ac/llvm: add better code for fsign

[mesa.git] / src / amd / llvm / ac_llvm_util.c

diff --git a/src/amd/llvm/ac_llvm_util.c b/src/amd/llvm/ac_llvm_util.c
index ddc8fee839be9a8934fa3ccbc9a302736e259d81..c7c8b991756ef5c882a44e6c1848ca4ed061e56c 100644 (file)
--- a/src/amd/llvm/ac_llvm_util.c
+++ b/src/amd/llvm/ac_llvm_util.c
@@ -69,16 +69,36 @@ static void ac_init_llvm_target()
 #if LLVM_VERSION_MAJOR >= 10
                /* Atomic optimizations require LLVM 10.0 for gfx10 support. */
                "-amdgpu-atomic-optimizations=true",
+#endif
+#if LLVM_VERSION_MAJOR >= 11
+               /* This was disabled by default in: https://reviews.llvm.org/D77228 */
+               "-structurizecfg-skip-uniform-regions",
 #endif
        };
        LLVMParseCommandLineOptions(ARRAY_SIZE(argv), argv, NULL);
 }
 
-static once_flag ac_init_llvm_target_once_flag = ONCE_FLAG_INIT;
+PUBLIC void ac_init_shared_llvm_once(void)
+{
+       static once_flag ac_init_llvm_target_once_flag = ONCE_FLAG_INIT;
+       call_once(&ac_init_llvm_target_once_flag, ac_init_llvm_target);
+}
+
+#if !LLVM_IS_SHARED
+static once_flag ac_init_static_llvm_target_once_flag = ONCE_FLAG_INIT;
+static void ac_init_static_llvm_once(void)
+{
+       call_once(&ac_init_static_llvm_target_once_flag, ac_init_llvm_target);
+}
+#endif
 
 void ac_init_llvm_once(void)
 {
-       call_once(&ac_init_llvm_target_once_flag, ac_init_llvm_target);
+#if LLVM_IS_SHARED
+       ac_init_shared_llvm_once();
+#else
+       ac_init_static_llvm_once();
+#endif
 }
 
 static LLVMTargetRef ac_get_llvm_target(const char *triple)
@@ -153,6 +173,9 @@ const char *ac_get_llvm_processor_name(enum radeon_family family)
                return "gfx1011";
        case CHIP_NAVI14:
                return "gfx1012";
+       case CHIP_SIENNA_CICHLID:
+       case CHIP_NAVY_FLOUNDER:
+               return "gfx1030";
        default:
                return "";
        }
@@ -169,14 +192,13 @@ static LLVMTargetMachineRef ac_create_target_machine(enum radeon_family family,
        LLVMTargetRef target = ac_get_llvm_target(triple);
 
        snprintf(features, sizeof(features),
-                "+DumpCode,-fp32-denormals,+fp64-denormals%s%s%s%s%s%s",
+                "+DumpCode%s%s%s%s%s",
+                LLVM_VERSION_MAJOR >= 11 ? "" : ",-fp32-denormals,+fp64-denormals",
                 family >= CHIP_NAVI10 && !(tm_options & AC_TM_WAVE32) ?
                         ",+wavefrontsize64,-wavefrontsize32" : "",
-                tm_options & AC_TM_SISCHED ? ",+si-scheduler" : "",
-                tm_options & AC_TM_FORCE_ENABLE_XNACK ? ",+xnack" : "",
-                tm_options & AC_TM_FORCE_DISABLE_XNACK ? ",-xnack" : "",
-                tm_options & AC_TM_PROMOTE_ALLOCA_TO_SCRATCH ? ",-promote-alloca" : "",
-                tm_options & AC_TM_NO_LOAD_STORE_OPT ? ",-load-store-opt" : "");
+                family <= CHIP_NAVI14 && tm_options & AC_TM_FORCE_ENABLE_XNACK ? ",+xnack" : "",
+                family <= CHIP_NAVI14 && tm_options & AC_TM_FORCE_DISABLE_XNACK ? ",-xnack" : "",
+                tm_options & AC_TM_PROMOTE_ALLOCA_TO_SCRATCH ? ",-promote-alloca" : "");
 
        LLVMTargetMachineRef tm = LLVMCreateTargetMachine(
                                     target,