From 7ff64fcaa7b7ba62d12758e49643f31dd9e90ece Mon Sep 17 00:00:00 2001 From: Steven Fuerst Date: Wed, 15 Aug 2012 15:07:14 -0700 Subject: Rename i2f() to int2float(), and make it global so one copy can be removed. Remove the copy of i2f() in r600_blit_kms.c We rename the function to something longer now that it is a global symbol. This reduces the likelyhood of unintended clashes later. This might be a candidate for inclusion inside general drm infrastructure. However, at the moment only the radeon driver uses it. Signed-off-by: Steven Fuerst --- drivers/gpu/drm/radeon/r600_blit.c | 66 +++++++++++++++++++------------------- 1 file changed, 33 insertions(+), 33 deletions(-) (limited to 'drivers/gpu/drm/radeon/r600_blit.c') diff --git a/drivers/gpu/drm/radeon/r600_blit.c b/drivers/gpu/drm/radeon/r600_blit.c index 3c031a48205..ee1b815b568 100644 --- a/drivers/gpu/drm/radeon/r600_blit.c +++ b/drivers/gpu/drm/radeon/r600_blit.c @@ -489,7 +489,7 @@ set_default_state(drm_radeon_private_t *dev_priv) ADVANCE_RING(); } -static uint32_t i2f(uint32_t input) +uint32_t int2float(uint32_t input) { u32 result, i, exponent, fraction; @@ -632,20 +632,20 @@ r600_blit_copy(struct drm_device *dev, vb = r600_nomm_get_vb_ptr(dev); } - vb[0] = i2f(dst_x); + vb[0] = int2float(dst_x); vb[1] = 0; - vb[2] = i2f(src_x); + vb[2] = int2float(src_x); vb[3] = 0; - vb[4] = i2f(dst_x); - vb[5] = i2f(h); - vb[6] = i2f(src_x); - vb[7] = i2f(h); + vb[4] = int2float(dst_x); + vb[5] = int2float(h); + vb[6] = int2float(src_x); + vb[7] = int2float(h); - vb[8] = i2f(dst_x + cur_size); - vb[9] = i2f(h); - vb[10] = i2f(src_x + cur_size); - vb[11] = i2f(h); + vb[8] = int2float(dst_x + cur_size); + vb[9] = int2float(h); + vb[10] = int2float(src_x + cur_size); + vb[11] = int2float(h); /* src */ set_tex_resource(dev_priv, FMT_8, @@ -721,20 +721,20 @@ r600_blit_copy(struct drm_device *dev, vb = r600_nomm_get_vb_ptr(dev); } - vb[0] = i2f(dst_x / 4); + vb[0] = int2float(dst_x / 4); vb[1] = 0; - vb[2] = i2f(src_x / 4); + vb[2] = int2float(src_x / 4); vb[3] = 0; - vb[4] = i2f(dst_x / 4); - vb[5] = i2f(h); - vb[6] = i2f(src_x / 4); - vb[7] = i2f(h); + vb[4] = int2float(dst_x / 4); + vb[5] = int2float(h); + vb[6] = int2float(src_x / 4); + vb[7] = int2float(h); - vb[8] = i2f((dst_x + cur_size) / 4); - vb[9] = i2f(h); - vb[10] = i2f((src_x + cur_size) / 4); - vb[11] = i2f(h); + vb[8] = int2float((dst_x + cur_size) / 4); + vb[9] = int2float(h); + vb[10] = int2float((src_x + cur_size) / 4); + vb[11] = int2float(h); /* src */ set_tex_resource(dev_priv, FMT_8_8_8_8, @@ -804,20 +804,20 @@ r600_blit_swap(struct drm_device *dev, dx2 = dx + w; dy2 = dy + h; - vb[0] = i2f(dx); - vb[1] = i2f(dy); - vb[2] = i2f(sx); - vb[3] = i2f(sy); + vb[0] = int2float(dx); + vb[1] = int2float(dy); + vb[2] = int2float(sx); + vb[3] = int2float(sy); - vb[4] = i2f(dx); - vb[5] = i2f(dy2); - vb[6] = i2f(sx); - vb[7] = i2f(sy2); + vb[4] = int2float(dx); + vb[5] = int2float(dy2); + vb[6] = int2float(sx); + vb[7] = int2float(sy2); - vb[8] = i2f(dx2); - vb[9] = i2f(dy2); - vb[10] = i2f(sx2); - vb[11] = i2f(sy2); + vb[8] = int2float(dx2); + vb[9] = int2float(dy2); + vb[10] = int2float(sx2); + vb[11] = int2float(sy2); switch(cpp) { case 4: -- cgit v1.2.3-70-g09d2 From 747f49ba67b8895a5831ab539de551b916f3738c Mon Sep 17 00:00:00 2001 From: Steven Fuerst Date: Wed, 15 Aug 2012 15:07:15 -0700 Subject: Replace int2float() with an optimized version. We use __fls() to find the most significant bit. Using that, the loop can be avoided. A second trick is to use the behaviour of the rotate instructions to expand the range of the unsigned int to float conversion to the full 32 bits in a branchless way. The routine is now exact up to 2^24. Above that, we truncate which is equivalent to rounding towards zero. Signed-off-by: Steven Fuerst --- drivers/gpu/drm/radeon/r600_blit.c | 51 +++++++++++++++++++++----------------- 1 file changed, 28 insertions(+), 23 deletions(-) (limited to 'drivers/gpu/drm/radeon/r600_blit.c') diff --git a/drivers/gpu/drm/radeon/r600_blit.c b/drivers/gpu/drm/radeon/r600_blit.c index ee1b815b568..7d8ac42e384 100644 --- a/drivers/gpu/drm/radeon/r600_blit.c +++ b/drivers/gpu/drm/radeon/r600_blit.c @@ -489,31 +489,36 @@ set_default_state(drm_radeon_private_t *dev_priv) ADVANCE_RING(); } -uint32_t int2float(uint32_t input) +/* 23 bits of float fractional data */ +#define I2F_FRAC_BITS 23 +#define I2F_MASK ((1 << I2F_FRAC_BITS) - 1) + +/* + * Converts unsigned integer into 32-bit IEEE floating point representation. + * Will be exact from 0 to 2^24. Above that, we round towards zero + * as the fractional bits will not fit in a float. (It would be better to + * round towards even as the fpu does, but that is slower.) + */ +uint32_t int2float(uint32_t x) { - u32 result, i, exponent, fraction; - - if ((input & 0x3fff) == 0) - result = 0; /* 0 is a special case */ - else { - exponent = 140; /* exponent biased by 127; */ - fraction = (input & 0x3fff) << 10; /* cheat and only - handle numbers below 2^^15 */ - for (i = 0; i < 14; i++) { - if (fraction & 0x800000) - break; - else { - fraction = fraction << 1; /* keep - shifting left until top bit = 1 */ - exponent = exponent - 1; - } - } - result = exponent << 23 | (fraction & 0x7fffff); /* mask - off top bit; assumed 1 */ - } - return result; -} + uint32_t msb, exponent, fraction; + + /* Zero is special */ + if (!x) return 0; + + /* Get location of the most significant bit */ + msb = __fls(x); + /* + * Use a rotate instead of a shift because that works both leftwards + * and rightwards due to the mod(32) behaviour. This means we don't + * need to check to see if we are above 2^24 or not. + */ + fraction = ror32(x, (msb - I2F_FRAC_BITS) & 0x1f) & I2F_MASK; + exponent = (127 + msb) << I2F_FRAC_BITS; + + return fraction + exponent; +} static int r600_nomm_get_vb(struct drm_device *dev) { -- cgit v1.2.3-70-g09d2 From 9e9eb7c60d57620bfe46b2a489e7f56a5925115a Mon Sep 17 00:00:00 2001 From: Steven Fuerst Date: Wed, 15 Aug 2012 15:07:16 -0700 Subject: Annotate int2float() as being a pure function. This allows gcc to fold duplicate calls into a single call. Since the current users do actually call it multiple times with the same arguments, this is an obvious win. Signed-off-by: Steven Fuerst --- drivers/gpu/drm/radeon/r600_blit.c | 2 +- drivers/gpu/drm/radeon/r600_blit_shaders.h | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) (limited to 'drivers/gpu/drm/radeon/r600_blit.c') diff --git a/drivers/gpu/drm/radeon/r600_blit.c b/drivers/gpu/drm/radeon/r600_blit.c index 7d8ac42e384..661fec2a2cc 100644 --- a/drivers/gpu/drm/radeon/r600_blit.c +++ b/drivers/gpu/drm/radeon/r600_blit.c @@ -499,7 +499,7 @@ set_default_state(drm_radeon_private_t *dev_priv) * as the fractional bits will not fit in a float. (It would be better to * round towards even as the fpu does, but that is slower.) */ -uint32_t int2float(uint32_t x) +__pure uint32_t int2float(uint32_t x) { uint32_t msb, exponent, fraction; diff --git a/drivers/gpu/drm/radeon/r600_blit_shaders.h b/drivers/gpu/drm/radeon/r600_blit_shaders.h index e17c2cbc662..2f3ce7a7597 100644 --- a/drivers/gpu/drm/radeon/r600_blit_shaders.h +++ b/drivers/gpu/drm/radeon/r600_blit_shaders.h @@ -35,5 +35,5 @@ extern const u32 r6xx_default_state[]; extern const u32 r6xx_ps_size, r6xx_vs_size; extern const u32 r6xx_default_size, r7xx_default_size; -uint32_t int2float(uint32_t x); +__pure uint32_t int2float(uint32_t x); #endif -- cgit v1.2.3-70-g09d2