From 13c34cf8ca43d0f9c1e1a663e6a3783b0938dfd9 Mon Sep 17 00:00:00 2001 From: =?utf8?q?Marek=20Ol=C5=A1=C3=A1k?= Date: Fri, 2 Dec 2016 19:40:40 +0100 Subject: [PATCH] radeonsi: wait for outstanding LDS instructions in memory barriers if needed MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit Cc: 13.0 Reviewed-by: Nicolai Hähnle --- src/gallium/drivers/radeonsi/si_shader.c | 18 +++++++++++++++++- 1 file changed, 17 insertions(+), 1 deletion(-) diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c index 53a8493dfce..236534906f5 100644 --- a/src/gallium/drivers/radeonsi/si_shader.c +++ b/src/gallium/drivers/radeonsi/si_shader.c @@ -3291,6 +3291,7 @@ static void emit_optimization_barrier(struct si_shader_context *ctx) #endif /* Combine these with & instead of |. */ +#define NOOP_WAITCNT 0xf7f #define LGKM_CNT 0x07f #define VM_CNT 0xf70 @@ -3311,8 +3312,23 @@ static void membar_emit( struct lp_build_emit_data *emit_data) { struct si_shader_context *ctx = si_shader_context(bld_base); + LLVMValueRef src0 = lp_build_emit_fetch(bld_base, emit_data->inst, 0, 0); + unsigned flags = LLVMConstIntGetZExtValue(src0); + unsigned waitcnt = NOOP_WAITCNT; - emit_waitcnt(ctx, VM_CNT); + if (flags & TGSI_MEMBAR_THREAD_GROUP) + waitcnt &= VM_CNT & LGKM_CNT; + + if (flags & (TGSI_MEMBAR_ATOMIC_BUFFER | + TGSI_MEMBAR_SHADER_BUFFER | + TGSI_MEMBAR_SHADER_IMAGE)) + waitcnt &= VM_CNT; + + if (flags & TGSI_MEMBAR_SHARED) + waitcnt &= LGKM_CNT; + + if (waitcnt != NOOP_WAITCNT) + emit_waitcnt(ctx, waitcnt); } static LLVMValueRef -- 2.30.2