llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
/**************************************************************************
|
|
|
|
*
|
|
|
|
* Copyright 2010-2021 VMware, Inc.
|
|
|
|
* All Rights Reserved.
|
|
|
|
*
|
|
|
|
* Permission is hereby granted, free of charge, to any person obtaining a
|
|
|
|
* copy of this software and associated documentation files (the
|
|
|
|
* "Software"), to deal in the Software without restriction, including
|
|
|
|
* without limitation the rights to use, copy, modify, merge, publish,
|
|
|
|
* distribute, sub license, and/or sell copies of the Software, and to
|
|
|
|
* permit persons to whom the Software is furnished to do so, subject to
|
|
|
|
* the following conditions:
|
|
|
|
*
|
|
|
|
* The above copyright notice and this permission notice (including the
|
|
|
|
* next paragraph) shall be included in all copies or substantial portions
|
|
|
|
* of the Software.
|
|
|
|
*
|
|
|
|
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
|
|
|
|
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
|
|
|
|
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
|
|
|
|
* IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
|
|
|
|
* ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
|
|
|
|
* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
|
|
|
|
* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
|
|
|
|
*
|
|
|
|
**************************************************************************/
|
|
|
|
|
|
|
|
#include <limits.h>
|
|
|
|
|
|
|
|
#include "pipe/p_defines.h"
|
|
|
|
#include "util/u_inlines.h"
|
|
|
|
#include "util/u_memory.h"
|
|
|
|
#include "util/u_pointer.h"
|
|
|
|
#include "util/format/u_format.h"
|
|
|
|
#include "util/u_dump.h"
|
|
|
|
#include "util/u_string.h"
|
|
|
|
#include "util/os_time.h"
|
|
|
|
#include "pipe/p_shader_tokens.h"
|
|
|
|
#include "draw/draw_context.h"
|
|
|
|
#include "tgsi/tgsi_dump.h"
|
|
|
|
#include "tgsi/tgsi_scan.h"
|
|
|
|
#include "tgsi/tgsi_parse.h"
|
|
|
|
#include "gallivm/lp_bld_type.h"
|
|
|
|
#include "gallivm/lp_bld_const.h"
|
|
|
|
#include "gallivm/lp_bld_conv.h"
|
|
|
|
#include "gallivm/lp_bld_init.h"
|
|
|
|
#include "gallivm/lp_bld_intr.h"
|
|
|
|
#include "gallivm/lp_bld_logic.h"
|
|
|
|
#include "gallivm/lp_bld_tgsi.h"
|
|
|
|
#include "gallivm/lp_bld_swizzle.h"
|
|
|
|
#include "gallivm/lp_bld_flow.h"
|
|
|
|
#include "gallivm/lp_bld_printf.h"
|
|
|
|
#include "gallivm/lp_bld_debug.h"
|
2022-02-25 00:00:25 +00:00
|
|
|
#include "gallivm/lp_bld_nir.h"
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
|
|
|
#include "lp_bld_alpha.h"
|
|
|
|
#include "lp_bld_blend.h"
|
|
|
|
#include "lp_bld_depth.h"
|
|
|
|
#include "lp_bld_interp.h"
|
|
|
|
#include "lp_context.h"
|
|
|
|
#include "lp_debug.h"
|
|
|
|
#include "lp_perf.h"
|
|
|
|
#include "lp_screen.h"
|
|
|
|
#include "lp_setup.h"
|
|
|
|
#include "lp_state.h"
|
|
|
|
#include "lp_tex_sample.h"
|
|
|
|
#include "lp_flush.h"
|
|
|
|
#include "lp_state_fs.h"
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Sampler.
|
|
|
|
*/
|
|
|
|
struct linear_sampler
|
|
|
|
{
|
|
|
|
struct lp_build_sampler_aos base;
|
|
|
|
LLVMValueRef texels_ptrs[LP_MAX_LINEAR_TEXTURES];
|
|
|
|
LLVMValueRef counter;
|
|
|
|
unsigned instance;
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Provide texels to the TGSI translation.
|
|
|
|
*
|
|
|
|
* We don't actually do any texture sampling here, but simply hand the
|
|
|
|
* precomputed row of texels.
|
|
|
|
*/
|
|
|
|
static LLVMValueRef
|
|
|
|
emit_fetch_texel_linear(const struct lp_build_sampler_aos *base,
|
|
|
|
struct lp_build_context *bld,
|
2022-06-22 19:42:38 +01:00
|
|
|
enum tgsi_texture_type target,
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
unsigned unit,
|
|
|
|
LLVMValueRef coords,
|
|
|
|
const struct lp_derivatives derivs,
|
|
|
|
enum lp_build_tex_modifier modifier)
|
|
|
|
{
|
|
|
|
struct linear_sampler *sampler = (struct linear_sampler *)base;
|
|
|
|
|
|
|
|
if (sampler->instance >= LP_MAX_LINEAR_TEXTURES) {
|
|
|
|
assert(FALSE);
|
|
|
|
return bld->undef;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Pointer to a row of texels */
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef texels_ptr = sampler->texels_ptrs[sampler->instance];
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef texel = lp_build_pointer_get(bld->gallivm->builder,
|
|
|
|
texels_ptr, sampler->counter);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
assert(LLVMTypeOf(texel) == bld->vec_type);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We have a struct lp_linear_sampler instance per TEX instruction,
|
2022-05-17 05:11:02 +01:00
|
|
|
* _not_ per unit, as each TEX instruction will need separate storage
|
|
|
|
* for the texels.
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
*/
|
|
|
|
(void)unit;
|
|
|
|
++sampler->instance;
|
|
|
|
|
|
|
|
return texel;
|
|
|
|
}
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
/**
|
|
|
|
* Generates the main body of the fragment shader
|
|
|
|
* Supports generating code for 4 pixel blocks and individual pixels
|
|
|
|
*/
|
|
|
|
static LLVMValueRef
|
|
|
|
llvm_fragment_body(struct lp_build_context *bld,
|
|
|
|
struct lp_fragment_shader *shader,
|
|
|
|
struct lp_fragment_shader_variant *variant,
|
|
|
|
struct linear_sampler* sampler,
|
|
|
|
LLVMValueRef *inputs_ptrs,
|
|
|
|
LLVMValueRef consts_ptr,
|
|
|
|
LLVMValueRef blend_color,
|
|
|
|
LLVMValueRef alpha_ref,
|
|
|
|
struct lp_type fs_type,
|
|
|
|
LLVMValueRef dst)
|
|
|
|
{
|
2022-05-17 05:11:02 +01:00
|
|
|
static const unsigned char bgra_swizzles[4] = {2, 1, 0, 3};
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
LLVMValueRef inputs[PIPE_MAX_SHADER_INPUTS];
|
|
|
|
LLVMValueRef outputs[PIPE_MAX_SHADER_OUTPUTS];
|
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
|
|
|
struct gallivm_state *gallivm = bld->gallivm;
|
|
|
|
LLVMValueRef result = NULL;
|
|
|
|
|
|
|
|
sampler->instance = 0;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Advance inputs
|
|
|
|
*/
|
2022-05-17 05:11:02 +01:00
|
|
|
unsigned i;
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
for (i = 0; i < shader->info.base.num_inputs; ++i) {
|
2022-05-17 05:11:02 +01:00
|
|
|
inputs[i] =
|
|
|
|
lp_build_pointer_get(builder, inputs_ptrs[i], sampler->counter);
|
|
|
|
assert(LLVMTypeOf(inputs[i]) == bld->vec_type);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
}
|
|
|
|
for ( ; i < PIPE_MAX_SHADER_INPUTS; ++i) {
|
|
|
|
inputs[i] = bld->undef;
|
|
|
|
}
|
|
|
|
|
|
|
|
for (i = 0; i < PIPE_MAX_SHADER_OUTPUTS; ++i) {
|
|
|
|
outputs[i] = bld->undef;
|
|
|
|
}
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
if (shader->base.type == PIPE_SHADER_IR_TGSI) {
|
2022-02-25 00:00:25 +00:00
|
|
|
lp_build_tgsi_aos(gallivm, shader->base.tokens, fs_type,
|
|
|
|
bgra_swizzles,
|
|
|
|
consts_ptr, inputs, outputs,
|
|
|
|
&sampler->base,
|
|
|
|
&shader->info.base);
|
2022-05-17 05:11:02 +01:00
|
|
|
} else {
|
2022-02-25 00:00:25 +00:00
|
|
|
nir_shader *clone = nir_shader_clone(NULL, shader->base.ir.nir);
|
|
|
|
lp_build_nir_aos(gallivm, clone, fs_type,
|
|
|
|
bgra_swizzles,
|
|
|
|
consts_ptr, inputs, outputs,
|
|
|
|
&sampler->base,
|
|
|
|
&shader->info.base);
|
|
|
|
ralloc_free(clone);
|
|
|
|
}
|
|
|
|
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
/*
|
|
|
|
* Blend output color
|
|
|
|
*/
|
|
|
|
for (i = 0; i < shader->info.base.num_outputs; ++i) {
|
|
|
|
if (!outputs[i])
|
|
|
|
continue;
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef output = LLVMBuildLoad(builder, outputs[i], "");
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
lp_build_name(output, "output%u", i);
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
unsigned cbuf = shader->info.base.output_semantic_index[i];
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
lp_build_name(output, "cbuf%u", cbuf);
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
if (shader->info.base.output_semantic_name[i]
|
|
|
|
!= TGSI_SEMANTIC_COLOR || cbuf != 0) {
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
continue;
|
2022-05-17 05:11:02 +01:00
|
|
|
}
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
|
|
|
/* Perform alpha test if necessary */
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef mask = NULL;
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
if (variant->key.alpha.enabled) {
|
|
|
|
LLVMTypeRef vec_type = lp_build_vec_type(gallivm, fs_type);
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef broadcast_alpha = lp_build_broadcast(gallivm, vec_type,
|
|
|
|
alpha_ref);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
mask = lp_build_cmp(bld, variant->key.alpha.func, output,
|
|
|
|
broadcast_alpha);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
/* XXX is 4 correct? */
|
|
|
|
mask = lp_build_swizzle_scalar_aos(bld, mask, bgra_swizzles[3], 4);
|
|
|
|
|
|
|
|
lp_build_name(mask, "alpha_test_mask");
|
|
|
|
}
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef src1 = lp_build_zero(gallivm, fs_type);
|
|
|
|
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
result = lp_build_blend_aos(gallivm,
|
|
|
|
&variant->key.blend,
|
|
|
|
variant->key.cbuf_format[i],
|
|
|
|
fs_type,
|
|
|
|
cbuf, /* rt */
|
|
|
|
output, /* src */
|
|
|
|
NULL, /* src_alpha */
|
|
|
|
src1, /* src1 */
|
|
|
|
NULL, /* src1_alpha */
|
|
|
|
dst,
|
|
|
|
mask,
|
|
|
|
blend_color, /* const_ */
|
|
|
|
NULL, /* const_alpha */
|
|
|
|
bgra_swizzles,
|
|
|
|
4);
|
|
|
|
}
|
|
|
|
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
/**
|
|
|
|
* Generate a function that executes the fragment shader in a linear fashion.
|
2022-05-17 05:11:02 +01:00
|
|
|
* The shader operates on unorm8[16] vectors.
|
|
|
|
* See lp_state_fs_analysis for the "linear" conditions.
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
*/
|
|
|
|
void
|
|
|
|
llvmpipe_fs_variant_linear_llvm(struct llvmpipe_context *lp,
|
|
|
|
struct lp_fragment_shader *shader,
|
|
|
|
struct lp_fragment_shader_variant *variant)
|
|
|
|
{
|
2022-05-17 05:11:02 +01:00
|
|
|
assert(shader->kind == LP_FS_KIND_BLIT_RGBA ||
|
|
|
|
shader->kind == LP_FS_KIND_BLIT_RGB1 ||
|
|
|
|
shader->kind == LP_FS_KIND_LLVM_LINEAR);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
struct gallivm_state *gallivm = variant->gallivm;
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
LLVMTypeRef int8t = LLVMInt8TypeInContext(gallivm->context);
|
|
|
|
LLVMTypeRef int32t = LLVMInt32TypeInContext(gallivm->context);
|
|
|
|
LLVMTypeRef pint8t = LLVMPointerType(int8t, 0);
|
|
|
|
LLVMTypeRef pixelt = LLVMVectorType(int32t, 4);
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
// unorm8[16] vector type
|
|
|
|
struct lp_type fs_type;
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
memset(&fs_type, 0, sizeof fs_type);
|
|
|
|
fs_type.floating = FALSE;
|
|
|
|
fs_type.sign = FALSE;
|
|
|
|
fs_type.norm = TRUE;
|
|
|
|
fs_type.width = 8;
|
|
|
|
fs_type.length = 16;
|
|
|
|
|
|
|
|
if (LP_DEBUG & DEBUG_TGSI) {
|
2022-05-17 05:11:02 +01:00
|
|
|
if (shader->base.tokens) {
|
|
|
|
tgsi_dump(shader->base.tokens, 0);
|
|
|
|
}
|
|
|
|
if (shader->base.ir.nir) {
|
|
|
|
nir_print_shader(shader->base.ir.nir, stderr);
|
|
|
|
}
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Generate the function prototype. Any change here must be reflected in
|
|
|
|
* lp_jit.h's lp_jit_frag_func function pointer type, and vice-versa.
|
|
|
|
*/
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
char func_name[256];
|
2022-02-22 23:26:05 +00:00
|
|
|
snprintf(func_name, sizeof(func_name), "fs_variant_linear");
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMTypeRef ret_type = pint8t;
|
|
|
|
LLVMTypeRef arg_types[4];
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
arg_types[0] = variant->jit_linear_context_ptr_type; /* context */
|
2022-05-17 05:11:02 +01:00
|
|
|
arg_types[1] = int32t; /* x */
|
|
|
|
arg_types[2] = int32t; /* y */
|
|
|
|
arg_types[3] = int32t; /* width */
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMTypeRef func_type =
|
|
|
|
LLVMFunctionType(ret_type, arg_types, ARRAY_SIZE(arg_types), 0);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef function =
|
|
|
|
LLVMAddFunction(gallivm->module, func_name, func_type);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
LLVMSetFunctionCallConv(function, LLVMCCallConv);
|
|
|
|
|
|
|
|
variant->linear_function = function;
|
|
|
|
|
|
|
|
/* XXX: need to propagate noalias down into color param now we are
|
|
|
|
* passing a pointer-to-pointer?
|
|
|
|
*/
|
2022-05-17 05:11:02 +01:00
|
|
|
for (unsigned i = 0; i < ARRAY_SIZE(arg_types); ++i) {
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
if (LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind) {
|
|
|
|
lp_add_function_attr(function, i + 1, LP_FUNC_ATTR_NOALIAS);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2022-02-22 23:26:05 +00:00
|
|
|
if (variant->gallivm->cache->data_size)
|
|
|
|
return;
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef context_ptr = LLVMGetParam(function, 0);
|
|
|
|
LLVMValueRef x = LLVMGetParam(function, 1);
|
|
|
|
LLVMValueRef y = LLVMGetParam(function, 2);
|
|
|
|
LLVMValueRef width = LLVMGetParam(function, 3);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
|
|
|
lp_build_name(context_ptr, "context");
|
|
|
|
lp_build_name(x, "x");
|
|
|
|
lp_build_name(y, "y");
|
|
|
|
lp_build_name(width, "width");
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Function body
|
|
|
|
*/
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMBasicBlockRef block =
|
|
|
|
LLVMAppendBasicBlockInContext(gallivm->context, function, "entry");
|
|
|
|
LLVMBuilderRef builder = gallivm->builder;
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
|
|
|
LLVMPositionBuilderAtEnd(builder, block);
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
struct lp_build_context bld;
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
lp_build_context_init(&bld, gallivm, fs_type);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Get context data
|
|
|
|
*/
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef consts_ptr =
|
|
|
|
lp_jit_linear_context_constants(gallivm, context_ptr);
|
|
|
|
LLVMValueRef interpolators_ptr =
|
|
|
|
lp_jit_linear_context_inputs(gallivm, context_ptr);
|
|
|
|
LLVMValueRef samplers_ptr =
|
|
|
|
lp_jit_linear_context_tex(gallivm, context_ptr);
|
|
|
|
|
|
|
|
LLVMValueRef color0_ptr =
|
|
|
|
lp_jit_linear_context_color0(gallivm, context_ptr);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
color0_ptr = LLVMBuildLoad(builder, color0_ptr, "");
|
2022-05-17 05:11:02 +01:00
|
|
|
color0_ptr = LLVMBuildBitCast(builder, color0_ptr,
|
|
|
|
LLVMPointerType(bld.vec_type, 0), "");
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef blend_color =
|
|
|
|
lp_jit_linear_context_blend_color(gallivm, context_ptr);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
blend_color = LLVMBuildLoad(builder, blend_color, "");
|
2022-05-17 05:11:02 +01:00
|
|
|
blend_color = lp_build_broadcast(gallivm, LLVMVectorType(int32t, 4),
|
|
|
|
blend_color);
|
|
|
|
blend_color = LLVMBuildBitCast(builder, blend_color,
|
|
|
|
LLVMVectorType(int8t, 16), "");
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef alpha_ref =
|
|
|
|
lp_jit_linear_context_alpha_ref(gallivm, context_ptr);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
alpha_ref = LLVMBuildLoad(builder, alpha_ref, "");
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Invoke the input interpolators
|
|
|
|
*/
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef inputs_ptrs[LP_MAX_LINEAR_INPUTS];
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
for (unsigned attrib = 0; attrib < shader->info.base.num_inputs; ++attrib) {
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
assert(attrib < LP_MAX_LINEAR_INPUTS);
|
|
|
|
if (attrib >= LP_MAX_LINEAR_INPUTS) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef index = LLVMConstInt(int32t, attrib, 0);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef elem =
|
|
|
|
lp_build_array_get(bld.gallivm, interpolators_ptr, index);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
assert(LLVMGetTypeKind(LLVMTypeOf(elem)) == LLVMPointerTypeKind);
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef fetch_ptr = lp_build_pointer_get(builder, elem,
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
LLVMConstInt(int32t, 0, 0));
|
|
|
|
assert(LLVMGetTypeKind(LLVMTypeOf(fetch_ptr)) == LLVMPointerTypeKind);
|
|
|
|
|
|
|
|
/* Pointer to a row of interpolated inputs */
|
|
|
|
elem = LLVMBuildBitCast(builder, elem, pint8t, "");
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef inputs_ptr = LLVMBuildCall(builder, fetch_ptr, &elem, 1, "");
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
assert(LLVMGetTypeKind(LLVMTypeOf(inputs_ptr)) == LLVMPointerTypeKind);
|
|
|
|
|
|
|
|
/* Mark the function read-only so that LLVM can optimize it away */
|
|
|
|
lp_add_function_attr(inputs_ptr, -1, LP_FUNC_ATTR_READONLY);
|
|
|
|
lp_add_function_attr(inputs_ptr, -1, LP_FUNC_ATTR_NOUNWIND);
|
|
|
|
|
|
|
|
lp_build_name(inputs_ptr, "input%u_ptr", attrib);
|
|
|
|
|
|
|
|
inputs_ptrs[attrib] = inputs_ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Invoke and hook up the texture samplers.
|
|
|
|
*/
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
struct linear_sampler sampler;
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
memset(&sampler, 0, sizeof sampler);
|
|
|
|
sampler.base.emit_fetch_texel = &emit_fetch_texel_linear;
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
for (unsigned attrib = 0; attrib < shader->info.num_texs; ++attrib) {
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
assert(attrib < LP_MAX_LINEAR_TEXTURES);
|
|
|
|
if (attrib >= LP_MAX_LINEAR_TEXTURES) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef index = LLVMConstInt(int32t, attrib, 0);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef elem = lp_build_array_get(bld.gallivm, samplers_ptr, index);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
assert(LLVMGetTypeKind(LLVMTypeOf(elem)) == LLVMPointerTypeKind);
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef fetch_ptr =
|
|
|
|
lp_build_pointer_get(builder, elem, LLVMConstInt(int32t, 0, 0));
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
assert(LLVMGetTypeKind(LLVMTypeOf(fetch_ptr)) == LLVMPointerTypeKind);
|
|
|
|
|
|
|
|
/* Pointer to a row of texels */
|
|
|
|
elem = LLVMBuildBitCast(builder, elem, pint8t, "");
|
2022-05-17 05:11:02 +01:00
|
|
|
LLVMValueRef texels_ptr = LLVMBuildCall(builder, fetch_ptr, &elem, 1, "");
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
assert(LLVMGetTypeKind(LLVMTypeOf(texels_ptr)) == LLVMPointerTypeKind);
|
|
|
|
|
|
|
|
/* Mark the function read-only so that LLVM can optimize it away */
|
|
|
|
lp_add_function_attr(texels_ptr, -1, LP_FUNC_ATTR_READONLY);
|
|
|
|
lp_add_function_attr(texels_ptr, -1, LP_FUNC_ATTR_NOUNWIND);
|
|
|
|
|
|
|
|
lp_build_name(texels_ptr, "tex%u_ptr", attrib);
|
|
|
|
|
|
|
|
sampler.texels_ptrs[attrib] = texels_ptr;
|
|
|
|
}
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
/* excess = width & 0x3 */
|
|
|
|
LLVMValueRef excess =
|
|
|
|
LLVMBuildAnd(builder, width, LLVMConstInt(int32t, 3, 0), "");
|
|
|
|
/* width *= 4 */
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
width = LLVMBuildLShr(builder, width, LLVMConstInt(int32t, 2, 0), "");
|
|
|
|
|
|
|
|
/* Loop over blocks of 4 pixels */
|
2022-05-17 05:11:02 +01:00
|
|
|
/* for loop.counter = 0; loop.counter < width; loop.counter++) { */
|
|
|
|
struct lp_build_for_loop_state loop;
|
|
|
|
lp_build_for_loop_begin(&loop, gallivm, LLVMConstInt(int32t, 0, 0),
|
|
|
|
LLVMIntULT, width, LLVMConstInt(int32t, 1, 0));
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
{
|
|
|
|
LLVMValueRef value;
|
|
|
|
sampler.counter = loop.counter;
|
|
|
|
|
|
|
|
/* Read 4 pixels */
|
2022-05-17 05:11:02 +01:00
|
|
|
value = lp_build_pointer_get_unaligned(builder, color0_ptr,
|
|
|
|
loop.counter, 4);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
|
|
|
/* Perform fragment shader body */
|
2022-05-17 05:11:02 +01:00
|
|
|
value = llvm_fragment_body(&bld, shader, variant, &sampler, inputs_ptrs,
|
|
|
|
consts_ptr, blend_color, alpha_ref, fs_type,
|
|
|
|
value);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
|
|
|
/* Write 4 pixels */
|
2022-05-17 05:11:02 +01:00
|
|
|
lp_build_pointer_set_unaligned(builder, color0_ptr, loop.counter,
|
|
|
|
value, 4);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
}
|
|
|
|
lp_build_for_loop_end(&loop);
|
|
|
|
|
|
|
|
/* Compute the edge pixels (width % 4) */
|
2022-05-17 05:11:02 +01:00
|
|
|
struct lp_build_if_state ifstate;
|
|
|
|
lp_build_if(&ifstate, gallivm, LLVMBuildICmp(builder, LLVMIntNE, excess,
|
|
|
|
LLVMConstInt(int32t, 0, 0), ""));
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
{
|
|
|
|
struct lp_build_loop_state loop_read, loop_write;
|
|
|
|
LLVMValueRef buf, elem, result, pixel_ptr;
|
|
|
|
LLVMValueRef buf_ptr = lp_build_alloca(gallivm, pixelt, "");
|
|
|
|
|
|
|
|
sampler.counter = width;
|
|
|
|
|
|
|
|
/* Get the i32* pixel pointer from the <i16x8>* element pointer */
|
|
|
|
pixel_ptr = LLVMBuildGEP(gallivm->builder, color0_ptr, &width, 1, "");
|
2022-05-17 05:11:02 +01:00
|
|
|
pixel_ptr = LLVMBuildBitCast(gallivm->builder, pixel_ptr,
|
|
|
|
LLVMPointerType(int32t, 0), "");
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
|
|
|
/* Copy individual pixels from memory to local buffer */
|
|
|
|
lp_build_loop_begin(&loop_read, gallivm, LLVMConstInt(int32t, 0, 0));
|
|
|
|
{
|
2022-05-17 05:11:02 +01:00
|
|
|
elem = lp_build_pointer_get(gallivm->builder,
|
|
|
|
pixel_ptr, loop_read.counter);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
|
|
|
buf = LLVMBuildLoad(gallivm->builder, buf_ptr, "");
|
2022-05-17 05:11:02 +01:00
|
|
|
buf = LLVMBuildInsertElement(builder, buf, elem,
|
|
|
|
loop_read.counter, "");
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
LLVMBuildStore(builder, buf, buf_ptr);
|
|
|
|
}
|
2022-05-17 05:11:02 +01:00
|
|
|
lp_build_loop_end_cond(&loop_read, excess,
|
|
|
|
LLVMConstInt(int32t, 1, 0), LLVMIntUGE);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
|
|
|
/* Perform fragment shader body */
|
|
|
|
buf = LLVMBuildLoad(gallivm->builder, buf_ptr, "");
|
|
|
|
buf = LLVMBuildBitCast(builder, buf, bld.vec_type, "");
|
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
result = llvm_fragment_body(&bld, shader, variant, &sampler,
|
|
|
|
inputs_ptrs, consts_ptr, blend_color,
|
|
|
|
alpha_ref, fs_type, buf);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
result = LLVMBuildBitCast(builder, result, pixelt, "");
|
|
|
|
|
|
|
|
/* Write individual pixels from local buffer to the memory */
|
|
|
|
lp_build_loop_begin(&loop_write, gallivm, LLVMConstInt(int32t, 0, 0));
|
|
|
|
{
|
2022-05-17 05:11:02 +01:00
|
|
|
elem = LLVMBuildExtractElement(builder, result,
|
|
|
|
loop_write.counter, "");
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
|
2022-05-17 05:11:02 +01:00
|
|
|
lp_build_pointer_set(gallivm->builder, pixel_ptr,
|
|
|
|
loop_write.counter, elem);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
}
|
2022-05-17 05:11:02 +01:00
|
|
|
lp_build_loop_end_cond(&loop_write, excess,
|
|
|
|
LLVMConstInt(int32t, 1, 0), LLVMIntUGE);
|
llvmpipe: Add a linear rasterizer optimized for 2D rendering.
This change adds:
- an alternative rasterizer, which rasterizes bins in a left->right &
top->bottom linear fashion;
- triangle -> rectangle detection;
- 1:1 blit detection;
- a special TGSI -> LLVM IR code generation that uses 8-bit SSE integers
in AoS fashion (as opposed to 32bits floats.)
Altogether these changes yield a 2x to 3x performance improvement for 2D
workloads. It was designed to render Windows 7 Aero and other Windows
built-in 3D applications (like Windows Media Player, Internet Explorer
11, UWP applications) with minimum CPU utilization, but it should be
generally applicable to other 2D-on-3D applications, like desktop
compositors, HTML browsers, 3D based UI toolkits, etc.
This was mostly the brainchild of Keith Whitwell back in 2010. I wrote
TGSI -> AoS translation. And many others added bug-fixes and
enhancements over the years: Roland Scheidegger, Brian Paul, and James
Benton.
Known issues:
- piglit spec@!opengl 1.1@quad-invariance will warn that "left and right
half should match" due to rounding error difference
- These optimized paths to kick in is that depth-buffer must not be
used, so some applications which want to benefit from these improvements
might need to be modified to ensure they use painter's algorithm instead
of depth-buffers.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Keith Whitwell <keithw@vmware.com>
v2: Incorporate Dave Airlie feedback: cleanup LP_DEBUG_xx; shrink 3+
empty lines.
v3: silence unused var warning, adapt to new upstream code (point setup)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11969>
2021-05-07 13:49:07 +01:00
|
|
|
}
|
|
|
|
lp_build_endif(&ifstate);
|
|
|
|
|
|
|
|
color0_ptr = LLVMBuildBitCast(builder, color0_ptr, pint8t, "");
|
|
|
|
|
|
|
|
LLVMBuildRet(builder, color0_ptr);
|
|
|
|
|
|
|
|
gallivm_verify_function(gallivm, function);
|
|
|
|
}
|