Add psxpress MDEC API and psn00bsdk_target_incbin_a

4 files changed, 383 insertions, 25 deletions

diff --git a/doc/cmake_reference.md b/doc/cmake_reference.md
index 60c46c0..25a89ec 100644
--- a/doc/cmake_reference.md
+++ b/doc/cmake_reference.md
@@ -4,17 +4,17 @@
 ## Setup
 
 The only requirement to use the SDK in CMake is to set the
-`CMAKE_TOOLCHAIN_FILE` variable to `INSTALL_PATH/lib/libpsn00b/cmake/sdk.cmake`
-(where `INSTALL_PATH` is the install prefix PSn00bSDK is installed to). This
-can be done on the command line (`-DCMAKE_TOOLCHAIN_FILE=...`), in
-`CMakeLists.txt` (before calling `project()`) or using a
-[preset](https://cmake.org/cmake/help/latest/manual/cmake-presets.7.html).
+`CMAKE_TOOLCHAIN_FILE` variable to the absolute path to
+`lib/libpsn00b/cmake/sdk.cmake` within the PSn00bSDK installation directory.
+This can be done on the command line (`-DCMAKE_TOOLCHAIN_FILE=...`), in
+`CMakeLists.txt` (`set(CMAKE_TOOLCHAIN_FILE ...)` before `project()`) or using
+[presets](https://cmake.org/cmake/help/latest/manual/cmake-presets.7.html).
 
 It's suggested to have a default preset that sets `CMAKE_TOOLCHAIN_FILE` to
-`$env{PSN00BSDK_LIBS}/cmake/sdk.cmake`, so the `PSN00BSDK_LIBS` environment
-variable (used by former PSn00bSDK versions) is respected. Such a preset can be
-created by placing a `CMakePresets.json` file in the project's root with the
-following contents:
+`$env{PSN00BSDK_LIBS}/cmake/sdk.cmake`, taking advantage of the
+`PSN00BSDK_LIBS` environment variable (used by former PSn00bSDK versions) to
+automatically find the SDK. Such a preset can be created by placing a
+`CMakePresets.json` file in the project's root with the following contents:
 
 ```json
 {
@@ -135,9 +135,12 @@ of these names.
 
   A new symbol/object will be created with the given name, escaped by replacing
   non-alphanumeric characters with underscores. The contents of the file will
-  be aligned to 4 bytes and placed in the `.data` section. Once added the file
-  and its size can be accessed by C/C++ code by declaring the symbol as an
-  extern array and `<symbol name>_size` as an integer, like this:
+  be aligned to 4 bytes and placed in the `.data` section. An unsigned 32-bit
+  integer named `<symbol name>_size` will also be defined and set to the length
+  of the file in bytes (without taking alignment/padding into account).
+
+  Once added the file and its size can be accessed by C/C++ code by declaring
+  the respective symbols as an extern array and as an integer, like this:
 
   ```c
   extern const uint8_t my_file[];
@@ -154,6 +157,13 @@ of these names.
   must be enabled by specifying `LANGUAGES C ASM` (or `LANGUAGES C CXX ASM` if
   C++ is also used) when invoking `project()`.
 
+- `psn00bsdk_target_incbin_a(<target> <PRIVATE|PUBLIC|INTERFACE> <symbol name> <size symbol name> <binary file> <alignment>)`
+
+  Advanced variant of `psn00bsdk_target_incbin()` that allows specifying a
+  custom name for the size symbol and changing the default alignment setting.
+  Note that the size integer is always aligned to a multiple of 4 bytes as the
+  MIPS architecture doesn't support unaligned reads.
+
 ## Definitions
 
 When compiling executables and libraries using the above commands the following
@@ -259,4 +269,4 @@ the build script.
   LZP archives as part of the build pipeline.
 
 -----------------------------------------
-_Last updated on 2021-12-29 by spicyjpeg_
+_Last updated on 2022-02-26 by spicyjpeg_
diff --git a/libpsn00b/cmake/internal_setup.cmake b/libpsn00b/cmake/internal_setup.cmake
index 8fb6482..d293127 100644
--- a/libpsn00b/cmake/internal_setup.cmake
+++ b/libpsn00b/cmake/internal_setup.cmake
@@ -32,7 +32,7 @@ set(
 		psxgte
 		psxspu
 		psxcd
-		#psxpress
+		psxpress
 		psxsio
 		psxetc
 		psxapi
@@ -186,7 +186,7 @@ function(psn00bsdk_add_cd_image name image_name config_file)
 
 	add_custom_target(
 		${name} ALL
-		COMMAND    ${MKPSXISO} -y -q ${CD_CONFIG_FILE}
+		COMMAND    ${MKPSXISO} -y ${CD_CONFIG_FILE}
 		BYPRODUCTS ${image_name}.bin ${image_name}.cue
 		COMMENT    "Building CD image ${image_name}"
 		${ARGN}
@@ -195,13 +195,17 @@ endfunction()
 
 ## Helper functions for assets
 
-# psn00bsdk_target_incbin(
+# psn00bsdk_target_incbin_a(
 #   <existing target name> <PRIVATE|PUBLIC|INTERFACE>
-#   <symbol name>
+#   <data symbol name>
+#   <size symbol name>
 #   <path to binary file>
+#   <linker section name>
+#   <alignment>
 # )
-function(psn00bsdk_target_incbin name type symbol_name path)
+function(psn00bsdk_target_incbin_a name type symbol_name size_name path section align)
 	string(MAKE_C_IDENTIFIER ${symbol_name} _id)
+	string(MAKE_C_IDENTIFIER ${size_name}   _size)
 	cmake_path(ABSOLUTE_PATH path OUTPUT_VARIABLE _path)
 
 	# Generate an assembly source file that includes the binary file and add it
@@ -212,26 +216,27 @@ function(psn00bsdk_target_incbin name type symbol_name path)
 		CONFIGURE
 		OUTPUT  ${_asm_file}
 		CONTENT [[
-.section .data.${_id}
-.balign 4
+.section ${section}
+.balign ${align}
 
 .global ${_id}
 .type ${_id}, @object
-${_id}:	
+${_id}:
 	.incbin "${_path}"
 
 .local ${_id}_end
 ${_id}_end:
 
+.balign ${align}
 .balign 4
 
-.global ${_id}_size
-.type ${_id}_size, @object
-${_id}_size:
+.global ${_size}
+.type ${_size}, @object
+${_size}:
 	.int (${_id}_end - ${_id})
 
 .size ${_id}, (${_id}_end - ${_id})
-.size ${_id}_size, 4
+.size ${_size}, 4
 ]]
 		ESCAPE_QUOTES
 		NEWLINE_STYLE LF
@@ -240,3 +245,22 @@ ${_id}_size:
 	target_sources(${name} ${type} ${_asm_file})
 	set_source_files_properties(${_asm_file} PROPERTIES OBJECT_DEPENDS ${_path})
 endfunction()
+
+# psn00bsdk_target_incbin(
+#   <existing target name> <PRIVATE|PUBLIC|INTERFACE>
+#   <symbol name>
+#   <path to binary file>
+# )
+function(psn00bsdk_target_incbin name type symbol_name path)
+	string(MAKE_C_IDENTIFIER ${symbol_name} _id)
+
+	psn00bsdk_target_incbin_a(
+		${name}
+		${type}
+		${_id}
+		${_id}_size
+		${path}
+		.data.${_id}
+		4
+	)
+endfunction()
diff --git a/libpsn00b/include/psxpress.h b/libpsn00b/include/psxpress.h
new file mode 100644
index 0000000..ad5f6a3
--- /dev/null
+++ b/libpsn00b/include/psxpress.h
@@ -0,0 +1,152 @@
+/*
+ * PSn00bSDK MDEC library
+ * (C) 2022 spicyjpeg - MPL licensed
+ */
+
+#ifndef __PSXPRESS_H
+#define __PSXPRESS_H
+
+#include <stdint.h>
+
+/* Structure definitions */
+
+typedef struct _DECDCTENV {
+	uint8_t iq_y[64];	// Luma quantization table, stored in zigzag order
+	uint8_t iq_c[64];	// Chroma quantization table, stored in zigzag order
+	int16_t dct[64];	// Inverse DCT matrix (2.14 fixed-point)
+} DECDCTENV;
+
+typedef enum _DECDCTMODE {
+	DECDCT_MODE_24BPP		= 1,
+	DECDCT_MODE_16BPP		= 0,
+	DECDCT_MODE_16BPP_BIT15	= 2,
+	DECDCT_MODE_RAW			= -1
+} DECDCTMODE;
+
+/* Public API */
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/**
+ * @brief Resets the MDEC and aborts any MDEC DMA transfers. If mode = 0, the
+ * default IDCT matrix and quantization tables are also loaded and the MDEC is
+ * put into color output mode, discarding any custom environment previously set
+ * with DecDCTPutEnv().
+ *
+ * DecDCTReset(0) must be called at least once prior to using the MDEC.
+ *
+ * @param mode
+ */
+void DecDCTReset(int32_t mode);
+
+/**
+ * @brief Uploads the specified decoding environment's quantization tables and
+ * IDCT matrix to the MDEC, or restores the default tables if a null pointer is
+ * passed. Calling this function is normally not required as DecDCTReset(0)
+ * initializes the MDEC with the default tables, but it may be useful for e.g.
+ * decoding JPEG or a format with custom quantization tables.
+ *
+ * The second argument, not present in the official SDK, specifies whether the
+ * MDEC shall be put into color (0) or monochrome (1) output mode. In
+ * monochrome mode each DCT block decoded from the input stream is transformed
+ * into an 8x8x8bpp bitmap, while in color mode each group of 6 DCT blocks (Cr,
+ * Cb, Y1-4) is used to form a 16x16 RGB bitmap.
+ *
+ * This function uses DecDCTinSync() to wait for the MDEC to become ready and
+ * should not be called during decoding or after calling DecDCTin().
+ *
+ * @param env Pointer to DECDCTENV or 0 for default tables
+ * @param mono 0 for color (normal), 1 for monochrome
+ */
+void DecDCTPutEnv(const DECDCTENV *env, int32_t mono);
+
+/**
+ * @brief Sets up the MDEC to start fetching and decoding a stream from the
+ * given address in main RAM. The first 32-bit word is initially copied to the
+ * MDEC0 register, then all subsequent data is read in 128-byte (32-word)
+ * chunks. The length of the stream (in 32-bit units, minus the first word)
+ * must be encoded in the lower 16 bits of the first word, as expected by the
+ * MDEC.
+ *
+ * The mode argument optionally specifies the output color depth (0 for 16bpp,
+ * 1 for 24bpp) if not already set in the first word. Passing -1 will result in
+ * DecDCTin() copying the first word as-is to MDEC0 without manipulating any of
+ * its bits.
+ *
+ * @param data
+ * @param mode DECDCT_MODE_* or -1
+ */
+void DecDCTin(const uint32_t *data, int32_t mode);
+
+/**
+ * @brief Configures the MDEC to automatically fetch data (the input stream,
+ * IDCT matrix or quantization tables) in 128-byte (32-word) chunks from the
+ * specified address in main RAM. The transfer is stopped, and any callback
+ * registered with DMACallback(0) is fired, once a certain number of 32-bit
+ * words have been read; usually the length should match the number of input
+ * words expected by the MDEC. If the MDEC expects more data its operation will
+ * be paused and can be resumed by calling DecDCTinRaw() again.
+ *
+ * This is a low-level variant of DecDCTin() that only sets up the DMA transfer
+ * and does not write anything to the MDEC0 register. The actual transfer won't
+ * start until the MDEC is given a valid command.
+ * 
+ * @param data
+ * @param length Number of 32-bit words to read (must be multiple of 32)
+ */
+void DecDCTinRaw(const uint32_t *data, size_t length);
+
+/**
+ * @brief Waits for the MDEC to finish decoding the input stream (if mode = 0)
+ * or returns whether it is busy (if mode = 1). MDEC commands can be issued
+ * only when the MDEC isn't busy.
+ *
+ * WARNING: DecDCTinSync(0) might time out and return -1 if the MDEC can't
+ * output decoded data, e.g. if the length passed DecDCTout() was too small and
+ * no callback is registered to set up further transfers. DecDCTinSync(0) shall
+ * only be used alongside DMACallback(1) or if the entirety of the decoded
+ * stream (usually a whole frame) is being written to main RAM.
+ *
+ * @param mode
+ * @return 0 or -1 in case of a timeout (mode = 0) / MDEC busy flag (mode = 1)
+ */
+int32_t DecDCTinSync(int32_t mode);
+
+/**
+ * @brief Configures the MDEC to automatically transfer decoded image data in
+ * 128-byte (32-word) chunks to the specified address in main RAM. MDEC
+ * operation is paused once a certain number of 32-bit words have been output
+ * and can be resumed by calling DecDCTout() again: the MDEC will continue
+ * decoding the input stream from where it left off. Any callback registered
+ * with DMACallback(1) is also fired whenever the transfer ends.
+ *
+ * This behavior allows the MDEC's output to be buffered into 16-pixel-wide
+ * vertical strips in main RAM, which can then be uploaded to VRAM using
+ * LoadImage().
+ *
+ * @param data
+ * @param length Number of 32-bit words to output (must be multiple of 32)
+ */
+void DecDCTout(uint32_t *data, size_t length);
+
+/**
+ * @brief Waits until the transfer set up by DecDCTout() finishes (if mode = 0)
+ * or returns whether it is still in progress (if mode = 1).
+ *
+ * WARNING: DecDCToutSync(0) might time out and return -1 if the MDEC is unable
+ * to consume enough input data in order to produce the desired amount of data.
+ * If the input stream isn't contiguous in memory, DMACallback(0) shall be used
+ * to register a callback that calls DecDCTin() to feed the MDEC.
+ *
+ * @param mode
+ * @return 0 or -1 in case of a timeout (mode = 0) / DMA busy flag (mode = 1)
+ */
+int32_t DecDCToutSync(int32_t mode);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
diff --git a/libpsn00b/psxpress/mdec.c b/libpsn00b/psxpress/mdec.c
new file mode 100644
index 0000000..82e2465
--- /dev/null
+++ b/libpsn00b/psxpress/mdec.c
@@ -0,0 +1,172 @@
+/*
+ * PSn00bSDK MDEC library (low-level MDEC/DMA API)
+ * (C) 2022 spicyjpeg - MPL licensed
+ */
+
+#include <stdint.h>
+#include <stdio.h>
+#include <psxapi.h>
+#include <psxpress.h>
+#include <hwregs_c.h>
+
+#define MDEC_SYNC_TIMEOUT 0x1000000
+
+/* Default IDCT matrix */
+
+#define S0 0x5a82	// 0x4000 * cos(0/16 * pi) * sqrt(2)
+#define S1 0x7d8a	// 0x4000 * cos(1/16 * pi) * 2
+#define S2 0x7641	// 0x4000 * cos(2/16 * pi) * 2
+#define S3 0x6a6d	// 0x4000 * cos(3/16 * pi) * 2
+#define S4 0x5a82	// 0x4000 * cos(4/16 * pi) * 2
+#define S5 0x471c	// 0x4000 * cos(5/16 * pi) * 2
+#define S6 0x30fb	// 0x4000 * cos(6/16 * pi) * 2
+#define S7 0x18f8	// 0x4000 * cos(7/16 * pi) * 2
+
+static const int16_t _default_idct_matrix[] = {
+	S0,  S0,  S0,  S0,  S0,  S0,  S0,  S0,
+	S1,  S3,  S5,  S7, -S7, -S5, -S3, -S1,
+	S2,  S6, -S6, -S2, -S2, -S6,  S6,  S2,
+	S3, -S7, -S1, -S5,  S5,  S1,  S7, -S3,
+	S4, -S4, -S4,  S4,  S4, -S4, -S4,  S4,
+	S5, -S1,  S7,  S3, -S3, -S7,  S1, -S5,
+	S6, -S2,  S2, -S6, -S6,  S2, -S2,  S6,
+	S7, -S5,  S3, -S1,  S1, -S3,  S5, -S7
+};
+
+/* Default quantization tables */
+
+// The default luma and chroma quantization table is based on the MPEG-1
+// quantization table, with the only difference being the first value (2
+// instead of 8). Note that quantization tables are stored in zigzag order
+// rather than row- or column-major.
+// https://problemkaputt.de/psx-spx.htm#mdecdecompression
+static const uint8_t _default_quant_table[] = {
+	  2,  16,  16,  19,  16,  19,  22,  22,
+	 22,  22,  22,  22,  26,  24,  26,  27,
+	 27,  27,  26,  26,  26,  26,  27,  27,
+	 27,  29,  29,  29,  34,  34,  34,  29,
+	 29,  29,  27,  27,  29,  29,  32,  32,
+	 34,  34,  37,  38,  37,  35,  35,  34,
+	 35,  38,  38,  40,  40,  40,  48,  48,
+	 46,  46,  56,  56,  58,  69,  69,  83
+};
+/*static const uint8_t _jpeg_y_quant_table[] = {
+	 16,  11,  12,  14,  12,  10,  16,  14,
+	 13,  14,  18,  17,  16,  19,  24,  40,
+	 26,  24,  22,  22,  24,  49,  35,  37,
+	 29,  40,  58,  51,  61,  60,  57,  51,
+	 56,  55,  64,  72,  92,  78,  64,  68,
+	 87,  69,  55,  56,  80, 109,  81,  87,
+	 95,  98, 103, 104, 103,  62,  77, 113,
+	121, 112, 100, 120,  92, 101, 103,  99
+};
+static const uint8_t _jpeg_c_quant_table[] = {
+	 17,  18,  18,  24,  21,  24,  47,  26,
+	 26,  47,  99,  66,  56,  66,  99,  99,
+	 99,  99,  99,  99,  99,  99,  99,  99,
+	 99,  99,  99,  99,  99,  99,  99,  99,
+	 99,  99,  99,  99,  99,  99,  99,  99,
+	 99,  99,  99,  99,  99,  99,  99,  99,
+	 99,  99,  99,  99,  99,  99,  99,  99,
+	 99,  99,  99,  99,  99,  99,  99,  99
+};*/
+
+/* Public API */
+
+void DecDCTReset(int32_t mode) {
+	EnterCriticalSection();
+
+	DMA_DPCR   |= 0x000000bb; // Enable DMA0 and DMA1
+	DMA_CHCR(0) = 0x00000201; // Stop DMA0
+	DMA_CHCR(1) = 0x00000200; // Stop DMA1
+	MDEC1       = 0x80000000; // Reset MDEC
+	MDEC1       = 0x60000000; // Enable DMA in/out requests
+
+	ExitCriticalSection();
+	if (!mode)
+		DecDCTPutEnv(0, 0);
+}
+
+void DecDCTPutEnv(const DECDCTENV *env, int32_t mono) {
+	const int16_t *dct  = env ? env->dct  : _default_idct_matrix;
+	const uint8_t *iq_y = env ? env->iq_y : _default_quant_table;
+	const uint8_t *iq_c = env ? env->iq_c : _default_quant_table;
+
+	DecDCTinSync(0);
+
+	MDEC0 = 0x60000000; // Set IDCT matrix
+	DecDCTinRaw((const uint32_t *) dct, 32);
+	DecDCTinSync(0);
+
+	MDEC0 = 0x40000000 | (mono ? 0 : 1); // Set table(s)
+	DecDCTinRaw((const uint32_t *) iq_y, 16);
+	DecDCTinSync(0);
+
+	if (!mono) {
+		DecDCTinRaw((const uint32_t *) iq_c, 16);
+		DecDCTinSync(0);
+	}
+}
+
+void DecDCTin(const uint32_t *data, int32_t mode) {
+	uint32_t header = *data;
+	if (mode == DECDCT_MODE_RAW)
+		MDEC0 = header;
+	else if (mode & DECDCT_MODE_24BPP)
+		MDEC0 = header | 0x30000000;
+	else
+		MDEC0 = header | 0x38000000 | ((mode & 2) << 24); // Bit 25 = mask
+
+	DecDCTinRaw((const uint32_t *) &(data[1]), header & 0xffff);
+}
+
+// This is a PSn00bSDK-only function that behaves like DecDCTout(), taking the
+// data length as an argument rather than parsing it from the first 4 bytes of
+// the stream.
+void DecDCTinRaw(const uint32_t *data, size_t length) {
+	DMA_MADR(0) = (uint32_t) data;
+	if (length < 32)
+		DMA_BCR(0) = 0x00010000 | length;
+	else
+		DMA_BCR(0) = 0x00000020 | ((length / 32) << 16);
+
+	DMA_CHCR(0) = 0x01000201;
+}
+
+int32_t DecDCTinSync(int32_t mode) {
+	if (mode)
+		return (MDEC1 >> 29) & 1;
+
+	for (uint32_t i = MDEC_SYNC_TIMEOUT; i; i--) {
+		if (!(MDEC1 & (1 << 29)))
+			return 0;
+	}
+
+	printf("psxpress: DecDCTinSync() timeout\n");
+	return -1;
+}
+
+void DecDCTout(uint32_t *data, size_t length) {
+	DecDCToutSync(0);
+
+	DMA_MADR(1) = (uint32_t) data;
+	if (length < 32)
+		DMA_BCR(1) = 0x00010000 | length;
+	else
+		DMA_BCR(1) = 0x00000020 | ((length / 32) << 16);
+
+	DMA_CHCR(1) = 0x01000200;
+}
+
+int32_t DecDCToutSync(int32_t mode) {
+	if (mode)
+		return (DMA_CHCR(1) >> 24) & 1;
+
+	for (uint32_t i = MDEC_SYNC_TIMEOUT; i; i--) {
+		if (!(DMA_CHCR(1) & (1 << 24)))
+			return 0;
+	}
+
+	printf("psxpress: DecDCToutSync() timeout\n");
+	return -1;
+}