[PATCH i-g-t 3/3] lib/amdgpu: fix gfx11 issue about amd_dispatch

[vitaly prosyak]

Looks good to me.

Reviewed-by: Vitaly Prosyak <vitaly.prosyak at amd.com>

On 2024-06-20 22:50, Jesse Zhang wrote:
> add some initialized register and fix some uncorrected register setting.
>
> Signed-off-by: Jesse Zhang <Jesse.Zhang at amd.com>
> Signed-off-by: Jiadong Zhu <Jiadong.Zhu at amd.com>
> ---
>  lib/amdgpu/amd_dispatch.c         |  2 +-
>  lib/amdgpu/amd_dispatch_helpers.c | 34 ++++++++++++++++++++-----------
>  lib/amdgpu/amd_shaders.c          |  3 +--
>  3 files changed, 24 insertions(+), 15 deletions(-)
>
> diff --git a/lib/amdgpu/amd_dispatch.c b/lib/amdgpu/amd_dispatch.c
> index b739ce64d..df7d56ea7 100644
> --- a/lib/amdgpu/amd_dispatch.c
> +++ b/lib/amdgpu/amd_dispatch.c
> @@ -345,7 +345,7 @@ amdgpu_memcpy_dispatch_hang_slow_test(amdgpu_device_handle device_handle,
>  	int r, r2;
>  
>  	int bo_dst_size = 0x4000000;
> -	int bo_shader_size = 0x400000;
> +	int bo_shader_size = 0x4000000;
>  	int bo_cmd_size = 4096;
>  
>  	struct amdgpu_cs_request ibs_request = {0};
> diff --git a/lib/amdgpu/amd_dispatch_helpers.c b/lib/amdgpu/amd_dispatch_helpers.c
> index 1c7413922..b0a5f550e 100644
> --- a/lib/amdgpu/amd_dispatch_helpers.c
> +++ b/lib/amdgpu/amd_dispatch_helpers.c
> @@ -70,8 +70,20 @@
>  		base->emit(base, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
>  		base->emit(base, 0x7b);
>  		base->emit(base, 0x20);
> +	} else if (version == 11) {
> +		base->emit(base, PACKET3_COMPUTE(PKT3_SET_SH_REG, 1));
> +		base->emit(base, 0x222);
> +		base->emit(base, 0);
> +		base->emit(base, PACKET3_COMPUTE(PKT3_SET_SH_REG, 4));
> +		base->emit(base, 0x224);
> +		base->emit(base, 0);
> +		base->emit(base, 0);
> +		base->emit(base, 0);
> +		base->emit(base, 0);
> +		base->emit(base, PACKET3_COMPUTE(PKT3_SET_SH_REG, 1));
> +		base->emit(base, 0x22a);
> +		base->emit(base, 0);
>  	}
> -
>  	return base->cdw - i;
>  }
>  
> @@ -85,27 +97,20 @@ int amdgpu_dispatch_write_cumask(struct amdgpu_cmd_base * base, uint32_t version
>   		base->emit(base, 0x216);
>   		base->emit(base, 0xffffffff);
>   		base->emit(base, 0xffffffff);
> - 	} else if(version == 10) {
> +	} else if((version == 10) || (version == 11)) {
>  		/* set mmCOMPUTE_STATIC_THREAD_MGMT_SE1 - mmCOMPUTE_STATIC_THREAD_MGMT_SE0 */
>   		base->emit(base, PACKET3_COMPUTE(PKT3_SET_SH_REG_INDEX, 2));
>   		base->emit(base, 0x30000216);
>   		base->emit(base, 0xffffffff);
>   		base->emit(base, 0xffffffff);
> -		/* set mmCOMPUTE_STATIC_THREAD_MGMT_SE3 - mmCOMPUTE_STATIC_THREAD_MGMT_SE2 */
> - 		base->emit(base, PACKET3_COMPUTE(PKT3_SET_SH_REG_INDEX, 2));
> - 		base->emit(base, 0x30000219);
> - 		base->emit(base, 0xffffffff);
> - 		base->emit(base, 0xffffffff);
>  	}
> -
> -
> - 	/* set mmCOMPUTE_STATIC_THREAD_MGMT_SE3 - mmCOMPUTE_STATIC_THREAD_MGMT_SE2 */
> - 	base->emit(base, PACKET3_COMPUTE(PKT3_SET_SH_REG, 2));
> +	/* set mmCOMPUTE_STATIC_THREAD_MGMT_SE3 - mmCOMPUTE_STATIC_THREAD_MGMT_SE2 */
> +	base->emit(base, PACKET3_COMPUTE(PKT3_SET_SH_REG_INDEX, 2));
>   	base->emit(base, 0x219);
>   	base->emit(base, 0xffffffff);
>   	base->emit(base, 0xffffffff);
>  
> - 	return base->cdw - offset_prev;
> +	return base->cdw - offset_prev;
>   }
>  
>  
> @@ -161,6 +166,11 @@ int amdgpu_dispatch_write2hw(struct amdgpu_cmd_base * base, uint64_t shader_addr
>  		base->emit(base, PACKET3_COMPUTE(PKT3_SET_SH_REG, 1));
>  		base->emit(base,0x228);
>  		base->emit(base, 0 );
> +	} else if (version == 11) {
> +		/* mmCOMPUTE_PGM_RSRC3 */
> +		base->emit(base, PACKET3_COMPUTE(PKT3_SET_SH_REG, 1));
> +		base->emit(base,0x228);
> +		base->emit(base, 0x3f0 );
>  	}
>  	return base->cdw - offset_prev;
>  }
> diff --git a/lib/amdgpu/amd_shaders.c b/lib/amdgpu/amd_shaders.c
> index fad6dbced..e4931961c 100644
> --- a/lib/amdgpu/amd_shaders.c
> +++ b/lib/amdgpu/amd_shaders.c
> @@ -132,7 +132,6 @@ int amdgpu_dispatch_load_cs_shader_hang_slow(uint32_t *ptr, uint32_t family_id)
>  	    0xd7460000, 0x04010c08, 0xe00c2000, 0x80000100,
>  	    0xbf8c0f70, 0xe01ca000, 0x80010100, 0xbf810000
>  	};
> -
>  	struct amdgpu_test_shader memcpy_cs_hang_slow_nv = {
>  	        memcpy_cs_hang_slow_nv_codes,
>  	        4,
> @@ -140,7 +139,7 @@ int amdgpu_dispatch_load_cs_shader_hang_slow(uint32_t *ptr, uint32_t family_id)
>  	        1
>  	};
>  	struct amdgpu_test_shader *shader;
> -	int i, loop = 0x10000;
> +	int i, loop = 0x100000;
>  
>  	switch (family_id) {
>  		case AMDGPU_FAMILY_AI: