent_atomic_fadd_ret_f32__amdgpu_no_fine_grained_memory:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -201,6 +215,18 @@ define float @global_agent_atomic_fadd_ret_f32__amdgpu_no_fine_grained_memory(pt
}
define float @global_agent_atomic_fadd_ret_f32__offset12b_pos__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_agent_atomic_fadd_ret_f32__offset12b_pos__amdgpu_no_fine_grained_memory:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v0, v[0:1], v2, off offset:2044 th:TH_ATOMIC_RETURN scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_agent_atomic_fadd_ret_f32__offset12b_pos__amdgpu_no_fine_grained_memory:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -388,6 +414,18 @@ define float @global_agent_atomic_fadd_ret_f32__offset12b_pos__amdgpu_no_fine_gr
}
define float @global_agent_atomic_fadd_ret_f32__offset12b_neg__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_agent_atomic_fadd_ret_f32__offset12b_neg__amdgpu_no_fine_grained_memory:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v0, v[0:1], v2, off offset:-2048 th:TH_ATOMIC_RETURN scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_agent_atomic_fadd_ret_f32__offset12b_neg__amdgpu_no_fine_grained_memory:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -585,6 +623,18 @@ define float @global_agent_atomic_fadd_ret_f32__offset12b_neg__amdgpu_no_fine_gr
}
define void @global_agent_atomic_fadd_noret_f32__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_agent_atomic_fadd_noret_f32__amdgpu_no_fine_grained_memory:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v[0:1], v2, off scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_agent_atomic_fadd_noret_f32__amdgpu_no_fine_grained_memory:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -764,6 +814,18 @@ define void @global_agent_atomic_fadd_noret_f32__amdgpu_no_fine_grained_memory(p
}
define void @global_agent_atomic_fadd_noret_f32__offset12b_pos__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_agent_atomic_fadd_noret_f32__offset12b_pos__amdgpu_no_fine_grained_memory:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v[0:1], v2, off offset:2044 scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_agent_atomic_fadd_noret_f32__offset12b_pos__amdgpu_no_fine_grained_memory:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -946,6 +1008,18 @@ define void @global_agent_atomic_fadd_noret_f32__offset12b_pos__amdgpu_no_fine_g
}
define void @global_agent_atomic_fadd_noret_f32__offset12b_neg__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_agent_atomic_fadd_noret_f32__offset12b_neg__amdgpu_no_fine_grained_memory:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v[0:1], v2, off offset:-2048 scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_agent_atomic_fadd_noret_f32__offset12b_neg__amdgpu_no_fine_grained_memory:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -1136,6 +1210,18 @@ define void @global_agent_atomic_fadd_noret_f32__offset12b_neg__amdgpu_no_fine_g
}
define float @global_system_atomic_fadd_ret_f32__offset12b_pos__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_system_atomic_fadd_ret_f32__offset12b_pos__amdgpu_no_fine_grained_memory:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_SYS
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v0, v[0:1], v2, off offset:2044 th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_SYS
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_system_atomic_fadd_ret_f32__offset12b_pos__amdgpu_no_fine_grained_memory:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -1326,6 +1412,18 @@ define float @global_system_atomic_fadd_ret_f32__offset12b_pos__amdgpu_no_fine_g
}
define void @global_system_atomic_fadd_noret_f32__offset12b_pos__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_system_atomic_fadd_noret_f32__offset12b_pos__amdgpu_no_fine_grained_memory:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_SYS
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v[0:1], v2, off offset:2044 scope:SCOPE_SYS
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_SYS
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_system_atomic_fadd_noret_f32__offset12b_pos__amdgpu_no_fine_grained_memory:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -1511,6 +1609,18 @@ define void @global_system_atomic_fadd_noret_f32__offset12b_pos__amdgpu_no_fine_
}
define float @global_agent_atomic_fadd_ret_f32_maybe_remote(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_agent_atomic_fadd_ret_f32_maybe_remote:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v0, v[0:1], v2, off offset:2044 th:TH_ATOMIC_RETURN scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_agent_atomic_fadd_ret_f32_maybe_remote:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -1714,6 +1824,18 @@ define float @global_agent_atomic_fadd_ret_f32_maybe_remote(ptr addrspace(1) %pt
}
define float @global_agent_atomic_fadd_ret_f32_maybe_remote__amdgpu_ignore_denormal_mode(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_agent_atomic_fadd_ret_f32_maybe_remote__amdgpu_ignore_denormal_mode:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v0, v[0:1], v2, off offset:2044 th:TH_ATOMIC_RETURN scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_agent_atomic_fadd_ret_f32_maybe_remote__amdgpu_ignore_denormal_mode:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -1917,6 +2039,18 @@ define float @global_agent_atomic_fadd_ret_f32_maybe_remote__amdgpu_ignore_denor
}
define void @global_agent_atomic_fadd_noret_f32_maybe_remote__amdgpu_ignore_denormal_mode(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_agent_atomic_fadd_noret_f32_maybe_remote__amdgpu_ignore_denormal_mode:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v[0:1], v2, off offset:2044 scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_agent_atomic_fadd_noret_f32_maybe_remote__amdgpu_ignore_denormal_mode:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -2113,6 +2247,18 @@ define void @global_agent_atomic_fadd_noret_f32_maybe_remote__amdgpu_ignore_deno
}
define float @global_agent_atomic_fadd_ret_f32___amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_agent_atomic_fadd_ret_f32___amdgpu_no_fine_grained_memory:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v0, v[0:1], v2, off offset:2044 th:TH_ATOMIC_RETURN scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_agent_atomic_fadd_ret_f32___amdgpu_no_fine_grained_memory:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -2300,6 +2446,18 @@ define float @global_agent_atomic_fadd_ret_f32___amdgpu_no_fine_grained_memory(p
}
define float @global_agent_atomic_fadd_ret_f32___amdgpu_no_fine_grained_memory__amdgpu_ignore_denormal_mode(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_agent_atomic_fadd_ret_f32___amdgpu_no_fine_grained_memory__amdgpu_ignore_denormal_mode:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v0, v[0:1], v2, off offset:2044 th:TH_ATOMIC_RETURN scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_agent_atomic_fadd_ret_f32___amdgpu_no_fine_grained_memory__amdgpu_ignore_denormal_mode:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -2473,6 +2631,18 @@ define float @global_agent_atomic_fadd_ret_f32___amdgpu_no_fine_grained_memory__
}
define float @global_agent_atomic_fadd_ret_f32_amdgpu_ignore_denormal_mode(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_agent_atomic_fadd_ret_f32_amdgpu_ignore_denormal_mode:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v0, v[0:1], v2, off offset:2044 th:TH_ATOMIC_RETURN scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_agent_atomic_fadd_ret_f32_amdgpu_ignore_denormal_mode:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -2676,6 +2846,18 @@ define float @global_agent_atomic_fadd_ret_f32_amdgpu_ignore_denormal_mode(ptr a
}
define void @global_agent_atomic_fadd_noret_f32_maybe_remote(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_agent_atomic_fadd_noret_f32_maybe_remote:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v[0:1], v2, off offset:2044 scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_agent_atomic_fadd_noret_f32_maybe_remote:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -2872,6 +3054,18 @@ define void @global_agent_atomic_fadd_noret_f32_maybe_remote(ptr addrspace(1) %p
}
define void @global_agent_atomic_fadd_noret_f32___amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_agent_atomic_fadd_noret_f32___amdgpu_no_fine_grained_memory:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v[0:1], v2, off offset:2044 scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_agent_atomic_fadd_noret_f32___amdgpu_no_fine_grained_memory:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -3054,6 +3248,18 @@ define void @global_agent_atomic_fadd_noret_f32___amdgpu_no_fine_grained_memory(
}
define void @global_agent_atomic_fadd_noret_f32___amdgpu_no_fine_grained_memory__amdgpu_ignore_denormal_mode(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_agent_atomic_fadd_noret_f32___amdgpu_no_fine_grained_memory__amdgpu_ignore_denormal_mode:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v[0:1], v2, off offset:2044 scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_agent_atomic_fadd_noret_f32___amdgpu_no_fine_grained_memory__amdgpu_ignore_denormal_mode:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -3210,6 +3416,18 @@ define void @global_agent_atomic_fadd_noret_f32___amdgpu_no_fine_grained_memory_
}
define void @global_agent_atomic_fadd_noret_f32_amdgpu_ignore_denormal_mode(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_agent_atomic_fadd_noret_f32_amdgpu_ignore_denormal_mode:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v[0:1], v2, off offset:2044 scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_agent_atomic_fadd_noret_f32_amdgpu_ignore_denormal_mode:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -3406,6 +3624,18 @@ define void @global_agent_atomic_fadd_noret_f32_amdgpu_ignore_denormal_mode(ptr
}
define float @global_agent_atomic_fadd_ret_f32__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_agent_atomic_fadd_ret_f32__amdgpu_no_remote_memory:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v0, v[0:1], v2, off th:TH_ATOMIC_RETURN scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_agent_atomic_fadd_ret_f32__amdgpu_no_remote_memory:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -3607,6 +3837,18 @@ define float @global_agent_atomic_fadd_ret_f32__amdgpu_no_remote_memory(ptr addr
}
define void @global_agent_atomic_fadd_noret_f32__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_agent_atomic_fadd_noret_f32__amdgpu_no_remote_memory:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v[0:1], v2, off scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_inv scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_loadcnt 0x0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+;
; GFX12-LABEL: global_agent_atomic_fadd_noret_f32__amdgpu_no_remote_memory:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
@@ -3800,6 +4042,18 @@ define void @global_agent_atomic_fadd_noret_f32__amdgpu_no_remote_memory(ptr add
}
define float @global_agent_atomic_fadd_ret_f32__amdgpu_no_remote_memory__amdgpu_ignore_denormal_mode(ptr addrspace(1) %ptr, float %val) #0 {
+; GFX1250-LABEL: global_agent_atomic_fadd_ret_f32__amdgpu_no_remote_memory__amdgpu_ignore_denormal_mode:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: global_wb scope:SCOPE_DEV
+; GFX1250-NEXT: s_wait_storecnt 0x0
+; GFX1250-NEXT: global_atomic_add_f32 v0, v[0:1], v2, off th:TH_ATOMIC_RETURN scope:SCOPE_DEV
+; GFX1250-NEXT: ...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/159817
More information about the llvm-commits
mailing list