[Mesa-dev] [PATCH] radeonsi: add a workaround for a bug in LLVM <= 3.8

[Nicolai Hähnle]

On 18.05.2016 11:13, Marek Olšák wrote:
> From: Marek Olšák <marek.olsak at amd.com>
>
> This is not directly applicable to stable and needs to be backported.

This is because LLVM gets confused / misled about SGPRs vs. VGPRs, 
right? It's unfortunate, but I see no better way, so...

Reviewed-by: Nicolai Hähnle <nicolai.haehnle at amd.com>

> ---
>   src/gallium/drivers/radeonsi/si_shader.c | 7 +++++++
>   1 file changed, 7 insertions(+)
>
> diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
> index 6166403..3df7820 100644
> --- a/src/gallium/drivers/radeonsi/si_shader.c
> +++ b/src/gallium/drivers/radeonsi/si_shader.c
> @@ -556,6 +556,13 @@ static LLVMValueRef get_bounded_indirect_index(struct si_shader_context *ctx,
>   	LLVMValueRef c_max = LLVMConstInt(ctx->i32, num - 1, 0);
>   	LLVMValueRef cc;
>
> +	/* LLVM 3.8: If indirect resource indexing is used:
> +	 * - SI & CIK hang
> +	 * - VI crashes
> +	 */
> +	if (HAVE_LLVM <= 0x0308)
> +		return LLVMGetUndef(ctx->i32);
> +
>   	if (util_is_power_of_two(num)) {
>   		result = LLVMBuildAnd(builder, result, c_max, "");
>   	} else {
>