[PATCH 04/10] amdgpu/dc: allow inlining constant int to fixed a lot better.

[Harry Wentland]

On 2017-09-29 03:13 AM, Dave Airlie wrote:
> From: Dave Airlie <airlied at redhat.com>
> 
> This uses two things that might be Linux specific,
> __builtin_constant_p (might be gcc)
> and
> BUILD_BUG_ON. (maybe other dm's can have something similiar).
> 

We can always use ifdef in our internal tree to keep Linux specific bits
without affecting other users of our code. As long as it's not all over
the place and the payoff is worthwhile, which it is in this case.

Harry

> This saves 4k in the bw calcs code.
> 
> Signed-off-by: Dave Airlie <airlied at redhat.com>
> ---
>  drivers/gpu/drm/amd/display/dc/calcs/bw_fixed.c | 37 ++++++++-----------------
>  drivers/gpu/drm/amd/display/dc/inc/bw_fixed.h   | 26 +++++++++++++++--
>  2 files changed, 35 insertions(+), 28 deletions(-)
> 
> diff --git a/drivers/gpu/drm/amd/display/dc/calcs/bw_fixed.c b/drivers/gpu/drm/amd/display/dc/calcs/bw_fixed.c
> index 0de6fa1..6ca288f 100644
> --- a/drivers/gpu/drm/amd/display/dc/calcs/bw_fixed.c
> +++ b/drivers/gpu/drm/amd/display/dc/calcs/bw_fixed.c
> @@ -25,13 +25,6 @@
>  #include "dm_services.h"
>  #include "bw_fixed.h"
>  
> -#define BITS_PER_FRACTIONAL_PART 24
> -
> -#define MIN_I32 \
> -	(int64_t)(-(1LL << (63 - BITS_PER_FRACTIONAL_PART)))
> -
> -#define MAX_I32 \
> -	(int64_t)((1ULL << (63 - BITS_PER_FRACTIONAL_PART)) - 1)
>  
>  #define MIN_I64 \
>  	(int64_t)(-(1LL << 63))
> @@ -40,10 +33,7 @@
>  	(int64_t)((1ULL << 63) - 1)
>  
>  #define FRACTIONAL_PART_MASK \
> -	((1ULL << BITS_PER_FRACTIONAL_PART) - 1)
> -
> -#define GET_INTEGER_PART(x) \
> -	((x) >> BITS_PER_FRACTIONAL_PART)
> +	((1ULL << BW_FIXED_BITS_PER_FRACTIONAL_PART) - 1)
>  
>  #define GET_FRACTIONAL_PART(x) \
>  	(FRACTIONAL_PART_MASK & (x))
> @@ -56,19 +46,14 @@ static uint64_t abs_i64(int64_t arg)
>  		return (uint64_t)(-arg);
>  }
>  
> -struct bw_fixed bw_int_to_fixed(int64_t value)
> +struct bw_fixed bw_int_to_fixed_nonconst(int64_t value)
>  {
>  	struct bw_fixed res;
> -	ASSERT(value < MAX_I32 && value > MIN_I32);
> -	res.value = value << BITS_PER_FRACTIONAL_PART;
> +	ASSERT(value < BW_FIXED_MAX_I32 && value > BW_FIXED_MIN_I32);
> +	res.value = value << BW_FIXED_BITS_PER_FRACTIONAL_PART;
>  	return res;
>  }
>  
> -int32_t bw_fixed_to_int(struct bw_fixed value)
> -{
> -	return GET_INTEGER_PART(value.value);
> -}
> -
>  struct bw_fixed bw_frc_to_fixed(int64_t numerator, int64_t denominator)
>  {
>  	struct bw_fixed res;
> @@ -87,11 +72,11 @@ struct bw_fixed bw_frc_to_fixed(int64_t numerator, int64_t denominator)
>  	arg2_value = abs_i64(denominator);
>  	res_value = div64_u64_rem(arg1_value, arg2_value, &remainder);
>  
> -	ASSERT(res_value <= MAX_I32);
> +	ASSERT(res_value <= BW_FIXED_MAX_I32);
>  
>  	/* determine fractional part */
>  	{
> -		uint32_t i = BITS_PER_FRACTIONAL_PART;
> +		uint32_t i = BW_FIXED_BITS_PER_FRACTIONAL_PART;
>  
>  		do
>  		{
> @@ -164,8 +149,8 @@ struct bw_fixed bw_mul(const struct bw_fixed arg1, const struct bw_fixed arg2)
>  	uint64_t arg1_value = abs_i64(arg1.value);
>  	uint64_t arg2_value = abs_i64(arg2.value);
>  
> -	uint64_t arg1_int = GET_INTEGER_PART(arg1_value);
> -	uint64_t arg2_int = GET_INTEGER_PART(arg2_value);
> +	uint64_t arg1_int = BW_FIXED_GET_INTEGER_PART(arg1_value);
> +	uint64_t arg2_int = BW_FIXED_GET_INTEGER_PART(arg2_value);
>  
>  	uint64_t arg1_fra = GET_FRACTIONAL_PART(arg1_value);
>  	uint64_t arg2_fra = GET_FRACTIONAL_PART(arg2_value);
> @@ -174,9 +159,9 @@ struct bw_fixed bw_mul(const struct bw_fixed arg1, const struct bw_fixed arg2)
>  
>  	res.value = arg1_int * arg2_int;
>  
> -	ASSERT(res.value <= MAX_I32);
> +	ASSERT(res.value <= BW_FIXED_MAX_I32);
>  
> -	res.value <<= BITS_PER_FRACTIONAL_PART;
> +	res.value <<= BW_FIXED_BITS_PER_FRACTIONAL_PART;
>  
>  	tmp = arg1_int * arg2_fra;
>  
> @@ -192,7 +177,7 @@ struct bw_fixed bw_mul(const struct bw_fixed arg1, const struct bw_fixed arg2)
>  
>  	tmp = arg1_fra * arg2_fra;
>  
> -	tmp = (tmp >> BITS_PER_FRACTIONAL_PART) +
> +	tmp = (tmp >> BW_FIXED_BITS_PER_FRACTIONAL_PART) +
>  		(tmp >= (uint64_t)(bw_frc_to_fixed(1, 2).value));
>  
>  	ASSERT(tmp <= (uint64_t)(MAX_I64 - res.value));
> diff --git a/drivers/gpu/drm/amd/display/dc/inc/bw_fixed.h b/drivers/gpu/drm/amd/display/dc/inc/bw_fixed.h
> index 4477e62..39ee8eba3 100644
> --- a/drivers/gpu/drm/amd/display/dc/inc/bw_fixed.h
> +++ b/drivers/gpu/drm/amd/display/dc/inc/bw_fixed.h
> @@ -26,10 +26,19 @@
>  #ifndef BW_FIXED_H_
>  #define BW_FIXED_H_
>  
> +#define BW_FIXED_BITS_PER_FRACTIONAL_PART 24
> +
> +#define BW_FIXED_GET_INTEGER_PART(x) ((x) >> BW_FIXED_BITS_PER_FRACTIONAL_PART)
>  struct bw_fixed {
>  	int64_t value;
>  };
>  
> +#define BW_FIXED_MIN_I32 \
> +	(int64_t)(-(1LL << (63 - BW_FIXED_BITS_PER_FRACTIONAL_PART)))
> +
> +#define BW_FIXED_MAX_I32 \
> +	(int64_t)((1ULL << (63 - BW_FIXED_BITS_PER_FRACTIONAL_PART)) - 1)
> +
>  static inline struct bw_fixed bw_min2(const struct bw_fixed arg1,
>  				      const struct bw_fixed arg2)
>  {
> @@ -56,9 +65,22 @@ static inline struct bw_fixed bw_max3(struct bw_fixed v1,
>  	return bw_max2(bw_max2(v1, v2), v3);
>  }
>  
> -struct bw_fixed bw_int_to_fixed(int64_t value);
> +struct bw_fixed bw_int_to_fixed_nonconst(int64_t value);
> +static inline struct bw_fixed bw_int_to_fixed(int64_t value)
> +{
> +	if (__builtin_constant_p(value)) {
> +		struct bw_fixed res;
> +		BUILD_BUG_ON(value > BW_FIXED_MAX_I32 || value < BW_FIXED_MIN_I32);
> +		res.value = value << BW_FIXED_BITS_PER_FRACTIONAL_PART;
> +		return res;
> +	} else
> +		return bw_int_to_fixed_nonconst(value);
> +}
>  
> -int32_t bw_fixed_to_int(struct bw_fixed value);
> +static inline int32_t bw_fixed_to_int(struct bw_fixed value)
> +{
> +	return BW_FIXED_GET_INTEGER_PART(value.value);
> +}
>  
>  struct bw_fixed bw_frc_to_fixed(int64_t num, int64_t denum);
>  
>