| Message ID | 772b3869cc992d894d8c1b4296f21100bcbed30d.1545075103.git.mirq-linux@rere.qmqm.pl |
|---|---|
| State | New |
| Headers | show |
| Series | clk: at91: optimize clk_round_rate() for AUDIO_PLL | expand |
Quoting Michał Mirosław (2018-12-17 11:37:05) > This makes for 3 clk_round_rate() calls instead of 64 of them on > SAMA5D2-based board when searching for 12.288MHz clock. Ok, but is that better? Worse? Can you more clearly describe what you're doing here? And it's not breaking some other corner case? > > Signed-off-by: Micha\u0142 Miros\u0142aw <mirq-linux@rere.qmqm.pl> This got all broken. Something on your end? > --- > drivers/clk/at91/clk-audio-pll.c | 8 +++++++- > 1 file changed, 7 insertions(+), 1 deletion(-) > > diff --git a/drivers/clk/at91/clk-audio-pll.c b/drivers/clk/at91/clk-audio-pll.c > index 36d77146a3bd..3a358282784e 100644 > --- a/drivers/clk/at91/clk-audio-pll.c > +++ b/drivers/clk/at91/clk-audio-pll.c > @@ -340,7 +340,11 @@ static long clk_audio_pll_pmc_round_rate(struct clk_hw *hw, unsigned long rate, > pr_debug("A PLL/PMC: %s, rate = %lu (parent_rate = %lu)\n", __func__, > rate, *parent_rate); > > - for (div = 1; div <= AUDIO_PLL_QDPMC_MAX; div++) { > + if (!rate) > + return 0; > + > + best_parent_rate = clk_round_rate(pclk->clk, 1); > + for (div = max(best_parent_rate / rate, 1ul); div <= AUDIO_PLL_QDPMC_MAX; div++) { Nitpick: Style is usually 1UL instead of lowercase 1ul. > best_parent_rate = clk_round_rate(pclk->clk, rate * div); > tmp_rate = best_parent_rate / div; > tmp_diff = abs(rate - tmp_rate); > @@ -350,6 +354,8 @@ static long clk_audio_pll_pmc_round_rate(struct clk_hw *hw, unsigned long rate, > best_rate = tmp_rate; > best_diff = tmp_diff; > tmp_qd = div; > + if (!best_diff) > + break; /* got exact match */ Seems sensible to break immediately if exact match is found. `
On Mon, Dec 17, 2018 at 03:22:50PM -0800, Stephen Boyd wrote: > Quoting Michał Mirosław (2018-12-17 11:37:05) > > This makes for 3 clk_round_rate() calls instead of 64 of them on > > SAMA5D2-based board when searching for 12.288MHz clock. > Ok, but is that better? Worse? Can you more clearly describe what you're > doing here? And it's not breaking some other corner case? > > > > Signed-off-by: Micha\u0142 Miros\u0142aw <mirq-linux@rere.qmqm.pl> > This got all broken. Something on your end? [...] linux-clk archives got it right: https://lore.kernel.org/linux-clk/772b3869cc992d894d8c1b4296f21100bcbed30d.1545075103.git.mirq-linux@rere.qmqm.pl/T/#u Best Regards, Michał Mirosław
Hello, On Mon, Dec 17, 2018 at 03:22:50PM -0800, Stephen Boyd wrote: > Quoting Michał Mirosław (2018-12-17 11:37:05) > > This makes for 3 clk_round_rate() calls instead of 64 of them on > > SAMA5D2-based board when searching for 12.288MHz clock. > > Ok, but is that better? Worse? Can you more clearly describe what you're > doing here? And it's not breaking some other corner case? I think there isn't even a policy how clk_round_rate() actually behaves, is there? If you could rely on it to (say) always round down cases like these could maybe be handled easier? (But I didn't really look at the patch, so this might not be one of these cases.) Best regards Uwe
diff --git a/drivers/clk/at91/clk-audio-pll.c b/drivers/clk/at91/clk-audio-pll.c index 36d77146a3bd..3a358282784e 100644 --- a/drivers/clk/at91/clk-audio-pll.c +++ b/drivers/clk/at91/clk-audio-pll.c @@ -340,7 +340,11 @@ static long clk_audio_pll_pmc_round_rate(struct clk_hw *hw, unsigned long rate, pr_debug("A PLL/PMC: %s, rate = %lu (parent_rate = %lu)\n", __func__, rate, *parent_rate); - for (div = 1; div <= AUDIO_PLL_QDPMC_MAX; div++) { + if (!rate) + return 0; + + best_parent_rate = clk_round_rate(pclk->clk, 1); + for (div = max(best_parent_rate / rate, 1ul); div <= AUDIO_PLL_QDPMC_MAX; div++) { best_parent_rate = clk_round_rate(pclk->clk, rate * div); tmp_rate = best_parent_rate / div; tmp_diff = abs(rate - tmp_rate); @@ -350,6 +354,8 @@ static long clk_audio_pll_pmc_round_rate(struct clk_hw *hw, unsigned long rate, best_rate = tmp_rate; best_diff = tmp_diff; tmp_qd = div; + if (!best_diff) + break; /* got exact match */ } }
This makes for 3 clk_round_rate() calls instead of 64 of them on SAMA5D2-based board when searching for 12.288MHz clock. Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl> --- drivers/clk/at91/clk-audio-pll.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-)