Replace gfx940 and gfx941 with gfx942 downstream (llvm#696)

ronlieb · web-flow · commit 537e0b63a40c · 2025-02-18T08:49:24.000-05:00
diff --git a/amd/comgr/docs/ReleaseNotes.md b/amd/comgr/docs/ReleaseNotes.md
@@ -218,8 +218,6 @@ now-removed Comgr actions, such as \*COMPILE\_SOURCE\_TO\_FATBIN.
 
 New Targets
 -----------
- - gfx940
- - gfx941
  - gfx942
  - gfx950
  - gfx1036
diff --git a/amd/comgr/src/comgr-isa-metadata.def b/amd/comgr/src/comgr-isa-metadata.def
@@ -69,8 +69,6 @@ HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx908",            true,  true, EF_AMDGPU_MAC
 HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx909",           false,  true, EF_AMDGPU_MACH_AMDGCN_GFX909,          true, 65536,  32,  4,   40, 1024,   16, 800, 102,    4, 256, 256)
 HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx90a",            true,  true, EF_AMDGPU_MACH_AMDGCN_GFX90A,          true, 65536,  32,  4,   40, 1024,   16, 800, 102,    4, 256, 256)
 HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx90c",           false,  true, EF_AMDGPU_MACH_AMDGCN_GFX90C,          true, 65536,  32,  4,   40, 1024,   16, 800, 102,    4, 256, 256)
-HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx940",            true,  true, EF_AMDGPU_MACH_AMDGCN_GFX940,          true, 65536,  32,  4,   40, 1024,   16, 800, 102,    4, 256, 256)
-HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx941",            true,  true, EF_AMDGPU_MACH_AMDGCN_GFX941,          true, 65536,  32,  4,   40, 1024,   16, 800, 102,    4, 256, 256)
 HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx942",            true,  true, EF_AMDGPU_MACH_AMDGCN_GFX942,          true, 65536,  32,  4,   40, 1024,   16, 800, 102,    4, 256, 256)
 HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx950",            true,  true, EF_AMDGPU_MACH_AMDGCN_GFX950,          true, 65536,  32,  4,   40, 1024,   16, 800, 102,    4, 256, 256)
 HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx1010",          false,  true, EF_AMDGPU_MACH_AMDGCN_GFX1010,         true, 65536,  32,  4,   40, 1024,  106, 800, 106,    8, 256, 256)
diff --git a/amd/comgr/test/get_data_isa_name_test.c b/amd/comgr/test/get_data_isa_name_test.c
@@ -86,8 +86,6 @@ static isa_features_t IsaFeatures[] = {
   {"amdgcn-amd-amdhsa--gfx909",          false,     false,     none,       true,      none,       false},
   {"amdgcn-amd-amdhsa--gfx90a",          false,     true,      none,       true,      none,       false},
   {"amdgcn-amd-amdhsa--gfx90c",          true,      false,     none,       true,      off,        false},
-  {"amdgcn-amd-amdhsa--gfx940",          false,     true,      none,       true,      none,       false},
-  {"amdgcn-amd-amdhsa--gfx941",          false,     true,      none,       true,      none,       false},
   {"amdgcn-amd-amdhsa--gfx942",          false,     true,      none,       true,      none,       false},
   {"amdgcn-amd-amdhsa--gfx950",          false,     true,      none,       true,      none,       false},
   {"amdgcn-amd-amdhsa--gfx1010",         false,     false,     none,       true,      none,       false},
diff --git a/amd/device-libs/ockl/src/cg.cl b/amd/device-libs/ockl/src/cg.cl
@@ -11,7 +11,7 @@
 #define AL(P,S) __opencl_atomic_load((__global atomic_uint *)P, memory_order_relaxed, S)
 #define AA(P,V,S) __opencl_atomic_fetch_add((__global atomic_uint *)P, V, memory_order_relaxed, S)
 
-#define AVOID_GWS() (__oclc_ISA_version == 9400 || __oclc_ISA_version == 9401 || __oclc_ISA_version == 9402 || __oclc_ISA_version == 9500 || __oclc_ISA_version >= 11000)
+#define AVOID_GWS() (__oclc_ISA_version == 9402 || __oclc_ISA_version == 9500 || __oclc_ISA_version >= 11000)
 
 // XXX do not change these two structs without changing the language runtime
 struct mg_sync {
diff --git a/amd/device-libs/ockl/src/dm.cl b/amd/device-libs/ockl/src/dm.cl
@@ -180,7 +180,7 @@ typedef struct heap_s {
 #define AFO(P, V, O) __opencl_atomic_fetch_or (P, V, O, memory_scope_device)
 #define ACE(P, E, V, O) __opencl_atomic_compare_exchange_strong(P, E, V, O, O, memory_scope_device)
 
-#define NEED_RELEASE __oclc_ISA_version >= 9400 && __oclc_ISA_version < 10000
+#define NEED_RELEASE __oclc_ISA_version >= 9402 && __oclc_ISA_version < 10000
 
 // get the heap pointer
 static __global heap_t *
diff --git a/amd/device-libs/oclc/src/isa_version_940.cl b/amd/device-libs/oclc/src/isa_version_940.cl
diff --git a/amd/device-libs/oclc/src/isa_version_941.cl b/amd/device-libs/oclc/src/isa_version_941.cl
diff --git a/llvm/test/CodeGen/AMDGPU/force-store-sc0-sc1.ll b/llvm/test/CodeGen/AMDGPU/force-store-sc0-sc1.ll
diff --git a/llvm/test/DebugInfo/AMDGPU/heterogeneous-dwarf.ll b/llvm/test/DebugInfo/AMDGPU/heterogeneous-dwarf.ll
@@ -1,5 +1,5 @@
 ; RUN: llc -O0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx906 -verify-machineinstrs -filetype=obj -emit-heterogeneous-dwarf-as-user-ops < %s | llvm-dwarfdump -v -debug-info - | FileCheck --check-prefixes=COMMON,FLAT-SCR-DIS %s
-; RUN: llc -O0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx940 -verify-machineinstrs -filetype=obj -emit-heterogeneous-dwarf-as-user-ops < %s | llvm-dwarfdump -v -debug-info - | FileCheck --check-prefixes=COMMON,FLAT-SCR-ENA %s
+; RUN: llc -O0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 -verify-machineinstrs -filetype=obj -emit-heterogeneous-dwarf-as-user-ops < %s | llvm-dwarfdump -v -debug-info - | FileCheck --check-prefixes=COMMON,FLAT-SCR-ENA %s
 
 source_filename = "heterogeneous-dwarf.cl"
 target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5-G1-ni:7"

-Original file line number
+Diff line change
 New Targets
 -----------
 - - gfx940
 - - gfx941
  - gfx942
  - gfx950
  - gfx1036