[Piglit] [PATCH 1/5] Basic tests of fast color clear functionality on i965/gen7+.
Ian Romanick
idr at freedesktop.org
Fri May 24 13:01:36 PDT 2013
Patches 1-3 and 5 are
Reviewed-by: Ian Romanick <ian.d.romanick at intel.com>
On 05/22/2013 10:20 AM, Paul Berry wrote:
> Fast color clears are a hardware feature in i965/gen7+ which allows
> the memory writes associated with clearing a color buffer to be
> deferred until the buffer is later written to or read from; in most
> cases this allows the clear writes to be write-combined with later
> rendering operations.
>
> For fast color clears to work correctly, the driver must maintain a
> state machine for each buffer to indicate whether there are deferred
> clears pending, and if so what the clear value was. This patch
> introduces simple shader_runner tests to validate corner cases of that
> state machine.
> ---
> tests/all.tests | 5 +
> tests/fast_color_clear/all-colors.shader_test | 140 +++++++++++++++++++++
> .../fast-slow-clear-interaction.shader_test | 39 ++++++
> .../non-redundant-clear.shader_test | 46 +++++++
> tests/fast_color_clear/redundant-clear.shader_test | 50 ++++++++
> 5 files changed, 280 insertions(+)
> create mode 100644 tests/fast_color_clear/all-colors.shader_test
> create mode 100644 tests/fast_color_clear/fast-slow-clear-interaction.shader_test
> create mode 100644 tests/fast_color_clear/non-redundant-clear.shader_test
> create mode 100644 tests/fast_color_clear/redundant-clear.shader_test
>
> diff --git a/tests/all.tests b/tests/all.tests
> index 0f80e0b..267e769 100644
> --- a/tests/all.tests
> +++ b/tests/all.tests
> @@ -2326,6 +2326,10 @@ add_plain_test(hiz, 'hiz-stencil-test-fbo-d24s8')
> add_plain_test(hiz, 'hiz-stencil-test-window-depth0')
> add_plain_test(hiz, 'hiz-stencil-test-window-depth1')
>
> +fast_color_clear = Group()
> +add_shader_test_dir(fast_color_clear, testsDir + '/fast_color_clear',
> + recursive=True)
> +
> asmparsertest = Group()
> def add_asmparsertest(group, shader):
> test = PlainExecTest(['asmparsertest', '-auto', group, testsDir + '/asmparsertest/shaders/' + group + '/' + shader])
> @@ -2852,6 +2856,7 @@ add_shader_test_dir(spec, os.path.join(generatedTestDir, 'spec'),
> import_glsl_parser_tests(profile.tests, generatedTestDir, ['spec'])
>
> profile.tests['hiz'] = hiz
> +profile.tests['fast_color_clear'] = fast_color_clear
> profile.tests['glean'] = glean
> profile.tests['glslparsertest'] = glslparsertest
> profile.tests['asmparsertest'] = asmparsertest
> diff --git a/tests/fast_color_clear/all-colors.shader_test b/tests/fast_color_clear/all-colors.shader_test
> new file mode 100644
> index 0000000..f4b8e87
> --- /dev/null
> +++ b/tests/fast_color_clear/all-colors.shader_test
> @@ -0,0 +1,140 @@
> +# Test the "fast clear" functionality of i965/gen7+ for all supported
> +# clear colors.
> +#
> +# Fast clear functionality on i965/gen7+ operates by deferring the
> +# memory writes that would normally be done during a buffer clear so
> +# that they happen either during rendering or at the time buffer data
> +# is read (e.g. by glReadPixels). It supports clearing to any color
> +# where each component is either 0.0 or 1.0.
> +#
> +# This test validates each supported clear color by (a) setting the
> +# clear color, (b) doing a clear, (c) making a draw call that touches
> +# a small number of pixels and outputs the same color, and then (d)
> +# reading the entire buffer to verify that it contains the desired
> +# color.
> +#
> +# The purpose of step (c) is to ensure that some writes are deferred
> +# to rendering time and others are deferred until glReadPixels time,
> +# so that we verify that correct data is written in both
> +# circumstances.
> +
> +[require]
> +GLSL >= 1.10
> +
> +[vertex shader]
> +attribute vec4 pos;
> +
> +void main()
> +{
> + gl_Position = pos;
> +}
> +
> +[fragment shader]
> +uniform vec4 color;
> +
> +void main()
> +{
> + gl_FragColor = color;
> +}
> +
> +[vertex data]
> +pos/float/4
> +0.0 0.0 0.0 1.0
> +1.0 1.0 0.0 1.0
> +
> +[test]
> +clear color 0.0 0.0 0.0 0.0
> +clear
> +uniform vec4 color 0.0 0.0 0.0 0.0
> +draw arrays GL_LINES 0 2
> +probe all rgba 0.0 0.0 0.0 0.0
> +
> +clear color 0.0 0.0 0.0 1.0
> +clear
> +uniform vec4 color 0.0 0.0 0.0 1.0
> +draw arrays GL_LINES 0 2
> +probe all rgba 0.0 0.0 0.0 1.0
> +
> +clear color 0.0 0.0 1.0 0.0
> +clear
> +uniform vec4 color 0.0 0.0 1.0 0.0
> +draw arrays GL_LINES 0 2
> +probe all rgba 0.0 0.0 1.0 0.0
> +
> +clear color 0.0 0.0 1.0 1.0
> +clear
> +uniform vec4 color 0.0 0.0 1.0 1.0
> +draw arrays GL_LINES 0 2
> +probe all rgba 0.0 0.0 1.0 1.0
> +
> +clear color 0.0 1.0 0.0 0.0
> +clear
> +uniform vec4 color 0.0 1.0 0.0 0.0
> +draw arrays GL_LINES 0 2
> +probe all rgba 0.0 1.0 0.0 0.0
> +
> +clear color 0.0 1.0 0.0 1.0
> +clear
> +uniform vec4 color 0.0 1.0 0.0 1.0
> +draw arrays GL_LINES 0 2
> +probe all rgba 0.0 1.0 0.0 1.0
> +
> +clear color 0.0 1.0 1.0 0.0
> +clear
> +uniform vec4 color 0.0 1.0 1.0 0.0
> +draw arrays GL_LINES 0 2
> +probe all rgba 0.0 1.0 1.0 0.0
> +
> +clear color 0.0 1.0 1.0 1.0
> +clear
> +uniform vec4 color 0.0 1.0 1.0 1.0
> +draw arrays GL_LINES 0 2
> +probe all rgba 0.0 1.0 1.0 1.0
> +
> +clear color 1.0 0.0 0.0 0.0
> +clear
> +uniform vec4 color 1.0 0.0 0.0 0.0
> +draw arrays GL_LINES 0 2
> +probe all rgba 1.0 0.0 0.0 0.0
> +
> +clear color 1.0 0.0 0.0 1.0
> +clear
> +uniform vec4 color 1.0 0.0 0.0 1.0
> +draw arrays GL_LINES 0 2
> +probe all rgba 1.0 0.0 0.0 1.0
> +
> +clear color 1.0 0.0 1.0 0.0
> +clear
> +uniform vec4 color 1.0 0.0 1.0 0.0
> +draw arrays GL_LINES 0 2
> +probe all rgba 1.0 0.0 1.0 0.0
> +
> +clear color 1.0 0.0 1.0 1.0
> +clear
> +uniform vec4 color 1.0 0.0 1.0 1.0
> +draw arrays GL_LINES 0 2
> +probe all rgba 1.0 0.0 1.0 1.0
> +
> +clear color 1.0 1.0 0.0 0.0
> +clear
> +uniform vec4 color 1.0 1.0 0.0 0.0
> +draw arrays GL_LINES 0 2
> +probe all rgba 1.0 1.0 0.0 0.0
> +
> +clear color 1.0 1.0 0.0 1.0
> +clear
> +uniform vec4 color 1.0 1.0 0.0 1.0
> +draw arrays GL_LINES 0 2
> +probe all rgba 1.0 1.0 0.0 1.0
> +
> +clear color 1.0 1.0 1.0 0.0
> +clear
> +uniform vec4 color 1.0 1.0 1.0 0.0
> +draw arrays GL_LINES 0 2
> +probe all rgba 1.0 1.0 1.0 0.0
> +
> +clear color 1.0 1.0 1.0 1.0
> +clear
> +uniform vec4 color 1.0 1.0 1.0 1.0
> +draw arrays GL_LINES 0 2
> +probe all rgba 1.0 1.0 1.0 1.0
> diff --git a/tests/fast_color_clear/fast-slow-clear-interaction.shader_test b/tests/fast_color_clear/fast-slow-clear-interaction.shader_test
> new file mode 100644
> index 0000000..e3a724c
> --- /dev/null
> +++ b/tests/fast_color_clear/fast-slow-clear-interaction.shader_test
> @@ -0,0 +1,39 @@
> +# Test the "fast clear" functionality of i965/gen7+ when two clears
> +# are done to the same buffer, but with drawing in between (so the
> +# second clear is not redundant).
> +#
> +# Fast clear functionality on i965/gen7+ operates by deferring the
> +# memory writes that would normally be done during a buffer clear so
> +# that they happen either during rendering or at the time buffer data
> +# is read (e.g. by glReadPixels). If two fast clears appear in a row,
> +# the second clear can be elided. But if other rendering (in this
> +# case a slow clear) occurrs between the two fast clears, the second
> +# fast clear is necessary, since some of the first fast clear's writes
> +# may already have been performed.
> +#
> +# In this test, we accomplish the slow clear by clearing to a color
> +# which is not supported by fast clear (0.5, 0.0, 0.0, 1.0).
> +
> +[require]
> +GLSL >= 1.10
> +
> +[vertex shader]
> +void main()
> +{
> + gl_Position = gl_Vertex;
> +}
> +
> +[fragment shader]
> +void main()
> +{
> + gl_FragColor = vec4(0.0);
> +}
> +
> +[test]
> +clear color 0.0 1.0 0.0 1.0
> +clear
> +clear color 0.5 0.0 0.0 1.0
> +clear
> +clear color 0.0 1.0 0.0 1.0
> +clear
> +probe all rgba 0.0 1.0 0.0 1.0
> diff --git a/tests/fast_color_clear/non-redundant-clear.shader_test b/tests/fast_color_clear/non-redundant-clear.shader_test
> new file mode 100644
> index 0000000..208091f
> --- /dev/null
> +++ b/tests/fast_color_clear/non-redundant-clear.shader_test
> @@ -0,0 +1,46 @@
> +# Test the "fast clear" functionality of i965/gen7+ when two clears
> +# are done to the same buffer, but with drawing in between (so the
> +# second clear is not redundant).
> +#
> +# Fast clear functionality on i965/gen7+ operates by deferring the
> +# memory writes that would normally be done during a buffer clear so
> +# that they happen either during rendering or at the time buffer data
> +# is read (e.g. by glReadPixels). If two fast clears appear in a row,
> +# the second clear can be elided. But if rendering occurrs between
> +# the two fast clears, the second clear is necessary, since some of
> +# the first clear's writes may already have been performed.
> +#
> +# This test verifies that in the case where two fast clears are
> +# separated by rendering, the second clear is actually performed.
> +
> +[require]
> +GLSL >= 1.10
> +
> +[vertex shader]
> +attribute vec4 pos;
> +
> +void main()
> +{
> + gl_Position = pos;
> +}
> +
> +[fragment shader]
> +uniform vec4 color;
> +
> +void main()
> +{
> + gl_FragColor = color;
> +}
> +
> +[vertex data]
> +pos/float/4
> +0.0 0.0 0.0 1.0
> +1.0 1.0 0.0 1.0
> +
> +[test]
> +clear color 0.0 1.0 0.0 1.0
> +clear
> +uniform vec4 color 1.0 0.0 0.0 1.0
> +draw arrays GL_LINES 0 2
> +clear
> +probe all rgba 0.0 1.0 0.0 1.0
> diff --git a/tests/fast_color_clear/redundant-clear.shader_test b/tests/fast_color_clear/redundant-clear.shader_test
> new file mode 100644
> index 0000000..ade0973
> --- /dev/null
> +++ b/tests/fast_color_clear/redundant-clear.shader_test
> @@ -0,0 +1,50 @@
> +# Test the "fast clear" functionality of i965/gen7+ when a redundant
> +# clear is done, but to a different clear color.
> +#
> +# Fast clear functionality on i965/gen7+ operates by deferring the
> +# memory writes that would normally be done during a buffer clear so
> +# that they happen either during rendering or at the time buffer data
> +# is read (e.g. by glReadPixels). If two fast clears appear in a row,
> +# the second clear can be elided (even if it clears to a different
> +# color, since none of the color writes have occurred yet).
> +#
> +# This test verifies that if two fast clears are performed in a row,
> +# the color that takes effect is the color from the second clear. As
> +# with fast-clear-all-colors.shader_test, we also make a draw call
> +# that touches a small number of pixels and outputs the same color.
> +# This is to ensure that some writes are deferred to rendering time
> +# and others are deferred until glReadPixels time, so that we verify
> +# that correct data is written in both circumstances.
> +
> +[require]
> +GLSL >= 1.10
> +
> +[vertex shader]
> +attribute vec4 pos;
> +
> +void main()
> +{
> + gl_Position = pos;
> +}
> +
> +[fragment shader]
> +uniform vec4 color;
> +
> +void main()
> +{
> + gl_FragColor = color;
> +}
> +
> +[vertex data]
> +pos/float/4
> +0.0 0.0 0.0 1.0
> +1.0 1.0 0.0 1.0
> +
> +[test]
> +clear color 1.0 0.0 0.0 1.0
> +clear
> +clear color 0.0 1.0 0.0 1.0
> +clear
> +uniform vec4 color 0.0 1.0 0.0 1.0
> +draw arrays GL_LINES 0 2
> +probe all rgba 0.0 1.0 0.0 1.0
>
More information about the Piglit
mailing list