[Mesa-dev] [PATCH] ac: fix WAITCNT flags for GFX9

[Samuel Pitoiset]

Thanks!

Reviewed-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>

On 8/22/18 3:19 AM, Marek Olšák wrote:
> From: Marek Olšák <marek.olsak at amd.com>
> 
> ---
>   src/amd/common/ac_llvm_build.h                    | 6 ++++++
>   src/amd/common/ac_nir_to_llvm.c                   | 4 ----
>   src/gallium/drivers/radeonsi/si_shader_internal.h | 5 -----
>   3 files changed, 6 insertions(+), 9 deletions(-)
> 
> diff --git a/src/amd/common/ac_llvm_build.h b/src/amd/common/ac_llvm_build.h
> index d11a6ebb1be..b080cca4cb7 100644
> --- a/src/amd/common/ac_llvm_build.h
> +++ b/src/amd/common/ac_llvm_build.h
> @@ -36,20 +36,26 @@ extern "C" {
>   
>   #define HAVE_32BIT_POINTERS (HAVE_LLVM >= 0x0700)
>   
>   enum {
>   	/* CONST is the only address space that selects SMEM loads */
>   	AC_CONST_ADDR_SPACE = HAVE_LLVM >= 0x700 ? 4 : 2,
>   	AC_LOCAL_ADDR_SPACE = 3,
>   	AC_CONST_32BIT_ADDR_SPACE = 6, /* same as CONST, but the pointer type has 32 bits */
>   };
>   
> +/* Combine these with & instead of |. */
> +#define NOOP_WAITCNT	0xcf7f
> +#define LGKM_CNT	0xc07f
> +#define EXP_CNT		0xcf0f
> +#define VM_CNT		0x0f70 /* On GFX9, vmcnt has 6 bits in [0:3] and [14:15] */
> +
>   struct ac_llvm_flow;
>   
>   struct ac_llvm_context {
>   	LLVMContextRef context;
>   	LLVMModuleRef module;
>   	LLVMBuilderRef builder;
>   
>   	LLVMTypeRef voidt;
>   	LLVMTypeRef i1;
>   	LLVMTypeRef i8;
> diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
> index 700e48e14b7..bfe35e4671b 100644
> --- a/src/amd/common/ac_nir_to_llvm.c
> +++ b/src/amd/common/ac_nir_to_llvm.c
> @@ -2528,24 +2528,20 @@ static LLVMValueRef visit_image_size(struct ac_nir_context *ctx,
>   	    glsl_get_sampler_dim(type) == GLSL_SAMPLER_DIM_1D &&
>   	    glsl_sampler_type_is_array(type)) {
>   		LLVMValueRef layers = LLVMBuildExtractElement(ctx->ac.builder, res, two, "");
>   		res = LLVMBuildInsertElement(ctx->ac.builder, res, layers,
>   						ctx->ac.i32_1, "");
>   
>   	}
>   	return res;
>   }
>   
> -#define NOOP_WAITCNT 0xf7f
> -#define LGKM_CNT 0x07f
> -#define VM_CNT 0xf70
> -
>   static void emit_membar(struct ac_llvm_context *ac,
>   			const nir_intrinsic_instr *instr)
>   {
>   	unsigned waitcnt = NOOP_WAITCNT;
>   
>   	switch (instr->intrinsic) {
>   	case nir_intrinsic_memory_barrier:
>   	case nir_intrinsic_group_memory_barrier:
>   		waitcnt &= VM_CNT & LGKM_CNT;
>   		break;
> diff --git a/src/gallium/drivers/radeonsi/si_shader_internal.h b/src/gallium/drivers/radeonsi/si_shader_internal.h
> index ac7784f7d46..aff7e0472db 100644
> --- a/src/gallium/drivers/radeonsi/si_shader_internal.h
> +++ b/src/gallium/drivers/radeonsi/si_shader_internal.h
> @@ -281,25 +281,20 @@ LLVMValueRef si_llvm_load_input_gs(struct ac_shader_abi *abi,
>   LLVMValueRef si_nir_lookup_interp_param(struct ac_shader_abi *abi,
>   					enum glsl_interp_mode interp,
>   					unsigned location);
>   
>   void si_llvm_emit_store(struct lp_build_tgsi_context *bld_base,
>   			const struct tgsi_full_instruction *inst,
>   			const struct tgsi_opcode_info *info,
>   			unsigned index,
>   			LLVMValueRef dst[4]);
>   
> -/* Combine these with & instead of |. */
> -#define NOOP_WAITCNT 0xf7f
> -#define LGKM_CNT 0x07f
> -#define VM_CNT 0xf70
> -
>   LLVMValueRef si_get_indirect_index(struct si_shader_context *ctx,
>   				   const struct tgsi_ind_register *ind,
>   				   unsigned addr_mul, int rel_index);
>   LLVMValueRef si_get_bounded_indirect_index(struct si_shader_context *ctx,
>   					   const struct tgsi_ind_register *ind,
>   					   int rel_index, unsigned num);
>   LLVMValueRef si_get_sample_id(struct si_shader_context *ctx);
>   
>   void si_shader_context_init_alu(struct lp_build_tgsi_context *bld_base);
>   void si_shader_context_init_mem(struct si_shader_context *ctx);
>