radv: disable the auto-waitcnt-before-barrier LLVM option

This option allows us to remove additional s_waitcnt instructions
because s_barrier internally does s_waitcnt 0.

Though, apparently there is a problem with LDS accesses that
causes rendering issues with FFXV and DXVK. Disable this
optimization for now (RadeonSI still uses it).

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107460
CC: 18.2 <mesa-stable@lists.freedesktop.org>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

src/amd/common/ac_llvm_util.c

@@ -149,7 +149,8 @@ static LLVMTargetMachineRef ac_create_target_machine(enum radeon_family family,
 	char features[256];
 	const char *triple = (tm_options & AC_TM_SUPPORTS_SPILL) ? "amdgcn-mesa-mesa3d" : "amdgcn--";
 	LLVMTargetRef target = ac_get_llvm_target(triple);
-	bool barrier_does_waitcnt = family != CHIP_VEGA20;
+	bool barrier_does_waitcnt = (tm_options & AC_TM_AUTO_WAITCNT_BEFORE_BARRIER) &&
+				    family != CHIP_VEGA20;
 
 	snprintf(features, sizeof(features),
 		 "+DumpCode,+vgpr-spilling,-fp32-denormals,+fp64-denormals%s%s%s%s%s",

src/amd/common/ac_llvm_util.h

@@ -65,6 +65,7 @@ enum ac_target_machine_options {
 	AC_TM_CHECK_IR = (1 << 5),
 	AC_TM_ENABLE_GLOBAL_ISEL = (1 << 6),
 	AC_TM_CREATE_LOW_OPT = (1 << 7),
+	AC_TM_AUTO_WAITCNT_BEFORE_BARRIER = (1 << 8),
 };
 
 enum ac_float_mode {

src/gallium/drivers/radeonsi/si_pipe.c

@@ -115,6 +115,7 @@ static void si_init_compiler(struct si_screen *sscreen,
 				       sscreen->info.chip_class <= VI;
 
 	enum ac_target_machine_options tm_options =
+		AC_TM_AUTO_WAITCNT_BEFORE_BARRIER |
 		(sscreen->debug_flags & DBG(SI_SCHED) ? AC_TM_SISCHED : 0) |
 		(sscreen->debug_flags & DBG(GISEL) ? AC_TM_ENABLE_GLOBAL_ISEL : 0) |
 		(sscreen->info.chip_class >= GFX9 ? AC_TM_FORCE_ENABLE_XNACK : 0) |