drm: move radeon_fixed.h to shared drm_fixed.h header

Will be used by nouveau driver also in the near future.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This commit is contained in:
Ben Skeggs 2010-04-28 11:46:42 +10:00 committed by Dave Airlie
parent 15a7df8db8
commit 68adac5e49
9 changed files with 398 additions and 398 deletions

View File

@ -26,7 +26,7 @@
#include <drm/drmP.h>
#include <drm/drm_crtc_helper.h>
#include <drm/radeon_drm.h>
#include "radeon_fixed.h"
#include <drm/drm_fixed.h>
#include "radeon.h"
#include "atom.h"
#include "atom-bits.h"

View File

@ -2686,53 +2686,53 @@ void r100_bandwidth_update(struct radeon_device *rdev)
fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff, crit_point_ff;
uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
fixed20_12 memtcas_ff[8] = {
fixed_init(1),
fixed_init(2),
fixed_init(3),
fixed_init(0),
fixed_init_half(1),
fixed_init_half(2),
fixed_init(0),
dfixed_init(1),
dfixed_init(2),
dfixed_init(3),
dfixed_init(0),
dfixed_init_half(1),
dfixed_init_half(2),
dfixed_init(0),
};
fixed20_12 memtcas_rs480_ff[8] = {
fixed_init(0),
fixed_init(1),
fixed_init(2),
fixed_init(3),
fixed_init(0),
fixed_init_half(1),
fixed_init_half(2),
fixed_init_half(3),
dfixed_init(0),
dfixed_init(1),
dfixed_init(2),
dfixed_init(3),
dfixed_init(0),
dfixed_init_half(1),
dfixed_init_half(2),
dfixed_init_half(3),
};
fixed20_12 memtcas2_ff[8] = {
fixed_init(0),
fixed_init(1),
fixed_init(2),
fixed_init(3),
fixed_init(4),
fixed_init(5),
fixed_init(6),
fixed_init(7),
dfixed_init(0),
dfixed_init(1),
dfixed_init(2),
dfixed_init(3),
dfixed_init(4),
dfixed_init(5),
dfixed_init(6),
dfixed_init(7),
};
fixed20_12 memtrbs[8] = {
fixed_init(1),
fixed_init_half(1),
fixed_init(2),
fixed_init_half(2),
fixed_init(3),
fixed_init_half(3),
fixed_init(4),
fixed_init_half(4)
dfixed_init(1),
dfixed_init_half(1),
dfixed_init(2),
dfixed_init_half(2),
dfixed_init(3),
dfixed_init_half(3),
dfixed_init(4),
dfixed_init_half(4)
};
fixed20_12 memtrbs_r4xx[8] = {
fixed_init(4),
fixed_init(5),
fixed_init(6),
fixed_init(7),
fixed_init(8),
fixed_init(9),
fixed_init(10),
fixed_init(11)
dfixed_init(4),
dfixed_init(5),
dfixed_init(6),
dfixed_init(7),
dfixed_init(8),
dfixed_init(9),
dfixed_init(10),
dfixed_init(11)
};
fixed20_12 min_mem_eff;
fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
@ -2763,7 +2763,7 @@ void r100_bandwidth_update(struct radeon_device *rdev)
}
}
min_mem_eff.full = rfixed_const_8(0);
min_mem_eff.full = dfixed_const_8(0);
/* get modes */
if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
@ -2784,28 +2784,28 @@ void r100_bandwidth_update(struct radeon_device *rdev)
mclk_ff = rdev->pm.mclk;
temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
temp_ff.full = rfixed_const(temp);
mem_bw.full = rfixed_mul(mclk_ff, temp_ff);
temp_ff.full = dfixed_const(temp);
mem_bw.full = dfixed_mul(mclk_ff, temp_ff);
pix_clk.full = 0;
pix_clk2.full = 0;
peak_disp_bw.full = 0;
if (mode1) {
temp_ff.full = rfixed_const(1000);
pix_clk.full = rfixed_const(mode1->clock); /* convert to fixed point */
pix_clk.full = rfixed_div(pix_clk, temp_ff);
temp_ff.full = rfixed_const(pixel_bytes1);
peak_disp_bw.full += rfixed_mul(pix_clk, temp_ff);
temp_ff.full = dfixed_const(1000);
pix_clk.full = dfixed_const(mode1->clock); /* convert to fixed point */
pix_clk.full = dfixed_div(pix_clk, temp_ff);
temp_ff.full = dfixed_const(pixel_bytes1);
peak_disp_bw.full += dfixed_mul(pix_clk, temp_ff);
}
if (mode2) {
temp_ff.full = rfixed_const(1000);
pix_clk2.full = rfixed_const(mode2->clock); /* convert to fixed point */
pix_clk2.full = rfixed_div(pix_clk2, temp_ff);
temp_ff.full = rfixed_const(pixel_bytes2);
peak_disp_bw.full += rfixed_mul(pix_clk2, temp_ff);
temp_ff.full = dfixed_const(1000);
pix_clk2.full = dfixed_const(mode2->clock); /* convert to fixed point */
pix_clk2.full = dfixed_div(pix_clk2, temp_ff);
temp_ff.full = dfixed_const(pixel_bytes2);
peak_disp_bw.full += dfixed_mul(pix_clk2, temp_ff);
}
mem_bw.full = rfixed_mul(mem_bw, min_mem_eff);
mem_bw.full = dfixed_mul(mem_bw, min_mem_eff);
if (peak_disp_bw.full >= mem_bw.full) {
DRM_ERROR("You may not have enough display bandwidth for current mode\n"
"If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
@ -2847,9 +2847,9 @@ void r100_bandwidth_update(struct radeon_device *rdev)
mem_tras = ((temp >> 12) & 0xf) + 4;
}
/* convert to FF */
trcd_ff.full = rfixed_const(mem_trcd);
trp_ff.full = rfixed_const(mem_trp);
tras_ff.full = rfixed_const(mem_tras);
trcd_ff.full = dfixed_const(mem_trcd);
trp_ff.full = dfixed_const(mem_trp);
tras_ff.full = dfixed_const(mem_tras);
/* Get values from the MEM_SDRAM_MODE_REG register...converting its */
temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
@ -2867,7 +2867,7 @@ void r100_bandwidth_update(struct radeon_device *rdev)
/* extra cas latency stored in bits 23-25 0-4 clocks */
data = (temp >> 23) & 0x7;
if (data < 5)
tcas_ff.full += rfixed_const(data);
tcas_ff.full += dfixed_const(data);
}
if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
@ -2904,72 +2904,72 @@ void r100_bandwidth_update(struct radeon_device *rdev)
if (rdev->flags & RADEON_IS_AGP) {
fixed20_12 agpmode_ff;
agpmode_ff.full = rfixed_const(radeon_agpmode);
temp_ff.full = rfixed_const_666(16);
sclk_eff_ff.full -= rfixed_mul(agpmode_ff, temp_ff);
agpmode_ff.full = dfixed_const(radeon_agpmode);
temp_ff.full = dfixed_const_666(16);
sclk_eff_ff.full -= dfixed_mul(agpmode_ff, temp_ff);
}
/* TODO PCIE lanes may affect this - agpmode == 16?? */
if (ASIC_IS_R300(rdev)) {
sclk_delay_ff.full = rfixed_const(250);
sclk_delay_ff.full = dfixed_const(250);
} else {
if ((rdev->family == CHIP_RV100) ||
rdev->flags & RADEON_IS_IGP) {
if (rdev->mc.vram_is_ddr)
sclk_delay_ff.full = rfixed_const(41);
sclk_delay_ff.full = dfixed_const(41);
else
sclk_delay_ff.full = rfixed_const(33);
sclk_delay_ff.full = dfixed_const(33);
} else {
if (rdev->mc.vram_width == 128)
sclk_delay_ff.full = rfixed_const(57);
sclk_delay_ff.full = dfixed_const(57);
else
sclk_delay_ff.full = rfixed_const(41);
sclk_delay_ff.full = dfixed_const(41);
}
}
mc_latency_sclk.full = rfixed_div(sclk_delay_ff, sclk_eff_ff);
mc_latency_sclk.full = dfixed_div(sclk_delay_ff, sclk_eff_ff);
if (rdev->mc.vram_is_ddr) {
if (rdev->mc.vram_width == 32) {
k1.full = rfixed_const(40);
k1.full = dfixed_const(40);
c = 3;
} else {
k1.full = rfixed_const(20);
k1.full = dfixed_const(20);
c = 1;
}
} else {
k1.full = rfixed_const(40);
k1.full = dfixed_const(40);
c = 3;
}
temp_ff.full = rfixed_const(2);
mc_latency_mclk.full = rfixed_mul(trcd_ff, temp_ff);
temp_ff.full = rfixed_const(c);
mc_latency_mclk.full += rfixed_mul(tcas_ff, temp_ff);
temp_ff.full = rfixed_const(4);
mc_latency_mclk.full += rfixed_mul(tras_ff, temp_ff);
mc_latency_mclk.full += rfixed_mul(trp_ff, temp_ff);
temp_ff.full = dfixed_const(2);
mc_latency_mclk.full = dfixed_mul(trcd_ff, temp_ff);
temp_ff.full = dfixed_const(c);
mc_latency_mclk.full += dfixed_mul(tcas_ff, temp_ff);
temp_ff.full = dfixed_const(4);
mc_latency_mclk.full += dfixed_mul(tras_ff, temp_ff);
mc_latency_mclk.full += dfixed_mul(trp_ff, temp_ff);
mc_latency_mclk.full += k1.full;
mc_latency_mclk.full = rfixed_div(mc_latency_mclk, mclk_ff);
mc_latency_mclk.full += rfixed_div(temp_ff, sclk_eff_ff);
mc_latency_mclk.full = dfixed_div(mc_latency_mclk, mclk_ff);
mc_latency_mclk.full += dfixed_div(temp_ff, sclk_eff_ff);
/*
HW cursor time assuming worst case of full size colour cursor.
*/
temp_ff.full = rfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
temp_ff.full = dfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
temp_ff.full += trcd_ff.full;
if (temp_ff.full < tras_ff.full)
temp_ff.full = tras_ff.full;
cur_latency_mclk.full = rfixed_div(temp_ff, mclk_ff);
cur_latency_mclk.full = dfixed_div(temp_ff, mclk_ff);
temp_ff.full = rfixed_const(cur_size);
cur_latency_sclk.full = rfixed_div(temp_ff, sclk_eff_ff);
temp_ff.full = dfixed_const(cur_size);
cur_latency_sclk.full = dfixed_div(temp_ff, sclk_eff_ff);
/*
Find the total latency for the display data.
*/
disp_latency_overhead.full = rfixed_const(8);
disp_latency_overhead.full = rfixed_div(disp_latency_overhead, sclk_ff);
disp_latency_overhead.full = dfixed_const(8);
disp_latency_overhead.full = dfixed_div(disp_latency_overhead, sclk_ff);
mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
@ -2997,16 +2997,16 @@ void r100_bandwidth_update(struct radeon_device *rdev)
/*
Find the drain rate of the display buffer.
*/
temp_ff.full = rfixed_const((16/pixel_bytes1));
disp_drain_rate.full = rfixed_div(pix_clk, temp_ff);
temp_ff.full = dfixed_const((16/pixel_bytes1));
disp_drain_rate.full = dfixed_div(pix_clk, temp_ff);
/*
Find the critical point of the display buffer.
*/
crit_point_ff.full = rfixed_mul(disp_drain_rate, disp_latency);
crit_point_ff.full += rfixed_const_half(0);
crit_point_ff.full = dfixed_mul(disp_drain_rate, disp_latency);
crit_point_ff.full += dfixed_const_half(0);
critical_point = rfixed_trunc(crit_point_ff);
critical_point = dfixed_trunc(crit_point_ff);
if (rdev->disp_priority == 2) {
critical_point = 0;
@ -3077,8 +3077,8 @@ void r100_bandwidth_update(struct radeon_device *rdev)
/*
Find the drain rate of the display buffer.
*/
temp_ff.full = rfixed_const((16/pixel_bytes2));
disp_drain_rate2.full = rfixed_div(pix_clk2, temp_ff);
temp_ff.full = dfixed_const((16/pixel_bytes2));
disp_drain_rate2.full = dfixed_div(pix_clk2, temp_ff);
grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
@ -3099,8 +3099,8 @@ void r100_bandwidth_update(struct radeon_device *rdev)
critical_point2 = 0;
else {
temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
temp_ff.full = rfixed_const(temp);
temp_ff.full = rfixed_mul(mclk_ff, temp_ff);
temp_ff.full = dfixed_const(temp);
temp_ff.full = dfixed_mul(mclk_ff, temp_ff);
if (sclk_ff.full < temp_ff.full)
temp_ff.full = sclk_ff.full;
@ -3108,15 +3108,15 @@ void r100_bandwidth_update(struct radeon_device *rdev)
if (mode1) {
temp_ff.full = read_return_rate.full - disp_drain_rate.full;
time_disp1_drop_priority.full = rfixed_div(crit_point_ff, temp_ff);
time_disp1_drop_priority.full = dfixed_div(crit_point_ff, temp_ff);
} else {
time_disp1_drop_priority.full = 0;
}
crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
crit_point_ff.full = rfixed_mul(crit_point_ff, disp_drain_rate2);
crit_point_ff.full += rfixed_const_half(0);
crit_point_ff.full = dfixed_mul(crit_point_ff, disp_drain_rate2);
crit_point_ff.full += dfixed_const_half(0);
critical_point2 = rfixed_trunc(crit_point_ff);
critical_point2 = dfixed_trunc(crit_point_ff);
if (rdev->disp_priority == 2) {
critical_point2 = 0;

View File

@ -299,24 +299,24 @@ void radeon_update_bandwidth_info(struct radeon_device *rdev)
sclk = radeon_get_engine_clock(rdev);
mclk = rdev->clock.default_mclk;
a.full = rfixed_const(100);
rdev->pm.sclk.full = rfixed_const(sclk);
rdev->pm.sclk.full = rfixed_div(rdev->pm.sclk, a);
rdev->pm.mclk.full = rfixed_const(mclk);
rdev->pm.mclk.full = rfixed_div(rdev->pm.mclk, a);
a.full = dfixed_const(100);
rdev->pm.sclk.full = dfixed_const(sclk);
rdev->pm.sclk.full = dfixed_div(rdev->pm.sclk, a);
rdev->pm.mclk.full = dfixed_const(mclk);
rdev->pm.mclk.full = dfixed_div(rdev->pm.mclk, a);
a.full = rfixed_const(16);
a.full = dfixed_const(16);
/* core_bandwidth = sclk(Mhz) * 16 */
rdev->pm.core_bandwidth.full = rfixed_div(rdev->pm.sclk, a);
rdev->pm.core_bandwidth.full = dfixed_div(rdev->pm.sclk, a);
} else {
sclk = radeon_get_engine_clock(rdev);
mclk = radeon_get_memory_clock(rdev);
a.full = rfixed_const(100);
rdev->pm.sclk.full = rfixed_const(sclk);
rdev->pm.sclk.full = rfixed_div(rdev->pm.sclk, a);
rdev->pm.mclk.full = rfixed_const(mclk);
rdev->pm.mclk.full = rfixed_div(rdev->pm.mclk, a);
a.full = dfixed_const(100);
rdev->pm.sclk.full = dfixed_const(sclk);
rdev->pm.sclk.full = dfixed_div(rdev->pm.sclk, a);
rdev->pm.mclk.full = dfixed_const(mclk);
rdev->pm.mclk.full = dfixed_div(rdev->pm.mclk, a);
}
}

View File

@ -633,37 +633,37 @@ calc_fb_div(struct radeon_pll *pll,
vco_freq = freq * post_div;
/* feedback_divider = vco_freq * ref_div / pll->reference_freq; */
a.full = rfixed_const(pll->reference_freq);
feedback_divider.full = rfixed_const(vco_freq);
feedback_divider.full = rfixed_div(feedback_divider, a);
a.full = rfixed_const(ref_div);
feedback_divider.full = rfixed_mul(feedback_divider, a);
a.full = dfixed_const(pll->reference_freq);
feedback_divider.full = dfixed_const(vco_freq);
feedback_divider.full = dfixed_div(feedback_divider, a);
a.full = dfixed_const(ref_div);
feedback_divider.full = dfixed_mul(feedback_divider, a);
if (pll->flags & RADEON_PLL_USE_FRAC_FB_DIV) {
/* feedback_divider = floor((feedback_divider * 10.0) + 0.5) * 0.1; */
a.full = rfixed_const(10);
feedback_divider.full = rfixed_mul(feedback_divider, a);
feedback_divider.full += rfixed_const_half(0);
feedback_divider.full = rfixed_floor(feedback_divider);
feedback_divider.full = rfixed_div(feedback_divider, a);
a.full = dfixed_const(10);
feedback_divider.full = dfixed_mul(feedback_divider, a);
feedback_divider.full += dfixed_const_half(0);
feedback_divider.full = dfixed_floor(feedback_divider);
feedback_divider.full = dfixed_div(feedback_divider, a);
/* *fb_div = floor(feedback_divider); */
a.full = rfixed_floor(feedback_divider);
*fb_div = rfixed_trunc(a);
a.full = dfixed_floor(feedback_divider);
*fb_div = dfixed_trunc(a);
/* *fb_div_frac = fmod(feedback_divider, 1.0) * 10.0; */
a.full = rfixed_const(10);
b.full = rfixed_mul(feedback_divider, a);
a.full = dfixed_const(10);
b.full = dfixed_mul(feedback_divider, a);
feedback_divider.full = rfixed_floor(feedback_divider);
feedback_divider.full = rfixed_mul(feedback_divider, a);
feedback_divider.full = dfixed_floor(feedback_divider);
feedback_divider.full = dfixed_mul(feedback_divider, a);
feedback_divider.full = b.full - feedback_divider.full;
*fb_div_frac = rfixed_trunc(feedback_divider);
*fb_div_frac = dfixed_trunc(feedback_divider);
} else {
/* *fb_div = floor(feedback_divider + 0.5); */
feedback_divider.full += rfixed_const_half(0);
feedback_divider.full = rfixed_floor(feedback_divider);
feedback_divider.full += dfixed_const_half(0);
feedback_divider.full = dfixed_floor(feedback_divider);
*fb_div = rfixed_trunc(feedback_divider);
*fb_div = dfixed_trunc(feedback_divider);
*fb_div_frac = 0;
}
@ -693,10 +693,10 @@ calc_fb_ref_div(struct radeon_pll *pll,
pll_out_max = pll->pll_out_max;
}
ffreq.full = rfixed_const(freq);
ffreq.full = dfixed_const(freq);
/* max_error = ffreq * 0.0025; */
a.full = rfixed_const(400);
max_error.full = rfixed_div(ffreq, a);
a.full = dfixed_const(400);
max_error.full = dfixed_div(ffreq, a);
for ((*ref_div) = pll->min_ref_div; (*ref_div) < pll->max_ref_div; ++(*ref_div)) {
if (calc_fb_div(pll, freq, post_div, (*ref_div), fb_div, fb_div_frac)) {
@ -707,9 +707,9 @@ calc_fb_ref_div(struct radeon_pll *pll,
continue;
/* pll_out = vco / post_div; */
a.full = rfixed_const(post_div);
pll_out.full = rfixed_const(vco);
pll_out.full = rfixed_div(pll_out, a);
a.full = dfixed_const(post_div);
pll_out.full = dfixed_const(vco);
pll_out.full = dfixed_div(pll_out, a);
if (pll_out.full >= ffreq.full) {
error.full = pll_out.full - ffreq.full;
@ -1099,15 +1099,15 @@ bool radeon_crtc_scaling_mode_fixup(struct drm_crtc *crtc,
}
if (radeon_crtc->rmx_type != RMX_OFF) {
fixed20_12 a, b;
a.full = rfixed_const(crtc->mode.vdisplay);
b.full = rfixed_const(radeon_crtc->native_mode.hdisplay);
radeon_crtc->vsc.full = rfixed_div(a, b);
a.full = rfixed_const(crtc->mode.hdisplay);
b.full = rfixed_const(radeon_crtc->native_mode.vdisplay);
radeon_crtc->hsc.full = rfixed_div(a, b);
a.full = dfixed_const(crtc->mode.vdisplay);
b.full = dfixed_const(radeon_crtc->native_mode.hdisplay);
radeon_crtc->vsc.full = dfixed_div(a, b);
a.full = dfixed_const(crtc->mode.hdisplay);
b.full = dfixed_const(radeon_crtc->native_mode.vdisplay);
radeon_crtc->hsc.full = dfixed_div(a, b);
} else {
radeon_crtc->vsc.full = rfixed_const(1);
radeon_crtc->hsc.full = rfixed_const(1);
radeon_crtc->vsc.full = dfixed_const(1);
radeon_crtc->hsc.full = dfixed_const(1);
}
return true;
}

View File

@ -26,7 +26,7 @@
#include <drm/drmP.h>
#include <drm/drm_crtc_helper.h>
#include <drm/radeon_drm.h>
#include "radeon_fixed.h"
#include <drm/drm_fixed.h>
#include "radeon.h"
#include "atom.h"

View File

@ -34,10 +34,10 @@
#include <drm_mode.h>
#include <drm_edid.h>
#include <drm_dp_helper.h>
#include <drm_fixed.h>
#include <linux/i2c.h>
#include <linux/i2c-id.h>
#include <linux/i2c-algo-bit.h>
#include "radeon_fixed.h"
struct radeon_bo;
struct radeon_device;

View File

@ -76,59 +76,59 @@ void rs690_pm_info(struct radeon_device *rdev)
/* Get various system informations from bios */
switch (crev) {
case 1:
tmp.full = rfixed_const(100);
rdev->pm.igp_sideport_mclk.full = rfixed_const(info->info.ulBootUpMemoryClock);
rdev->pm.igp_sideport_mclk.full = rfixed_div(rdev->pm.igp_sideport_mclk, tmp);
rdev->pm.igp_system_mclk.full = rfixed_const(le16_to_cpu(info->info.usK8MemoryClock));
rdev->pm.igp_ht_link_clk.full = rfixed_const(le16_to_cpu(info->info.usFSBClock));
rdev->pm.igp_ht_link_width.full = rfixed_const(info->info.ucHTLinkWidth);
tmp.full = dfixed_const(100);
rdev->pm.igp_sideport_mclk.full = dfixed_const(info->info.ulBootUpMemoryClock);
rdev->pm.igp_sideport_mclk.full = dfixed_div(rdev->pm.igp_sideport_mclk, tmp);
rdev->pm.igp_system_mclk.full = dfixed_const(le16_to_cpu(info->info.usK8MemoryClock));
rdev->pm.igp_ht_link_clk.full = dfixed_const(le16_to_cpu(info->info.usFSBClock));
rdev->pm.igp_ht_link_width.full = dfixed_const(info->info.ucHTLinkWidth);
break;
case 2:
tmp.full = rfixed_const(100);
rdev->pm.igp_sideport_mclk.full = rfixed_const(info->info_v2.ulBootUpSidePortClock);
rdev->pm.igp_sideport_mclk.full = rfixed_div(rdev->pm.igp_sideport_mclk, tmp);
rdev->pm.igp_system_mclk.full = rfixed_const(info->info_v2.ulBootUpUMAClock);
rdev->pm.igp_system_mclk.full = rfixed_div(rdev->pm.igp_system_mclk, tmp);
rdev->pm.igp_ht_link_clk.full = rfixed_const(info->info_v2.ulHTLinkFreq);
rdev->pm.igp_ht_link_clk.full = rfixed_div(rdev->pm.igp_ht_link_clk, tmp);
rdev->pm.igp_ht_link_width.full = rfixed_const(le16_to_cpu(info->info_v2.usMinHTLinkWidth));
tmp.full = dfixed_const(100);
rdev->pm.igp_sideport_mclk.full = dfixed_const(info->info_v2.ulBootUpSidePortClock);
rdev->pm.igp_sideport_mclk.full = dfixed_div(rdev->pm.igp_sideport_mclk, tmp);
rdev->pm.igp_system_mclk.full = dfixed_const(info->info_v2.ulBootUpUMAClock);
rdev->pm.igp_system_mclk.full = dfixed_div(rdev->pm.igp_system_mclk, tmp);
rdev->pm.igp_ht_link_clk.full = dfixed_const(info->info_v2.ulHTLinkFreq);
rdev->pm.igp_ht_link_clk.full = dfixed_div(rdev->pm.igp_ht_link_clk, tmp);
rdev->pm.igp_ht_link_width.full = dfixed_const(le16_to_cpu(info->info_v2.usMinHTLinkWidth));
break;
default:
tmp.full = rfixed_const(100);
tmp.full = dfixed_const(100);
/* We assume the slower possible clock ie worst case */
/* DDR 333Mhz */
rdev->pm.igp_sideport_mclk.full = rfixed_const(333);
rdev->pm.igp_sideport_mclk.full = dfixed_const(333);
/* FIXME: system clock ? */
rdev->pm.igp_system_mclk.full = rfixed_const(100);
rdev->pm.igp_system_mclk.full = rfixed_div(rdev->pm.igp_system_mclk, tmp);
rdev->pm.igp_ht_link_clk.full = rfixed_const(200);
rdev->pm.igp_ht_link_width.full = rfixed_const(8);
rdev->pm.igp_system_mclk.full = dfixed_const(100);
rdev->pm.igp_system_mclk.full = dfixed_div(rdev->pm.igp_system_mclk, tmp);
rdev->pm.igp_ht_link_clk.full = dfixed_const(200);
rdev->pm.igp_ht_link_width.full = dfixed_const(8);
DRM_ERROR("No integrated system info for your GPU, using safe default\n");
break;
}
} else {
tmp.full = rfixed_const(100);
tmp.full = dfixed_const(100);
/* We assume the slower possible clock ie worst case */
/* DDR 333Mhz */
rdev->pm.igp_sideport_mclk.full = rfixed_const(333);
rdev->pm.igp_sideport_mclk.full = dfixed_const(333);
/* FIXME: system clock ? */
rdev->pm.igp_system_mclk.full = rfixed_const(100);
rdev->pm.igp_system_mclk.full = rfixed_div(rdev->pm.igp_system_mclk, tmp);
rdev->pm.igp_ht_link_clk.full = rfixed_const(200);
rdev->pm.igp_ht_link_width.full = rfixed_const(8);
rdev->pm.igp_system_mclk.full = dfixed_const(100);
rdev->pm.igp_system_mclk.full = dfixed_div(rdev->pm.igp_system_mclk, tmp);
rdev->pm.igp_ht_link_clk.full = dfixed_const(200);
rdev->pm.igp_ht_link_width.full = dfixed_const(8);
DRM_ERROR("No integrated system info for your GPU, using safe default\n");
}
/* Compute various bandwidth */
/* k8_bandwidth = (memory_clk / 2) * 2 * 8 * 0.5 = memory_clk * 4 */
tmp.full = rfixed_const(4);
rdev->pm.k8_bandwidth.full = rfixed_mul(rdev->pm.igp_system_mclk, tmp);
tmp.full = dfixed_const(4);
rdev->pm.k8_bandwidth.full = dfixed_mul(rdev->pm.igp_system_mclk, tmp);
/* ht_bandwidth = ht_clk * 2 * ht_width / 8 * 0.8
* = ht_clk * ht_width / 5
*/
tmp.full = rfixed_const(5);
rdev->pm.ht_bandwidth.full = rfixed_mul(rdev->pm.igp_ht_link_clk,
tmp.full = dfixed_const(5);
rdev->pm.ht_bandwidth.full = dfixed_mul(rdev->pm.igp_ht_link_clk,
rdev->pm.igp_ht_link_width);
rdev->pm.ht_bandwidth.full = rfixed_div(rdev->pm.ht_bandwidth, tmp);
rdev->pm.ht_bandwidth.full = dfixed_div(rdev->pm.ht_bandwidth, tmp);
if (tmp.full < rdev->pm.max_bandwidth.full) {
/* HT link is a limiting factor */
rdev->pm.max_bandwidth.full = tmp.full;
@ -136,10 +136,10 @@ void rs690_pm_info(struct radeon_device *rdev)
/* sideport_bandwidth = (sideport_clk / 2) * 2 * 2 * 0.7
* = (sideport_clk * 14) / 10
*/
tmp.full = rfixed_const(14);
rdev->pm.sideport_bandwidth.full = rfixed_mul(rdev->pm.igp_sideport_mclk, tmp);
tmp.full = rfixed_const(10);
rdev->pm.sideport_bandwidth.full = rfixed_div(rdev->pm.sideport_bandwidth, tmp);
tmp.full = dfixed_const(14);
rdev->pm.sideport_bandwidth.full = dfixed_mul(rdev->pm.igp_sideport_mclk, tmp);
tmp.full = dfixed_const(10);
rdev->pm.sideport_bandwidth.full = dfixed_div(rdev->pm.sideport_bandwidth, tmp);
}
void rs690_mc_init(struct radeon_device *rdev)
@ -239,20 +239,20 @@ void rs690_crtc_bandwidth_compute(struct radeon_device *rdev,
return;
}
if (crtc->vsc.full > rfixed_const(2))
wm->num_line_pair.full = rfixed_const(2);
if (crtc->vsc.full > dfixed_const(2))
wm->num_line_pair.full = dfixed_const(2);
else
wm->num_line_pair.full = rfixed_const(1);
wm->num_line_pair.full = dfixed_const(1);
b.full = rfixed_const(mode->crtc_hdisplay);
c.full = rfixed_const(256);
a.full = rfixed_div(b, c);
request_fifo_depth.full = rfixed_mul(a, wm->num_line_pair);
request_fifo_depth.full = rfixed_ceil(request_fifo_depth);
if (a.full < rfixed_const(4)) {
b.full = dfixed_const(mode->crtc_hdisplay);
c.full = dfixed_const(256);
a.full = dfixed_div(b, c);
request_fifo_depth.full = dfixed_mul(a, wm->num_line_pair);
request_fifo_depth.full = dfixed_ceil(request_fifo_depth);
if (a.full < dfixed_const(4)) {
wm->lb_request_fifo_depth = 4;
} else {
wm->lb_request_fifo_depth = rfixed_trunc(request_fifo_depth);
wm->lb_request_fifo_depth = dfixed_trunc(request_fifo_depth);
}
/* Determine consumption rate
@ -261,23 +261,23 @@ void rs690_crtc_bandwidth_compute(struct radeon_device *rdev,
* vsc = vertical scaling ratio, defined as source/destination
* hsc = horizontal scaling ration, defined as source/destination
*/
a.full = rfixed_const(mode->clock);
b.full = rfixed_const(1000);
a.full = rfixed_div(a, b);
pclk.full = rfixed_div(b, a);
a.full = dfixed_const(mode->clock);
b.full = dfixed_const(1000);
a.full = dfixed_div(a, b);
pclk.full = dfixed_div(b, a);
if (crtc->rmx_type != RMX_OFF) {
b.full = rfixed_const(2);
b.full = dfixed_const(2);
if (crtc->vsc.full > b.full)
b.full = crtc->vsc.full;
b.full = rfixed_mul(b, crtc->hsc);
c.full = rfixed_const(2);
b.full = rfixed_div(b, c);
consumption_time.full = rfixed_div(pclk, b);
b.full = dfixed_mul(b, crtc->hsc);
c.full = dfixed_const(2);
b.full = dfixed_div(b, c);
consumption_time.full = dfixed_div(pclk, b);
} else {
consumption_time.full = pclk.full;
}
a.full = rfixed_const(1);
wm->consumption_rate.full = rfixed_div(a, consumption_time);
a.full = dfixed_const(1);
wm->consumption_rate.full = dfixed_div(a, consumption_time);
/* Determine line time
@ -285,18 +285,18 @@ void rs690_crtc_bandwidth_compute(struct radeon_device *rdev,
* LineTime = total number of horizontal pixels
* pclk = pixel clock period(ns)
*/
a.full = rfixed_const(crtc->base.mode.crtc_htotal);
line_time.full = rfixed_mul(a, pclk);
a.full = dfixed_const(crtc->base.mode.crtc_htotal);
line_time.full = dfixed_mul(a, pclk);
/* Determine active time
* ActiveTime = time of active region of display within one line,
* hactive = total number of horizontal active pixels
* htotal = total number of horizontal pixels
*/
a.full = rfixed_const(crtc->base.mode.crtc_htotal);
b.full = rfixed_const(crtc->base.mode.crtc_hdisplay);
wm->active_time.full = rfixed_mul(line_time, b);
wm->active_time.full = rfixed_div(wm->active_time, a);
a.full = dfixed_const(crtc->base.mode.crtc_htotal);
b.full = dfixed_const(crtc->base.mode.crtc_hdisplay);
wm->active_time.full = dfixed_mul(line_time, b);
wm->active_time.full = dfixed_div(wm->active_time, a);
/* Maximun bandwidth is the minimun bandwidth of all component */
rdev->pm.max_bandwidth = rdev->pm.core_bandwidth;
@ -304,8 +304,8 @@ void rs690_crtc_bandwidth_compute(struct radeon_device *rdev,
if (rdev->pm.max_bandwidth.full > rdev->pm.sideport_bandwidth.full &&
rdev->pm.sideport_bandwidth.full)
rdev->pm.max_bandwidth = rdev->pm.sideport_bandwidth;
read_delay_latency.full = rfixed_const(370 * 800 * 1000);
read_delay_latency.full = rfixed_div(read_delay_latency,
read_delay_latency.full = dfixed_const(370 * 800 * 1000);
read_delay_latency.full = dfixed_div(read_delay_latency,
rdev->pm.igp_sideport_mclk);
} else {
if (rdev->pm.max_bandwidth.full > rdev->pm.k8_bandwidth.full &&
@ -314,23 +314,23 @@ void rs690_crtc_bandwidth_compute(struct radeon_device *rdev,
if (rdev->pm.max_bandwidth.full > rdev->pm.ht_bandwidth.full &&
rdev->pm.ht_bandwidth.full)
rdev->pm.max_bandwidth = rdev->pm.ht_bandwidth;
read_delay_latency.full = rfixed_const(5000);
read_delay_latency.full = dfixed_const(5000);
}
/* sclk = system clocks(ns) = 1000 / max_bandwidth / 16 */
a.full = rfixed_const(16);
rdev->pm.sclk.full = rfixed_mul(rdev->pm.max_bandwidth, a);
a.full = rfixed_const(1000);
rdev->pm.sclk.full = rfixed_div(a, rdev->pm.sclk);
a.full = dfixed_const(16);
rdev->pm.sclk.full = dfixed_mul(rdev->pm.max_bandwidth, a);
a.full = dfixed_const(1000);
rdev->pm.sclk.full = dfixed_div(a, rdev->pm.sclk);
/* Determine chunk time
* ChunkTime = the time it takes the DCP to send one chunk of data
* to the LB which consists of pipeline delay and inter chunk gap
* sclk = system clock(ns)
*/
a.full = rfixed_const(256 * 13);
chunk_time.full = rfixed_mul(rdev->pm.sclk, a);
a.full = rfixed_const(10);
chunk_time.full = rfixed_div(chunk_time, a);
a.full = dfixed_const(256 * 13);
chunk_time.full = dfixed_mul(rdev->pm.sclk, a);
a.full = dfixed_const(10);
chunk_time.full = dfixed_div(chunk_time, a);
/* Determine the worst case latency
* NumLinePair = Number of line pairs to request(1=2 lines, 2=4 lines)
@ -340,13 +340,13 @@ void rs690_crtc_bandwidth_compute(struct radeon_device *rdev,
* ChunkTime = time it takes the DCP to send one chunk of data to the LB
* which consists of pipeline delay and inter chunk gap
*/
if (rfixed_trunc(wm->num_line_pair) > 1) {
a.full = rfixed_const(3);
wm->worst_case_latency.full = rfixed_mul(a, chunk_time);
if (dfixed_trunc(wm->num_line_pair) > 1) {
a.full = dfixed_const(3);
wm->worst_case_latency.full = dfixed_mul(a, chunk_time);
wm->worst_case_latency.full += read_delay_latency.full;
} else {
a.full = rfixed_const(2);
wm->worst_case_latency.full = rfixed_mul(a, chunk_time);
a.full = dfixed_const(2);
wm->worst_case_latency.full = dfixed_mul(a, chunk_time);
wm->worst_case_latency.full += read_delay_latency.full;
}
@ -360,34 +360,34 @@ void rs690_crtc_bandwidth_compute(struct radeon_device *rdev,
* of data to the LB which consists of
* pipeline delay and inter chunk gap
*/
if ((2+wm->lb_request_fifo_depth) >= rfixed_trunc(request_fifo_depth)) {
if ((2+wm->lb_request_fifo_depth) >= dfixed_trunc(request_fifo_depth)) {
tolerable_latency.full = line_time.full;
} else {
tolerable_latency.full = rfixed_const(wm->lb_request_fifo_depth - 2);
tolerable_latency.full = dfixed_const(wm->lb_request_fifo_depth - 2);
tolerable_latency.full = request_fifo_depth.full - tolerable_latency.full;
tolerable_latency.full = rfixed_mul(tolerable_latency, chunk_time);
tolerable_latency.full = dfixed_mul(tolerable_latency, chunk_time);
tolerable_latency.full = line_time.full - tolerable_latency.full;
}
/* We assume worst case 32bits (4 bytes) */
wm->dbpp.full = rfixed_const(4 * 8);
wm->dbpp.full = dfixed_const(4 * 8);
/* Determine the maximum priority mark
* width = viewport width in pixels
*/
a.full = rfixed_const(16);
wm->priority_mark_max.full = rfixed_const(crtc->base.mode.crtc_hdisplay);
wm->priority_mark_max.full = rfixed_div(wm->priority_mark_max, a);
wm->priority_mark_max.full = rfixed_ceil(wm->priority_mark_max);
a.full = dfixed_const(16);
wm->priority_mark_max.full = dfixed_const(crtc->base.mode.crtc_hdisplay);
wm->priority_mark_max.full = dfixed_div(wm->priority_mark_max, a);
wm->priority_mark_max.full = dfixed_ceil(wm->priority_mark_max);
/* Determine estimated width */
estimated_width.full = tolerable_latency.full - wm->worst_case_latency.full;
estimated_width.full = rfixed_div(estimated_width, consumption_time);
if (rfixed_trunc(estimated_width) > crtc->base.mode.crtc_hdisplay) {
wm->priority_mark.full = rfixed_const(10);
estimated_width.full = dfixed_div(estimated_width, consumption_time);
if (dfixed_trunc(estimated_width) > crtc->base.mode.crtc_hdisplay) {
wm->priority_mark.full = dfixed_const(10);
} else {
a.full = rfixed_const(16);
wm->priority_mark.full = rfixed_div(estimated_width, a);
wm->priority_mark.full = rfixed_ceil(wm->priority_mark);
a.full = dfixed_const(16);
wm->priority_mark.full = dfixed_div(estimated_width, a);
wm->priority_mark.full = dfixed_ceil(wm->priority_mark);
wm->priority_mark.full = wm->priority_mark_max.full - wm->priority_mark.full;
}
}
@ -439,58 +439,58 @@ void rs690_bandwidth_update(struct radeon_device *rdev)
WREG32(R_006D58_LB_MAX_REQ_OUTSTANDING, tmp);
if (mode0 && mode1) {
if (rfixed_trunc(wm0.dbpp) > 64)
a.full = rfixed_mul(wm0.dbpp, wm0.num_line_pair);
if (dfixed_trunc(wm0.dbpp) > 64)
a.full = dfixed_mul(wm0.dbpp, wm0.num_line_pair);
else
a.full = wm0.num_line_pair.full;
if (rfixed_trunc(wm1.dbpp) > 64)
b.full = rfixed_mul(wm1.dbpp, wm1.num_line_pair);
if (dfixed_trunc(wm1.dbpp) > 64)
b.full = dfixed_mul(wm1.dbpp, wm1.num_line_pair);
else
b.full = wm1.num_line_pair.full;
a.full += b.full;
fill_rate.full = rfixed_div(wm0.sclk, a);
fill_rate.full = dfixed_div(wm0.sclk, a);
if (wm0.consumption_rate.full > fill_rate.full) {
b.full = wm0.consumption_rate.full - fill_rate.full;
b.full = rfixed_mul(b, wm0.active_time);
a.full = rfixed_mul(wm0.worst_case_latency,
b.full = dfixed_mul(b, wm0.active_time);
a.full = dfixed_mul(wm0.worst_case_latency,
wm0.consumption_rate);
a.full = a.full + b.full;
b.full = rfixed_const(16 * 1000);
priority_mark02.full = rfixed_div(a, b);
b.full = dfixed_const(16 * 1000);
priority_mark02.full = dfixed_div(a, b);
} else {
a.full = rfixed_mul(wm0.worst_case_latency,
a.full = dfixed_mul(wm0.worst_case_latency,
wm0.consumption_rate);
b.full = rfixed_const(16 * 1000);
priority_mark02.full = rfixed_div(a, b);
b.full = dfixed_const(16 * 1000);
priority_mark02.full = dfixed_div(a, b);
}
if (wm1.consumption_rate.full > fill_rate.full) {
b.full = wm1.consumption_rate.full - fill_rate.full;
b.full = rfixed_mul(b, wm1.active_time);
a.full = rfixed_mul(wm1.worst_case_latency,
b.full = dfixed_mul(b, wm1.active_time);
a.full = dfixed_mul(wm1.worst_case_latency,
wm1.consumption_rate);
a.full = a.full + b.full;
b.full = rfixed_const(16 * 1000);
priority_mark12.full = rfixed_div(a, b);
b.full = dfixed_const(16 * 1000);
priority_mark12.full = dfixed_div(a, b);
} else {
a.full = rfixed_mul(wm1.worst_case_latency,
a.full = dfixed_mul(wm1.worst_case_latency,
wm1.consumption_rate);
b.full = rfixed_const(16 * 1000);
priority_mark12.full = rfixed_div(a, b);
b.full = dfixed_const(16 * 1000);
priority_mark12.full = dfixed_div(a, b);
}
if (wm0.priority_mark.full > priority_mark02.full)
priority_mark02.full = wm0.priority_mark.full;
if (rfixed_trunc(priority_mark02) < 0)
if (dfixed_trunc(priority_mark02) < 0)
priority_mark02.full = 0;
if (wm0.priority_mark_max.full > priority_mark02.full)
priority_mark02.full = wm0.priority_mark_max.full;
if (wm1.priority_mark.full > priority_mark12.full)
priority_mark12.full = wm1.priority_mark.full;
if (rfixed_trunc(priority_mark12) < 0)
if (dfixed_trunc(priority_mark12) < 0)
priority_mark12.full = 0;
if (wm1.priority_mark_max.full > priority_mark12.full)
priority_mark12.full = wm1.priority_mark_max.full;
d1mode_priority_a_cnt = rfixed_trunc(priority_mark02);
d2mode_priority_a_cnt = rfixed_trunc(priority_mark12);
d1mode_priority_a_cnt = dfixed_trunc(priority_mark02);
d2mode_priority_a_cnt = dfixed_trunc(priority_mark12);
if (rdev->disp_priority == 2) {
d1mode_priority_a_cnt |= S_006548_D1MODE_PRIORITY_A_ALWAYS_ON(1);
d2mode_priority_a_cnt |= S_006D48_D2MODE_PRIORITY_A_ALWAYS_ON(1);
@ -500,32 +500,32 @@ void rs690_bandwidth_update(struct radeon_device *rdev)
WREG32(R_006D48_D2MODE_PRIORITY_A_CNT, d2mode_priority_a_cnt);
WREG32(R_006D4C_D2MODE_PRIORITY_B_CNT, d2mode_priority_a_cnt);
} else if (mode0) {
if (rfixed_trunc(wm0.dbpp) > 64)
a.full = rfixed_mul(wm0.dbpp, wm0.num_line_pair);
if (dfixed_trunc(wm0.dbpp) > 64)
a.full = dfixed_mul(wm0.dbpp, wm0.num_line_pair);
else
a.full = wm0.num_line_pair.full;
fill_rate.full = rfixed_div(wm0.sclk, a);
fill_rate.full = dfixed_div(wm0.sclk, a);
if (wm0.consumption_rate.full > fill_rate.full) {
b.full = wm0.consumption_rate.full - fill_rate.full;
b.full = rfixed_mul(b, wm0.active_time);
a.full = rfixed_mul(wm0.worst_case_latency,
b.full = dfixed_mul(b, wm0.active_time);
a.full = dfixed_mul(wm0.worst_case_latency,
wm0.consumption_rate);
a.full = a.full + b.full;
b.full = rfixed_const(16 * 1000);
priority_mark02.full = rfixed_div(a, b);
b.full = dfixed_const(16 * 1000);
priority_mark02.full = dfixed_div(a, b);
} else {
a.full = rfixed_mul(wm0.worst_case_latency,
a.full = dfixed_mul(wm0.worst_case_latency,
wm0.consumption_rate);
b.full = rfixed_const(16 * 1000);
priority_mark02.full = rfixed_div(a, b);
b.full = dfixed_const(16 * 1000);
priority_mark02.full = dfixed_div(a, b);
}
if (wm0.priority_mark.full > priority_mark02.full)
priority_mark02.full = wm0.priority_mark.full;
if (rfixed_trunc(priority_mark02) < 0)
if (dfixed_trunc(priority_mark02) < 0)
priority_mark02.full = 0;
if (wm0.priority_mark_max.full > priority_mark02.full)
priority_mark02.full = wm0.priority_mark_max.full;
d1mode_priority_a_cnt = rfixed_trunc(priority_mark02);
d1mode_priority_a_cnt = dfixed_trunc(priority_mark02);
if (rdev->disp_priority == 2)
d1mode_priority_a_cnt |= S_006548_D1MODE_PRIORITY_A_ALWAYS_ON(1);
WREG32(R_006548_D1MODE_PRIORITY_A_CNT, d1mode_priority_a_cnt);
@ -535,32 +535,32 @@ void rs690_bandwidth_update(struct radeon_device *rdev)
WREG32(R_006D4C_D2MODE_PRIORITY_B_CNT,
S_006D4C_D2MODE_PRIORITY_B_OFF(1));
} else {
if (rfixed_trunc(wm1.dbpp) > 64)
a.full = rfixed_mul(wm1.dbpp, wm1.num_line_pair);
if (dfixed_trunc(wm1.dbpp) > 64)
a.full = dfixed_mul(wm1.dbpp, wm1.num_line_pair);
else
a.full = wm1.num_line_pair.full;
fill_rate.full = rfixed_div(wm1.sclk, a);
fill_rate.full = dfixed_div(wm1.sclk, a);
if (wm1.consumption_rate.full > fill_rate.full) {
b.full = wm1.consumption_rate.full - fill_rate.full;
b.full = rfixed_mul(b, wm1.active_time);
a.full = rfixed_mul(wm1.worst_case_latency,
b.full = dfixed_mul(b, wm1.active_time);
a.full = dfixed_mul(wm1.worst_case_latency,
wm1.consumption_rate);
a.full = a.full + b.full;
b.full = rfixed_const(16 * 1000);
priority_mark12.full = rfixed_div(a, b);
b.full = dfixed_const(16 * 1000);
priority_mark12.full = dfixed_div(a, b);
} else {
a.full = rfixed_mul(wm1.worst_case_latency,
a.full = dfixed_mul(wm1.worst_case_latency,
wm1.consumption_rate);
b.full = rfixed_const(16 * 1000);
priority_mark12.full = rfixed_div(a, b);
b.full = dfixed_const(16 * 1000);
priority_mark12.full = dfixed_div(a, b);
}
if (wm1.priority_mark.full > priority_mark12.full)
priority_mark12.full = wm1.priority_mark.full;
if (rfixed_trunc(priority_mark12) < 0)
if (dfixed_trunc(priority_mark12) < 0)
priority_mark12.full = 0;
if (wm1.priority_mark_max.full > priority_mark12.full)
priority_mark12.full = wm1.priority_mark_max.full;
d2mode_priority_a_cnt = rfixed_trunc(priority_mark12);
d2mode_priority_a_cnt = dfixed_trunc(priority_mark12);
if (rdev->disp_priority == 2)
d2mode_priority_a_cnt |= S_006D48_D2MODE_PRIORITY_A_ALWAYS_ON(1);
WREG32(R_006548_D1MODE_PRIORITY_A_CNT,

View File

@ -795,20 +795,20 @@ void rv515_crtc_bandwidth_compute(struct radeon_device *rdev,
return;
}
if (crtc->vsc.full > rfixed_const(2))
wm->num_line_pair.full = rfixed_const(2);
if (crtc->vsc.full > dfixed_const(2))
wm->num_line_pair.full = dfixed_const(2);
else
wm->num_line_pair.full = rfixed_const(1);
wm->num_line_pair.full = dfixed_const(1);
b.full = rfixed_const(mode->crtc_hdisplay);
c.full = rfixed_const(256);
a.full = rfixed_div(b, c);
request_fifo_depth.full = rfixed_mul(a, wm->num_line_pair);
request_fifo_depth.full = rfixed_ceil(request_fifo_depth);
if (a.full < rfixed_const(4)) {
b.full = dfixed_const(mode->crtc_hdisplay);
c.full = dfixed_const(256);
a.full = dfixed_div(b, c);
request_fifo_depth.full = dfixed_mul(a, wm->num_line_pair);
request_fifo_depth.full = dfixed_ceil(request_fifo_depth);
if (a.full < dfixed_const(4)) {
wm->lb_request_fifo_depth = 4;
} else {
wm->lb_request_fifo_depth = rfixed_trunc(request_fifo_depth);
wm->lb_request_fifo_depth = dfixed_trunc(request_fifo_depth);
}
/* Determine consumption rate
@ -817,23 +817,23 @@ void rv515_crtc_bandwidth_compute(struct radeon_device *rdev,
* vsc = vertical scaling ratio, defined as source/destination
* hsc = horizontal scaling ration, defined as source/destination
*/
a.full = rfixed_const(mode->clock);
b.full = rfixed_const(1000);
a.full = rfixed_div(a, b);
pclk.full = rfixed_div(b, a);
a.full = dfixed_const(mode->clock);
b.full = dfixed_const(1000);
a.full = dfixed_div(a, b);
pclk.full = dfixed_div(b, a);
if (crtc->rmx_type != RMX_OFF) {
b.full = rfixed_const(2);
b.full = dfixed_const(2);
if (crtc->vsc.full > b.full)
b.full = crtc->vsc.full;
b.full = rfixed_mul(b, crtc->hsc);
c.full = rfixed_const(2);
b.full = rfixed_div(b, c);
consumption_time.full = rfixed_div(pclk, b);
b.full = dfixed_mul(b, crtc->hsc);
c.full = dfixed_const(2);
b.full = dfixed_div(b, c);
consumption_time.full = dfixed_div(pclk, b);
} else {
consumption_time.full = pclk.full;
}
a.full = rfixed_const(1);
wm->consumption_rate.full = rfixed_div(a, consumption_time);
a.full = dfixed_const(1);
wm->consumption_rate.full = dfixed_div(a, consumption_time);
/* Determine line time
@ -841,27 +841,27 @@ void rv515_crtc_bandwidth_compute(struct radeon_device *rdev,
* LineTime = total number of horizontal pixels
* pclk = pixel clock period(ns)
*/
a.full = rfixed_const(crtc->base.mode.crtc_htotal);
line_time.full = rfixed_mul(a, pclk);
a.full = dfixed_const(crtc->base.mode.crtc_htotal);
line_time.full = dfixed_mul(a, pclk);
/* Determine active time
* ActiveTime = time of active region of display within one line,
* hactive = total number of horizontal active pixels
* htotal = total number of horizontal pixels
*/
a.full = rfixed_const(crtc->base.mode.crtc_htotal);
b.full = rfixed_const(crtc->base.mode.crtc_hdisplay);
wm->active_time.full = rfixed_mul(line_time, b);
wm->active_time.full = rfixed_div(wm->active_time, a);
a.full = dfixed_const(crtc->base.mode.crtc_htotal);
b.full = dfixed_const(crtc->base.mode.crtc_hdisplay);
wm->active_time.full = dfixed_mul(line_time, b);
wm->active_time.full = dfixed_div(wm->active_time, a);
/* Determine chunk time
* ChunkTime = the time it takes the DCP to send one chunk of data
* to the LB which consists of pipeline delay and inter chunk gap
* sclk = system clock(Mhz)
*/
a.full = rfixed_const(600 * 1000);
chunk_time.full = rfixed_div(a, rdev->pm.sclk);
read_delay_latency.full = rfixed_const(1000);
a.full = dfixed_const(600 * 1000);
chunk_time.full = dfixed_div(a, rdev->pm.sclk);
read_delay_latency.full = dfixed_const(1000);
/* Determine the worst case latency
* NumLinePair = Number of line pairs to request(1=2 lines, 2=4 lines)
@ -871,9 +871,9 @@ void rv515_crtc_bandwidth_compute(struct radeon_device *rdev,
* ChunkTime = time it takes the DCP to send one chunk of data to the LB
* which consists of pipeline delay and inter chunk gap
*/
if (rfixed_trunc(wm->num_line_pair) > 1) {
a.full = rfixed_const(3);
wm->worst_case_latency.full = rfixed_mul(a, chunk_time);
if (dfixed_trunc(wm->num_line_pair) > 1) {
a.full = dfixed_const(3);
wm->worst_case_latency.full = dfixed_mul(a, chunk_time);
wm->worst_case_latency.full += read_delay_latency.full;
} else {
wm->worst_case_latency.full = chunk_time.full + read_delay_latency.full;
@ -889,34 +889,34 @@ void rv515_crtc_bandwidth_compute(struct radeon_device *rdev,
* of data to the LB which consists of
* pipeline delay and inter chunk gap
*/
if ((2+wm->lb_request_fifo_depth) >= rfixed_trunc(request_fifo_depth)) {
if ((2+wm->lb_request_fifo_depth) >= dfixed_trunc(request_fifo_depth)) {
tolerable_latency.full = line_time.full;
} else {
tolerable_latency.full = rfixed_const(wm->lb_request_fifo_depth - 2);
tolerable_latency.full = dfixed_const(wm->lb_request_fifo_depth - 2);
tolerable_latency.full = request_fifo_depth.full - tolerable_latency.full;
tolerable_latency.full = rfixed_mul(tolerable_latency, chunk_time);
tolerable_latency.full = dfixed_mul(tolerable_latency, chunk_time);
tolerable_latency.full = line_time.full - tolerable_latency.full;
}
/* We assume worst case 32bits (4 bytes) */
wm->dbpp.full = rfixed_const(2 * 16);
wm->dbpp.full = dfixed_const(2 * 16);
/* Determine the maximum priority mark
* width = viewport width in pixels
*/
a.full = rfixed_const(16);
wm->priority_mark_max.full = rfixed_const(crtc->base.mode.crtc_hdisplay);
wm->priority_mark_max.full = rfixed_div(wm->priority_mark_max, a);
wm->priority_mark_max.full = rfixed_ceil(wm->priority_mark_max);
a.full = dfixed_const(16);
wm->priority_mark_max.full = dfixed_const(crtc->base.mode.crtc_hdisplay);
wm->priority_mark_max.full = dfixed_div(wm->priority_mark_max, a);
wm->priority_mark_max.full = dfixed_ceil(wm->priority_mark_max);
/* Determine estimated width */
estimated_width.full = tolerable_latency.full - wm->worst_case_latency.full;
estimated_width.full = rfixed_div(estimated_width, consumption_time);
if (rfixed_trunc(estimated_width) > crtc->base.mode.crtc_hdisplay) {
estimated_width.full = dfixed_div(estimated_width, consumption_time);
if (dfixed_trunc(estimated_width) > crtc->base.mode.crtc_hdisplay) {
wm->priority_mark.full = wm->priority_mark_max.full;
} else {
a.full = rfixed_const(16);
wm->priority_mark.full = rfixed_div(estimated_width, a);
wm->priority_mark.full = rfixed_ceil(wm->priority_mark);
a.full = dfixed_const(16);
wm->priority_mark.full = dfixed_div(estimated_width, a);
wm->priority_mark.full = dfixed_ceil(wm->priority_mark);
wm->priority_mark.full = wm->priority_mark_max.full - wm->priority_mark.full;
}
}
@ -945,58 +945,58 @@ void rv515_bandwidth_avivo_update(struct radeon_device *rdev)
WREG32(LB_MAX_REQ_OUTSTANDING, tmp);
if (mode0 && mode1) {
if (rfixed_trunc(wm0.dbpp) > 64)
a.full = rfixed_div(wm0.dbpp, wm0.num_line_pair);
if (dfixed_trunc(wm0.dbpp) > 64)
a.full = dfixed_div(wm0.dbpp, wm0.num_line_pair);
else
a.full = wm0.num_line_pair.full;
if (rfixed_trunc(wm1.dbpp) > 64)
b.full = rfixed_div(wm1.dbpp, wm1.num_line_pair);
if (dfixed_trunc(wm1.dbpp) > 64)
b.full = dfixed_div(wm1.dbpp, wm1.num_line_pair);
else
b.full = wm1.num_line_pair.full;
a.full += b.full;
fill_rate.full = rfixed_div(wm0.sclk, a);
fill_rate.full = dfixed_div(wm0.sclk, a);
if (wm0.consumption_rate.full > fill_rate.full) {
b.full = wm0.consumption_rate.full - fill_rate.full;
b.full = rfixed_mul(b, wm0.active_time);
a.full = rfixed_const(16);
b.full = rfixed_div(b, a);
a.full = rfixed_mul(wm0.worst_case_latency,
b.full = dfixed_mul(b, wm0.active_time);
a.full = dfixed_const(16);
b.full = dfixed_div(b, a);
a.full = dfixed_mul(wm0.worst_case_latency,
wm0.consumption_rate);
priority_mark02.full = a.full + b.full;
} else {
a.full = rfixed_mul(wm0.worst_case_latency,
a.full = dfixed_mul(wm0.worst_case_latency,
wm0.consumption_rate);
b.full = rfixed_const(16 * 1000);
priority_mark02.full = rfixed_div(a, b);
b.full = dfixed_const(16 * 1000);
priority_mark02.full = dfixed_div(a, b);
}
if (wm1.consumption_rate.full > fill_rate.full) {
b.full = wm1.consumption_rate.full - fill_rate.full;
b.full = rfixed_mul(b, wm1.active_time);
a.full = rfixed_const(16);
b.full = rfixed_div(b, a);
a.full = rfixed_mul(wm1.worst_case_latency,
b.full = dfixed_mul(b, wm1.active_time);
a.full = dfixed_const(16);
b.full = dfixed_div(b, a);
a.full = dfixed_mul(wm1.worst_case_latency,
wm1.consumption_rate);
priority_mark12.full = a.full + b.full;
} else {
a.full = rfixed_mul(wm1.worst_case_latency,
a.full = dfixed_mul(wm1.worst_case_latency,
wm1.consumption_rate);
b.full = rfixed_const(16 * 1000);
priority_mark12.full = rfixed_div(a, b);
b.full = dfixed_const(16 * 1000);
priority_mark12.full = dfixed_div(a, b);
}
if (wm0.priority_mark.full > priority_mark02.full)
priority_mark02.full = wm0.priority_mark.full;
if (rfixed_trunc(priority_mark02) < 0)
if (dfixed_trunc(priority_mark02) < 0)
priority_mark02.full = 0;
if (wm0.priority_mark_max.full > priority_mark02.full)
priority_mark02.full = wm0.priority_mark_max.full;
if (wm1.priority_mark.full > priority_mark12.full)
priority_mark12.full = wm1.priority_mark.full;
if (rfixed_trunc(priority_mark12) < 0)
if (dfixed_trunc(priority_mark12) < 0)
priority_mark12.full = 0;
if (wm1.priority_mark_max.full > priority_mark12.full)
priority_mark12.full = wm1.priority_mark_max.full;
d1mode_priority_a_cnt = rfixed_trunc(priority_mark02);
d2mode_priority_a_cnt = rfixed_trunc(priority_mark12);
d1mode_priority_a_cnt = dfixed_trunc(priority_mark02);
d2mode_priority_a_cnt = dfixed_trunc(priority_mark12);
if (rdev->disp_priority == 2) {
d1mode_priority_a_cnt |= MODE_PRIORITY_ALWAYS_ON;
d2mode_priority_a_cnt |= MODE_PRIORITY_ALWAYS_ON;
@ -1006,32 +1006,32 @@ void rv515_bandwidth_avivo_update(struct radeon_device *rdev)
WREG32(D2MODE_PRIORITY_A_CNT, d2mode_priority_a_cnt);
WREG32(D2MODE_PRIORITY_B_CNT, d2mode_priority_a_cnt);
} else if (mode0) {
if (rfixed_trunc(wm0.dbpp) > 64)
a.full = rfixed_div(wm0.dbpp, wm0.num_line_pair);
if (dfixed_trunc(wm0.dbpp) > 64)
a.full = dfixed_div(wm0.dbpp, wm0.num_line_pair);
else
a.full = wm0.num_line_pair.full;
fill_rate.full = rfixed_div(wm0.sclk, a);
fill_rate.full = dfixed_div(wm0.sclk, a);
if (wm0.consumption_rate.full > fill_rate.full) {
b.full = wm0.consumption_rate.full - fill_rate.full;
b.full = rfixed_mul(b, wm0.active_time);
a.full = rfixed_const(16);
b.full = rfixed_div(b, a);
a.full = rfixed_mul(wm0.worst_case_latency,
b.full = dfixed_mul(b, wm0.active_time);
a.full = dfixed_const(16);
b.full = dfixed_div(b, a);
a.full = dfixed_mul(wm0.worst_case_latency,
wm0.consumption_rate);
priority_mark02.full = a.full + b.full;
} else {
a.full = rfixed_mul(wm0.worst_case_latency,
a.full = dfixed_mul(wm0.worst_case_latency,
wm0.consumption_rate);
b.full = rfixed_const(16);
priority_mark02.full = rfixed_div(a, b);
b.full = dfixed_const(16);
priority_mark02.full = dfixed_div(a, b);
}
if (wm0.priority_mark.full > priority_mark02.full)
priority_mark02.full = wm0.priority_mark.full;
if (rfixed_trunc(priority_mark02) < 0)
if (dfixed_trunc(priority_mark02) < 0)
priority_mark02.full = 0;
if (wm0.priority_mark_max.full > priority_mark02.full)
priority_mark02.full = wm0.priority_mark_max.full;
d1mode_priority_a_cnt = rfixed_trunc(priority_mark02);
d1mode_priority_a_cnt = dfixed_trunc(priority_mark02);
if (rdev->disp_priority == 2)
d1mode_priority_a_cnt |= MODE_PRIORITY_ALWAYS_ON;
WREG32(D1MODE_PRIORITY_A_CNT, d1mode_priority_a_cnt);
@ -1039,32 +1039,32 @@ void rv515_bandwidth_avivo_update(struct radeon_device *rdev)
WREG32(D2MODE_PRIORITY_A_CNT, MODE_PRIORITY_OFF);
WREG32(D2MODE_PRIORITY_B_CNT, MODE_PRIORITY_OFF);
} else {
if (rfixed_trunc(wm1.dbpp) > 64)
a.full = rfixed_div(wm1.dbpp, wm1.num_line_pair);
if (dfixed_trunc(wm1.dbpp) > 64)
a.full = dfixed_div(wm1.dbpp, wm1.num_line_pair);
else
a.full = wm1.num_line_pair.full;
fill_rate.full = rfixed_div(wm1.sclk, a);
fill_rate.full = dfixed_div(wm1.sclk, a);
if (wm1.consumption_rate.full > fill_rate.full) {
b.full = wm1.consumption_rate.full - fill_rate.full;
b.full = rfixed_mul(b, wm1.active_time);
a.full = rfixed_const(16);
b.full = rfixed_div(b, a);
a.full = rfixed_mul(wm1.worst_case_latency,
b.full = dfixed_mul(b, wm1.active_time);
a.full = dfixed_const(16);
b.full = dfixed_div(b, a);
a.full = dfixed_mul(wm1.worst_case_latency,
wm1.consumption_rate);
priority_mark12.full = a.full + b.full;
} else {
a.full = rfixed_mul(wm1.worst_case_latency,
a.full = dfixed_mul(wm1.worst_case_latency,
wm1.consumption_rate);
b.full = rfixed_const(16 * 1000);
priority_mark12.full = rfixed_div(a, b);
b.full = dfixed_const(16 * 1000);
priority_mark12.full = dfixed_div(a, b);
}
if (wm1.priority_mark.full > priority_mark12.full)
priority_mark12.full = wm1.priority_mark.full;
if (rfixed_trunc(priority_mark12) < 0)
if (dfixed_trunc(priority_mark12) < 0)
priority_mark12.full = 0;
if (wm1.priority_mark_max.full > priority_mark12.full)
priority_mark12.full = wm1.priority_mark_max.full;
d2mode_priority_a_cnt = rfixed_trunc(priority_mark12);
d2mode_priority_a_cnt = dfixed_trunc(priority_mark12);
if (rdev->disp_priority == 2)
d2mode_priority_a_cnt |= MODE_PRIORITY_ALWAYS_ON;
WREG32(D1MODE_PRIORITY_A_CNT, MODE_PRIORITY_OFF);

View File

@ -21,41 +21,41 @@
*
* Authors: Dave Airlie
*/
#ifndef RADEON_FIXED_H
#define RADEON_FIXED_H
#ifndef DRM_FIXED_H
#define DRM_FIXED_H
typedef union rfixed {
typedef union dfixed {
u32 full;
} fixed20_12;
#define rfixed_const(A) (u32)(((A) << 12))/* + ((B + 0.000122)*4096)) */
#define rfixed_const_half(A) (u32)(((A) << 12) + 2048)
#define rfixed_const_666(A) (u32)(((A) << 12) + 2731)
#define rfixed_const_8(A) (u32)(((A) << 12) + 3277)
#define rfixed_mul(A, B) ((u64)((u64)(A).full * (B).full + 2048) >> 12)
#define fixed_init(A) { .full = rfixed_const((A)) }
#define fixed_init_half(A) { .full = rfixed_const_half((A)) }
#define rfixed_trunc(A) ((A).full >> 12)
#define dfixed_const(A) (u32)(((A) << 12))/* + ((B + 0.000122)*4096)) */
#define dfixed_const_half(A) (u32)(((A) << 12) + 2048)
#define dfixed_const_666(A) (u32)(((A) << 12) + 2731)
#define dfixed_const_8(A) (u32)(((A) << 12) + 3277)
#define dfixed_mul(A, B) ((u64)((u64)(A).full * (B).full + 2048) >> 12)
#define dfixed_init(A) { .full = dfixed_const((A)) }
#define dfixed_init_half(A) { .full = dfixed_const_half((A)) }
#define dfixed_trunc(A) ((A).full >> 12)
static inline u32 rfixed_floor(fixed20_12 A)
static inline u32 dfixed_floor(fixed20_12 A)
{
u32 non_frac = rfixed_trunc(A);
u32 non_frac = dfixed_trunc(A);
return rfixed_const(non_frac);
return dfixed_const(non_frac);
}
static inline u32 rfixed_ceil(fixed20_12 A)
static inline u32 dfixed_ceil(fixed20_12 A)
{
u32 non_frac = rfixed_trunc(A);
u32 non_frac = dfixed_trunc(A);
if (A.full > rfixed_const(non_frac))
return rfixed_const(non_frac + 1);
if (A.full > dfixed_const(non_frac))
return dfixed_const(non_frac + 1);
else
return rfixed_const(non_frac);
return dfixed_const(non_frac);
}
static inline u32 rfixed_div(fixed20_12 A, fixed20_12 B)
static inline u32 dfixed_div(fixed20_12 A, fixed20_12 B)
{
u64 tmp = ((u64)A.full << 13);